Sacramento City Council
November 18, 2025

Received: 11 December 2020 | Revised: 31 December 2021 | Accepted: 15 January 2022 DOI: 10.1002/jwmg.22216

RESEARCH ARTICLE
Utility‐scale solar impacts to volant wildlife
K. Shawn Smallwood
3108 Finch Street, Davis, CA 95616, USA
Correspondence
K. Shawn Smallwood, 3108 Finch Street,
Davis, CA 95616, USA.

Email: [email protected]
Abstract
To reduce carbon emissions from fossil fuel combustion, Uni- ted States government agencies, including those in California, initiated aggressive programs to hasten development of utility ‐ scale solar energy. Much of California's early development of solar energy occurred in deserts and annual grasslands, much of it on public land. Measurement of solar energy's impacts to wildlife has been limited to mortality caused by features of solar facilities, and has yet to include impacts from habitat loss and energy transmission. To estimate species ‐specific bird and bat fatality rates and statewide mortality, I reviewed reports of fatality monitoring from 1982 to 2018 at 14 projects, which varied in duration, level of sampling, search interval, search method, and carcass detection trials. Because most monitors performed carcass detection trials using species of birds whose members were larger than birds and bats found as fatalities, I bridged the monitors' onsite trial results to offsite trial results based on the same methods but which also measured detec- tion probabilities across the full range of body sizes of species represented by fatalities. This bridge preserved the project site's effects on detection probabilities while more fully ad- justing for the effects of body size. My fatality estimates consistently exceeded those reported. Projected to California's installed capacity of 1,948.8 MW of solar thermal and
12,220 MW of photovoltaic (PV) panels in 2020 (14,168.8 MW total), reported estimates would support an annual statewide fatality estimate of 37,546 birds and 207 bats, whereas I es- timated fatalities of 267,732 birds and 11,418 bats. Fatalities/ Journal of Wildlife Management 2022;e22216. wileyonlinelibrary.com/journal/jwmg | 1o f2 9 https://doi.org/10.1002/jwmg.22216
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2022 The Authors. The Journal of Wildlife Management published by Wiley Periodicals LLC on behalf of The Wildlife Society.

MW/year averaged 11.61 birds and 0.06 bats at PV projects and 64.61 birds and 5.49 bats at solar thermal projects.

Fatalities/km/year averaged 113.16 birds and zero bats at generation tie ‐ins, and 14.44 birds and 2.56 bats along peri- meter fences. Bird fatality rates averaged 3 times higher at PV projects searched by foot rather than car. They were usually biased low by insufficient monitoring duration and by the 22% of fatalities that monitors could not identify to species. I esti- mated that construction grading for solar projects removed habitat that otherwise would have supported nearly 300,000 birds/year. I recommend that utility ‐scale solar energy devel- opment be slowed to improve project decision ‐making, im- pacts assessment, fatality monitoring, mitigation efficacy, and oversight.

KEYWORDS
bats, birds, carcass detection trials, collision mortality, fatality monitoring, solar energy, wildlife impacts
In response to environmental and economic threats posed by carbon emissions from fossil fuel combustion, state and federal governments embarked on ambitious plans to develop renewable energy in California, USA (California Public Utilities Commission 2008). Plans such as the Desert Renewable Energy Conservation Plan (California Energy Commission et al. 2014) were formulated to overcome regulatory and permitting barriers for utility ‐scale renewable energy and its transmission to help Californians achieve the goal of obtaining 60% of its energy from renewable sources by 2030 and 100% of it by 2045. Because development of utility ‐scale renewable energy projects comes with environmental costs, including impacts to wildlife, state and federal agencies collaborated on an Avian ‐Solar Science Coordination Plan (Multiagency Avian ‐Solar Collaborative Working Group 2016) to identify and prioritize information needs related to impacts to birds. Just as understanding of impacts to wildlife lagged the rapid ex- pansion of wind energy development (May et al. 2015, 2017), understanding impacts to wildlife also lags the rapid expansion of utility ‐scale solar energy projects (Multiagency Avian ‐Solar Collaborative Working Group 2016, Kosciuch et al. 2020). For example, only recently was it learned that many of the animals killed at California's solar projects originated out of state (Conkling et al. 2020), thus elevating the impacts to the regional level.

Deliberation is needed concerning the environmental costs and benefits of utility ‐scale solar energy compared to other forms of energy generation (e.g., wind, geothermal, hydroelectric, fossil ‐fuels, nuclear, rooftop solar).

Accurate estimates of wildlife mortality would help determine impacts at project and regional levels. Researchers have calculated regional estimates of wind energy impacts to birds (Loss et al. 2013a, Smallwood 2013, Zimmerling et al. 2013, Erickson et al. 2014) and bats (Hayes 2013; Smallwood 2013, 2020; Zimmerling and Francis 2016).

Researchers have also discussed cumulative impacts (Calvert et al. 2013) and calculated regional estimates of wildlife mortality caused by other anthropogenic sources, such as from collisions with automobiles (Loss et al. 2014a), transmission lines (Rioux et al. 2013, Loss et al. 2014b), and buildings (Machtans et al. 2013, Loss et al. 2014c), and predation from house cats (Loss et al. 2013b). Accuracy of these regional estimates depends on representativeness of the monitored projects (or places) relative to the spatial distribution of all projects (Smallwood 2013, Huso and Dalthorp 2014). It also depends substantially on study design and field methods implemented at the project level.
2o f2 9 | SMALLWOOD
Before 2012, fatality monitoring at utility ‐scale solar projects was rare, largely because few utility ‐scale solar projects existed. The only wildlife fatality monitoring performed before 2012 had been at the 10 ‐MW Solar One project in 1982 ‒1983 near Barstow in San Bernardino, California (McCrary et al. 1986). Since monitoring at Solar One, much had been learned about fatality monitoring at wind projects from the late 1980s through the initial fatality monitoring plan for solar projects (Nicolai et al. 2011) and its update in 2016 (Huso et al. 2016). Managers formulated guidance documents to standardize survey methods for the purposes of accurately estimating and comparing fatalities among wind energy projects and to fatalities estimated for other forms of energy generation (Smallwood 2017b). Researchers deliberated field methods concerning their support of accurate fatality estimation (Smallwood 2007, Johnson et al. 2016, Reyes et al. 2016, Smallwood et al. 2018, Kitano et al. 2020), and the estimation methods themselves (Korner ‐Nievergelt et al. 2011, Kitano and Shiraki 2013, Péron et al. 2013, Warren‐ Hicks et al. 2013, Smallwood et al. 2018). A new framework for testing the efficacy of bat, golden eagle ( Aquila chrysaetos), and bald eagle ( Haliaeetus leucocephalus ) impact ‐reduction strategies at wind energy projects (Sinclair and DeGeorge 2016) was also an important step in measuring and responding to utility ‐scale renewable energy impacts to volant wildlife. Much of what has been learned from research of wind energy impacts to wildlife can contribute to improved fatality monitoring at utility ‐scale solar projects, but solar projects also pose different risk factors, including to many species less vulnerable to wind energy impacts.

Once constructed, utility ‐scale solar projects pose multiple fatality risk factors. Volant wildlife can collide with solar collectors, power block st ructures, project buildings, medium ‐voltage overhead lines, gen ‐tie lines (i.e., generator lead or transmission lines), fencing, a nd automobiles servicing the project. Some birds might collide with photovoltaic (PV) panels because of the lake effect, or the birds' perception of many closely spaced PV panels as a waterbody onto which t hey attempt to land (Kagan et al. 2014). Polarized light from PV panels might attract prey of insectivorous birds an d hence the birds themselves (Horváth et al. 2010), or it might fool birds into trying to lick water from the panel while in flight (Horváth et al. 2009, 2010). Reflected self ‐images on mirrors of solar thermal projects, or even of PV panels, mi ght elicit aggressive responses of birds motivated to defend territory (Hager and Craig 2014, Kahle et al.
2016). Collisions might result from high ‐speed predator ‐ prey encounters in which the pursuer or pur sued collide with a project feature (Dunn 1993). Bats might fail to detect angled collector panels or mirrors because of reduced echolocation output (Gorresen et al. 2017, Corcoran and Weller 2018) or confused echolocation feedback (P. R. L ong, retired military pilot, personal communication). Bats might als o misinterpret echolocation ‐detected flat panels as water bodies from which they attempt to drink while in flight (Greif and Siemers 2010). At power tower projects, birds and bats die because of acute exposure to the zone of solar flux (Kagan et al. 2014). Data summarized herein indicate that birds also perish because of electrocution on energized portions of the project, entrapment or entanglement with project infrastructure such as fencing, and drowning in solar evaporation ponds. Until 2018, however, data needed to test hypothesized causal factors and to estima te regional fatalities were largely unavailable to the public.

Reviews of project ‐level fatality estimates were reported by those with exclusive access to reports and data (Kagan et al. 2014,C o o p e r2016, Walston et al. 2016, Kosciuch et al. 2020). These reviews, along with Harrison et al. ( 2016), discussed barrier effects, habitat loss, an d displacement of wildlife, but they focused on the lesser impacts of mortality caused by operable p rojects. In the reviews other than Harrison et al. ( 2016), fatality rates at operative projects were compared mostly as they had been reported, and only Walston et al. (2016) attempted to adjust all project ‐level fatality counts by carcass detection probabilities. I obtained and reviewed reports, data, environmental reviews, and com munications related to bird and bat fatality monitoring at California utility ‐scale solar energy projects (Table 1). The variety of study desig ns and methods I observed inspired my own estimations of project ‐level fatality estimates. I became most concerned over liberal exclu- sions of fatalities from project ‐level fatality estimates, use of automobi les in fatality searches, and placements of inappropriate carcasses to measure searcher detectio n and carcass persistence in carcass detection trials.

My first objective was to compare my independent estim ates to originally reported estimates of bird and bat SOLAR IMPACTS | 3o f2 9
TABLE 1 Solar projects for which fatality data were made available for bats and bird mortalities 1982 ‒2018 at solar projects in California, USA, in response to United States Freedom of Information Act and California Public Records Act requests in 2018
Project Ref a Countyb
Features
surveyedc MW Ha Year(s)
Other
surveysd
California Valley Solar Ranch 1 SLO PV, OL, G, F 258 476 2012 –2014 Bat, BM Topaz Solar Farm 2 SLO PV, OL 550 677 2012 –2016 Nest, Use, Bat, BM
Campo Verde Solar Project 3 Imp PV, G, F 139 584 2013 –2016 None Centinela Solar Energy Project 4 Imp PV, G, F 170 665 2014 –2015 None Imperial Solar Energy Center
South
5 Imp PV, G 130 383 2014 None
Calipatria Solar Project 6 Imp PV 20 64 2016 None
Midway Solar Farm II 7 Imp PV 30 113 2018 None
Desert Sunlight Solar Farm 8 Riv PV, G, F 550 1,700 2015 –2016 None Stateline Solar Farm 9 SB PV, G 300 682 2017 –2018 None McCoy Solar Energy Project 10 Riv PV, G, F 250 914 2016 –2017 None Blythe Solar Power Project 11 Riv PV, G, F 235 769 2016 –2017 None Genesis Solar Energy Project 12 Riv SCA, PB, P, G, F 250 728 2015 –2017 None Solar One 13 SB PT, HM, F, P 10 32 1982 –1983 Use
Ivanpah Solar Electric
Generating System
14 SB PT, HM, G, F 377 1,457 2013 –2018 Use
aReferences: 1 = H.T. Harvey and Associates ( 2013, 2015a); 2 = Althouse and Meade ( 2012, 2014); 3 = Heritage Environmental Consultants ( 2014, 2015a 2016); 4 = Heritage Environmental Consultants ( 2015b), Chambers Group ( 2016); 5 = UltraSystems ( 2014a–e); 6 = Heritage Environmental Consultants ( 2017a,b,c); 7 = Shoener and Barrett's Biological Surveys ( 2018); 8 = Western EcoSystems Technology ( 2017a 2018a); 9 = Dudek ( 2018a); 10 = Martinson et al. ( 2018b); 11 = Doering and Santistevan ( 2013), Martinson et al. ( 2018a); 12 = Western EcoSystems Technology ( 2016, 2017c); 13 = McCrary et al. ( 1986); 14 = H.T. Harvey and Associates ( 2015b), Western EcoSystems Technology ( 2017b, 2018b, 2019).
bSLO = San Luis Obispo, Imp = Imperial, Riv = Riverside, SB = San Bernardino. cPV = photovoltaic panels mounted on ground ‐based metal frames, OL = medium voltage overhead lines, G = gen ‐tie, or generation tie‐ins to the nearest substation or transmission lines, F = perimeter security fence, SCA = solar collector arrays composed of parabolic mirrors, PB = power blocks, P = solar evaporation ponds, PT = power towers receiving reflected sunlight from heliostat mirrors, and HM = heliostat mirrors arranged to reflect solar energy to receiver on power tower. dBat = bat activity surveys performed using acoustic detectors; BM = background mortality surveys consisting of fatality searches in adjacent areas lacking solar project infrastructure; nest = nest survey, use = visual scans to detect and count birds during daylight hours.
fatality rates at monitored utility ‐scale solar projects. I predicted my estimates would usually be higher. My second objective was to estimate statewide fatalitie s of birds and bats caused by t he installed capacity of utility ‐solar projects in 2020. My third objective was to apport ion collision fatality impacts to project features to reveal which features warrant greatest concern. My fourth objective was to compare fatality rates among bird and bat species and among those with special status to identify species in relatively greater need of research and mitigation. My fifth objective was to ide ntify additional sources of error and bias that warrant further research to increase accuracy of fatality estimates. 4o f2 9 | SMALLWOOD

STUDY AREA
The fatality monitoring studies I reviewed had been performed from 1982 ‒2018 at 14 utility ‐scale solar energy projects on relatively flat terrain in annual grassland, desert, and agricultural environments within the desert of California (Figure 1; Table 1). Elevations ranged −4 m to 983 m. Climate was Mediterranean, with warmest tem- peratures and least rainfall during June –August, coolest temperatures and greatest rainfall during No- vember–February, and transitional periods with increasing temperatures and decreasing rain in March –May, and deceasing temperatures and increasing rain during September –October. Twelve projects were located in Mojave and Sonoran deserts, where annual rainfall averaged <10 cm, and natural ground cover was desert scrub including creosote ( Larrea tridentata ). Two project sites were in San Luis Obispo County, where annual rainfall averaged 10–38 cm, and natural ground cover was annual grassland composed of red brome ( Bromus madritensis rubens ), soft chess ( Bromus hordeaceus ), foxtail barley ( Hordeum murinum ), six ‐weeks fescue ( Festuca octoflora ) and wild oat (Avena fatua .). Wildlife typical of the desert sites included coyote ( Canis latrans ), desert kit fox ( Vulpes macrois ), bighorn sheep ( Ovis canadensis ), black ‐tailed jackrabbit ( Lepus californicus ), canyon bat ( Parastrellus hesperus ), common raven ( Corvus corax), desert sidewinder ( Crotalus cerastes), and Agassiz desert tortoise ( Gopherus agassizii).

FIGURE 1 Locations and rated capacities (MW) of solar energy project sites that were monitored for wildlife fatalities in California, 1982 ‒2018: 1 = California Valley Solar Ranch, San Luis Obispo County, 2 = Topaz Solar Farm, San Luis Obispo County, 3 = Campo Verde Solar Project, Imperial County, 4 = Centinela Solar Energy Project, Imperial County, 5 = Imperial Solar Energy Center South, Imperial County, 6 = Calipatria Solar Project, Imperial County, 7 = Midway Solar Farm II, Imperial County, 8 = Desert Sunlight Solar Farm, Riverside County, 9 = Stateline Solar Farm, San Bernardino County, 10 = McCoy Solar Energy Project, Riverside County, 11 = Blythe Solar Power Project, Riverside County, 12 = Genesis Solar Energy Project, Riverside County, 13 = Solar One, San Bernardino County, 14 = Ivanpah Solar Electric Generating System, San Bernardino County SOLAR IMPACTS | 5o f2 9
Wildlife typical of the grassland sites included coyote, San Joaquin kit fox ( Vulpes macrotis mutica ), black ‐tailed jackrabbit, giant kangaroo rat ( Dipodomys ingens ), and northern Pacific rattlesnake ( Crotalus oreganus oreganus ).

All monitored projects had been graded flat and natural vegetation had been removed. Of the 3 solar energy technologies monitored, 1 technology included flat PV panels on ground ‐mounted frames with solar trackers at most projects. The other 2 technologies were solar thermal, including solar ‐trough parabolic mirrors referred to as solar collector arrays (SCAs) at Genesis Solar Energy Project, and heliostat mirrors reflecting light to receivers, or boilers, atop power towers at Ivanpah Solar Electric Generating System (Ivanpah) and the since ‐decommissioned Solar One project. Solar projects connected to the grid via gen ‐ties, which resemble transmission lines and their towers. Solar projects were also surrounded by perimeter security fences. Some projects also included medium ‐ voltage overhead lines. Monitored projects totaled 3,259 MW, including 627 MW of solar thermal and 2,632 MW of PV. They included 163.35 km of security fence bordering 2,237 MW of solar arrays (0.073 km/MW) and

78.09 km of gen ‐tie serving 2,237 MW of solar arrays (0.037 km/MW) that were monitored for fatalities.

METHODS
Through use of California Public Records Act (PRA) and federal Freedom of Information Act (FOIA) requests to agencies with regulatory authority over biological resources, I obtained reports, data, environmental reviews, and communications related to fatality monitoring at utility ‐scale solar projects in California. Field and analytical methods of fatality monitoring varied among reports but all were intended to estimate the number of fatalities caused by the project. Monitors searched for fatalities by following protocols designed to account for proportions of fatalities not found, to extrapolate from sampling units to whole projects, and to compare fatality rates between project features and whole projects. Although monitors used various fatality estimators, the basic form was F =
F
ρ
 ,
where F was the fatality estimate, F was the number of fatalities found, and ρ represented carcass detection probability.

Monitors typically searched by foot or by car for fatalities according to a periodic schedule and along evenly spaced transects across sampled portions or all of a solar project. At 6 projects, monitors applied distance sampling from transects. They usually removed all fatality finds upon discovery. Most performed carcass detection trials, in which testers placed carcasses on the study site to quantify searcher detection rates and carcass persistence time, or the time until the placed carcass was no longer detectable by searchers. Outcomes of carcass detection trials were used to estimate ρ. For fatality estimation, monitors usually omitted fatalities found incidental to scheduled searches or outside search areas. Some monitors set 5 or 10 fatality detections/species as thresholds for species' inclusion, and some separately reported fatality estimates between known and unknown cause of death. A few reports provided estimates for 10 species with the highest fatality rates.

From information I received following PRA and FOIA requests, I constructed data sets I needed to in- dependently estimate fatalities at monitored solar projects (fatality data and their reported attributes, along with definitions of data fields [Table S1], are available in Supporting Information). I recorded each study's start date, duration of monitoring of each project feature, proportion of each project feature searched, whether searched by walking or car, transect width, whether distance sampling was used, fatality search interval, and fatality metric (Table 2). I recorded essential attributes of carcass detection trials that monitors used to represent ρ, including taxa, carcass source, body size classes into which trial carcasses were grouped, trial duration, and metric used to estimate the probability of carcasses persisting to the next search ( r), which could be derived from R = proportion of car- casses remaining on the ith day into a trial, R
C = mean daily carcass persistence upon the ith day into a trial, and t̅ = mean days to carcass removal (Table S2, available in Supporting Information). Some of the monitors' detection trials included placements of feather piles as a separate class from small and large birds. I did not use the results of such trials because all fatalities found as feather piles began as complete carcasses. Starting a trial from a feather pile, as if the rest of the carcass never existed, was unrealistic. 6o f2 9 | SMALLWOOD
TABLE 2 Bird and bat fatality search methods reported at utility ‐scale solar projects in California, USA, 1982–2018
Project Feature a Yrs Sample (%) Search Transect (m) b
I (days) in spring,
summer, fall,
winterc Fˆ metricd
CVSRe Gen‐tie 2.4 100 Walk 15 7, 7, 7, 7 F/(S × r)
CVSRe MVOH 1.8 100 Walk 18 7, 7, 7, 7 F/(S × r)
CVSRe Fence 2.0 20 Walk 3 7, 7, 7, 7 F/(S × r)
CVSRe PV 2.0 20 Walk 5 7, 7, 7, 7 F/(S × r)
Topaz MVOH 5.0 3.3 Walk 2.13 1 | 30 F/plot/search
Topaz PV 5.0 3.5 Walk 2.3 1 | 30 F/plot/search
Campo Verde Fence 2.5 100 Walk 1 | 30 None
Campo Verde Gen ‐tie 2.5 30 Walk 7.5 1 | 30 None
Campo Verde PV 2.5 30 Walk 1 | 30 None
Centinela Gen ‐tie 0.8 50 Walk 5 3.5, 3.5, 3.5, 3.5 F/(S × R C)
Centinela Fence 1.0 100 Walk 1 | 30
Centinela PV 1.0 10 Walk 4.2 1 | 30 F/0.2 yr
Imperial Gen ‐tie 0.4 100 Car 1 | 30 None
Imperial PV 0.4 10 Car 14.5 1 | 30 None
Calipatria PV 2.0 100 Walk 35 7, 21, 7, 21 Distance
f
Midway PV 0.4 Car 60 7, 0, 18, 0
Desert Sunlight Fence 1.0 7 Car 7, 23, 8, 19 F/(S × r)
Desert Sunlight Gen ‐tie 1.0 48 Walk 15 7, 23, 8, 19 F/(S × r) Desert Sunlight PV 1.0 30 Walk 70 7, 23, 8, 19 F/(S × r ×A ) Stateline Fence 1.0 100 Car 3 7, 21, 7, 21 F/(S × r)
Stateline Gen ‐tie 1.0 50 Car 15 7, 21, 7, 21 F/(S × r) Stateline PV 1.0 40 Car 120 7, 21, 7, 21 F/(S × r ×A )
McCoy Fence 2.0 98 Car 3 7, 21, 7, 21 F/(S × r)
McCoy Gen ‐tie 2.0 25 Walk 15 7, 21, 7, 21 F/(S × r)
McCoy PV 2.0 45 Car 160 7, 21, 7, 21 F/(S × r ×A )
Blythe Fence 2.0 92 Car 3 7, 21, 7, 21 F/(S × r)
Blythe Gen ‐tie 2.0 25 Walk 15 7, 21, 7, 21 F/(S × r ×A ) Blythe PV 2.0 41 Car 95 7, 21, 7, 21 F/(S × r ×A )
Genesis Fence 2.0 100 Car ? 7, 20, 7, 19 F/(S × r)
Genesis Gen ‐tie 2.0 25 Walk 15 7, 20, 7, 19 F/(S × r)
Genesis Ponds 2.0 100 Walk 7, 20, 7, 19 F/(S × r)
Genesis PBlocks 2.0 100 Walk 7.5 7, 20, 7, 19 F/(S × r) Genesis SCAs 2.0 30 –50 Car 30 7, 20, 7, 19 F/(S × r ×A ) (Continues)
SOLAR IMPACTS | 7o f2 9
Where monitors used distance sampling, I derived probabilities of searcher detection of trial carcasses within distance increments from transects, which were reported graphically. I multiplied each distance increment by transect length to calculate the search area, and hence the search area as a proportion of the project area ( A) associated with a probability of searcher detection. Because all but the gen ‐tie at Centinela did not report distances of fatalities from the transect, I usually could not assign fatalities to the distance increment for which detection probabilities had been characterized by trials. I therefore averaged S × A among the distance increments searched, where S was the proportion of placed trial carcasses found within the ith distance increment (of those available), and A was the searchable area between the car and the farthest point of the ith distance increment. Along with r, S × A composed ρ: F =
F
rSA××
 .

Where carcass detection trials were not implemented, insufficiently described, or based on <8 carcass placements/ category of bats, small birds, and large birds (i.e., California Valley Solar Ranch [CVSR], Campo Verde, Centinela, Imperial, S o l a rO n e ) ,Ia p p l i e dd e t e c t i o nr a t e sf r o mm yo w nt r i a l so fc a r c a s s e sa sp a r to f23‐year fatality monitoring studies in the Altamont Pass Wind Resource Area (APWRA). Given its short ‐stature annual grassland cover, carcass visibility to sca- vengers and fatality searchers should be similar between the APWRA and the 5 project sites to which I applied APWRA‐ derived carcass detection rates, but to minimize differences, I relied solely on trial carcasses that had been placed on open ground to simulate conditions of ground visibility at the 5 solar project sites. The APWRA trials were integrated into routine fatality monitoring with placements that were periodic throughout eac h study and randomized within fatality search areas. In 1 study, trials included 276 bird carcasses of 79 species that ranged in mass from 2 g to 6,650 g (Smallwood et al.2017a, 2018). In the other study, trials included 764 bird carcasses of 100 species ranging in mass from TABLE 2 (Continued)
Project Feature a Yrs Sample (%) Search Transect (m) b
I (days) in spring,
summer, fall,
winterc Fˆ metricd
Solar One Heliostats 0.8 100 Walk 7, 0, 7, 7 F/( r)
Solar One Tower 0.8 100 Walk 7, 0, 7, 7 F/( r)
Ivanpah Fence 2.0 100 Car 3 7, 25, 7, 25 F/(S × r)
Ivanpah Gen ‐tie 2.0 100 Car 15 7, 25, 7, 25 F/(S × r)
Ivanpah Heliostats 2.0 24 Walk 5 7, 25, 7, 25 F/(S × r) Ivanpah Heliostats 2.0 8 Walk 7, 0, 0, 21 F/(S × r)
Ivanpah Tower g 2.0 100 Walk 7, 25, 7, 25 F/(S × r)
Ivanpah Tower g 100 Walk 7, 21, 7, 21 F/(S × r)
aPV = photovoltaic panels mounted on ground ‐based metal frames, MVOH = medium voltage overhead lines, Gen ‐ tie = generation tie ‐ins to the nearest substation or transmission lines, Fence = perimeter security fence, SCA = solar collector arrays composed of parabolic mirrors, PBlocks = power blocks, Ponds = solar evaporation ponds, Tower = power towers receiving reflected sunlight from heliostat mirrors, and Heliostats = mirrors arranged to reflect solar energy to a receiver on power tower.
bWidth of search area to one side of transect.
c1 | 30 refers to daily searches for 7 consecutive days beginning again every 30 days. dFˆ = estimated number of fatalities, F = number of fatalities found, S = proportion of trial carcasses found, r = proportion of trial carcasses persisting to next search, A = proportion of search area that was searchable; empty spaces indicated the reporting was unclear which metric was used.
eCalifornia Valley Solar Ranch. Also performed daily searches for 5 consecutive days once every 4 weeks. fUsed Program Distance (Miller 2016), but none of the models fit the data well. gIncluded cleared ground around towers, power blocks, and inner rings of high ‐density heliostat mirrors. 8o f2 9 | SMALLWOOD
3 g to 6,000 g, and 144 bat carcasses of 16 species ranging in mass from 2.3 g to 23.6 g (Brown et al.2016). Trial carcasses were left indefinitely where placed. Whether searchers ever found the carcass determined each trial outcome. I logit‐ regressed trial outcomes of found or not found on log10 body mass (g) to estimate overall detection probabilityD:

D e
e= 1+ ,
abl o g M
abl o g M
− +×
− +×
10 10

where a and b were fitted coefficients andM w a sb o d ym a s s( F i g u r e2). I estimatedD separately for birds and bats, and among the 3 fatality search intervals that were implem ented between the 2 APWRA studies: 5, 7, and 28 days. To simulate a daily search interval, I estimatedD from carcass trial placements 1 day prior to the next search (Smallwood 2017a).
(A) (B)
FIGURE 2 Hypothetical fatality estimation from 1 bird of a 10 ‐g species found at a site where monitors used 78‐g bird carcasses to represent small birds of ≤100 g body mass in on ‐site carcass detection trials, and then calibrated to carcass detection trial outcomes on open ground in the Altamont Pass Wind Resource Area (APWRA), California, USA, 2012 ‒2015, where carcass detection probability was modeled to variation in body mass. A) Predicted probability of carcass detection (
D) and 95% confidence intervals based on outcomes of detection trials performed at 7 ‐day intervals and logit ‐regressed on log 10 body mass, and overlayed by overall detection measured for 78 ‐g bird carcasses in on ‐site trials at 2 sites (shaded and striped diamond symbols) and in the APWRA (open square) and calibrated to D of the 10 ‐g bird that was found (oversized circle among mean D). Also overlaid are the detection probabilities derived from trials with 78 ‐g birds calibrated to that of the 10 ‐g bird (circles patterned to match symbols representing uncalibrated detection probabilities). B) Resulting fatality estimates ( F) derived from 3 types of adjustment: conventional trials for carcass persistence r and searcher detection S using 78 ‐g birds to represent small birds, D of a 10 ‐g bird, and the calibration bridge to introduce the effect of body size on detection probability while preserving the site effect: C = rS go ns i t e D
rS APWRA
1 (× ) 78 (× )
g
g
10 78










SOLAR IMPACTS | 9o f2 9
Although logit regression calculates asymptotic confidence intervals along with predicted detection prob- abilities of each body mass used to develop the model, the body masses of fatality finds can differ from those of the trial carcasses used to develop the model. To extend the model's confidence intervals to fatality finds, I fit nonlinear regression models to logit regression's 95% lower and upper confidence limits (Smallwood et al. 2018).

To the 5 studies at solar projects in need of carcass detection probabilities from my offsite studies, D served as
ρ in the fatality estimator: F =
F
D

 . I assumed that detection probabilities of carcasses placed on open ground depended more on variation in carcass size than on variation in carcass removal rates among project sites. Spatial variation in carcass removal rates remains unknown, but the effect of body mass on detection probability is strong (Figure 2). Given the strong effect of body mass, D served a second purpose in this study. Because I monitored trial carcasses in the APWRA for availability to be found by searchers so that I could also estimate fatalities using the conventional terms r × S, I had the means to calibrate r × S to D , the importance of which is discussed below.

Most fatality monitors at solar projects deployed birds in detection trials that were larger than most of the birds and bats found as fatalities. This practice emerged as a potential bias because the placed birds represented broad size categories in which the body size ‐specific detection probabilities of the placed birds would be increasingly discrepant from the detection probabilities of increasingly smaller or larger species found as fatalities in the same broad size category (Figure 2). I therefore adjusted the results of detection trials to account for the effects of variation in body mass. I used integrated, overall detection trials performed in the APWRA to derive a unit ‐free scaler to adjust for the effects of variation in body mass on trial outcomes. I formulated a body mass scaler C as
follows:

C F
rS
D
rS= ×× ith solar project APWRA open ground,




















 
With C , I adjusted number of fatalities found for overall detection probability r × S based on the monitors' onsite trials, and then I adjusted fatalities by the ratio of overall detection probability D to overall detection probability r × S derived from the same conventional trials using the same ‐sized carcasses as at the solar project. In these cases, C served as ρ. In the dividend, r × S were from the same ‐sized species as r × S in the divisor, thereby serving to bridge the 2 sites, 1 site where D was additionally measured along a continuous range of body mass and the other site where F was additionally measured among species that varied continuously in body mass (Figure 2). This bridge preserved any differences in r and S between the sites. I applied this bridge to onsite detection trials and fatality data from Topaz, Desert Sunlight, McCoy, Midway, Calipatria, Blythe, Genesis, and Ivanpah. I was unable to apply this bridge to data from Stateline, to which I relied solely upon onsite detection trials.

I applied the body mass scaler to bird and bat fatalities where trial outcomes had been lumped into broad size categories. Monitors had performed detection trials for bats at only 1 solar project, but bats were represented by small birds in those trials. The scaling function I used for bats was D for bats relative to r × S for whichever bird species in my study represented small birds at the solar project, and again I relied only on bats and small birds in my study that had been placed on open ground (no occlusion) at Vasco Winds (Brown et al. 2016) or Sand Hill (Smallwood et al. 2018). To represent trial carcasses at solar projects, I selected APWRA bird trial carcasses ranging in size 24 –32 g for house sparrows ( Passer domesticus ), 36 –47 g for brown ‐headed cowbirds ( Molothrus ater ), 50–70 g for Japanese quail ( Coturnix japonica ) chicks, 108 –168 g for Japanese quail adults, 260 –400 g for rock pigeons ( Columba livia ), 1,050 –1,150 g for mallards ( Anas platyrhynchos ), and 1,200 –2,300 g for ring ‐necked pheasants ( Phasianus colchicus ). For application to fatality monitoring based on 21 ‐day search intervals, I averaged D between 7 ‐ and 28 ‐day intervals on open ground at Vasco Winds.

Only at CVSR were found fatalities sufficiently assigned to sampling units so that other researchers could estimate standard error attributable to variation among sampling units. Consistent with Smallwood et al. ( 2020), I used the delta method to estimate confidence intervals at CVSR. Otherwise, I relied on logit ‐regressed estimates of confidence intervals to represent confidence intervals of fatality estimates. I estimated fatalities of every species 10 of 29 | SMALLWOOD
represented by ≥1 fatality at operational solar projects. I assumed solar projects caused all fatalities found by monitors, but natural causes likely contributed to a few of the fatalities.

Although I endeavored to reduce the effects of bias on ρ by factoring in fatality detection probability as a function of interspecific variation in body mass, fatality estimates remained amply prone to error and bias (Smallwood 2007). To improve accuracy of fatality estimates going forward, I addressed 3 additional sources of error and probable bias. First, I compared monitoring outcomes between pedestrian and car surveys. I assumed bias would be indicated, though not verified, if estimated fatality rates differed between projects searched by foot and by car. Second, to indicate whether and to what degree wildlife species were represented in fatality monitoring results at solar projects, I related the cumulative number of species represented as detected fatalities by months since monitoring was initiated, similar to the approach used by Beston et al. ( 2015). To the fatality monitoring data of each project, I attempted to fit a nonlinear model for the purpose of identifying the asymptote of the cumulative number of species represented, and to determine whether the predicted asymptote had indeed been reached.

Third, I tallied fatalities that monitors had assigned to taxa or size groups other than the species level.

RESULTS
Attributes of fatality monitoring studies
Fatality monitoring studies varied greatly in attributes that affect fatality estimates (Tables 2, S1). As examples, fatality searches covered 3.3% to 100% of solar project features, and spanned 0.42 to 5 years in duration. Fatality search intervals ranged 1 to 25 days, but at 2 projects searches were daily for 5 to 7 consecutive days beginning at intervals of 28 to 30 days, respectively. Areas searched from transects varied within maximum distances of 2.1 to 160 m—a7 6‐fold range. Distance sampling was used from a transect along the gen ‐tie at Centinela, from transects intersecting PV arrays at Centinela, Imperial, Calipatria, Midway, Desert Sunlight, Midway, McCoy, and Blythe, and from transects intersecting SCAs at Genesis. The fatality metric at Topaz was fatalities/plot/search, which was incomparable to fatality metrics reported in other studies. No fatality estimates were reported at Campo Verde and Imperial, and none were reported for bats at Ivanpah.

Studies varied greatly in carcass detection trials needed to accurately represent ρ. Bat detection trials were implemented only at Genesis, where small birds served as surrogates for bats (Table S2). Bird carcasses were placed in all detection trials except for trials along the Centinela gen ‐tie, where colorful bird ‐shaped holiday ornaments were placed in searcher detection trials. Source of bird carcasses was unreported at all studies except CVSR, which used carcasses from wildlife rehabilitation facilities and fatalities found during monitoring. Two studies neglected to identify species used in detection trials. Monitors at Solar One and Calipatria did not perform searcher detection trials. Monitors at Centinela and Imperial did not perform detection trials of any type. Duration of carcass per- sistence trials ranged 5 to 42 days. The carcass persistence metric varied but was usually mean days to carcass removal (Table S2).

The majority of birds killed at solar projects were small ‐bodied; most were smaller than the bird carcasses used to represent them in detection trials (Figure 3). Of bird species found as fatalities among solar projects, the typical adult body mass of 30% were <22 g, 50% were <42 g, 63% were ≤100 g, and only 22% were at least as large as rock pigeons. Of bird species found as fatalities at 11 PV projects, the typical adult body mass of 30% were ≤25 g, 50% were <56 g, 57% were <100 g, and only 25% were at least as large as rock pigeons. Among 3 projects (McCoy, Blythe, and Desert Sunlight), the smallest species used in small ‐bird persistence trials was larger than 73% of the small birds found as fatalities (Figure S1, available in Supporting Information). The smallest species used in medium ‐ bird persistence trials was larger than 33% of the medium bird species found. The largest species used in large ‐bird persistence trials was smaller than 50% of the large birds found as fatalities. Accounting for the detection prob- abilities of fatalities of species smaller than detection trial carcasses increased fatality estimates of many small ‐bird SOLAR IMPACTS | 11 of 29
species, but it reduced estimates of a much smaller number of large ‐bird species (Figure S2, available in Supporting Information). The onsite ‐adjusted fatality estimates bridged to the APWRA's integrated overall detection trials related linearly to onsite ‐adjusted fatality estimates without the bridge, indicative of an unbiased adjustment for interspecific variation in body mass (Figure S3, available in Supporting Information).

Body‐size adjusted fatality estimates
My bird fatality estimates exceeded originally reported estimates by factors of 1.07 at Campo Verde, 1.5 at Desert Sunlight, 1.88 at CVSR, 4.59 at McCoy, 6.59 at Blythe, 10.49 at Genesis, 10.5 at Calipatria, and 18 –72 at Topaz (Table 3). No project ‐wide fatality estimates were reported at Ivanpah, but my estimates for birds at Ivanpah's heliostat mirror arrays and in the tower area exceeded originally reported estimates by factors of 10 and 3.8, respectively.

My estimate of bird fatalities among Ivanpah's heliostat mirrors exceed that of the Tower Area by a factor of 6.2 (Table 3; Figure S4, available in Supporting Information). Some unknown portion of the fatalities found among heliostat mirrors was likely caused by exposure to the zone of solar flux. Bird fatality rates among mirrors at Genesis averaged only 20% of those among heliostat mirrors at Ivanpah. Bird mortality was highest at solar thermal projects (Table 3). On a per ‐MW basis, the gen ‐tie was the most dangerous project feature to birds at some projects, whereas the PV panels or mirrors were the most dangerous at others. On a project ‐wide basis, however, more birds died by collision with solar collectors —mirrors and PV panels. (A) (B)
FIGURE 3 Body size ‐distribution of bird fatalities relative to typical body mass of species used in searcher detection and carcass persistence trials (dashed vertical lines) among 13 California solar projects (Solar One not included) monitored 2012 ‒2018, where the clear zone represents small birds ≤100 g, light gray represents medium birds >100 g to ≤1,000 g, and dark gray represents large birds >1,000 g: A) proportion of estimated number of fatalities (
Fˆ)b y1 0 ‐g increments of typical species' body mass, and B) estimated annual fatalities by body mass on a continuous scale. Carcasses of many found fatalities were smaller than the species used in on ‐site detection trials 12 of 29 | SMALLWOOD
TABLE 3 Reported estimates and my own estimates of wildlife fatality rates at California utility ‐scale solar energy projects and project features through 2020. I cannot explain the estimates that were not reported by monitors, but I did not estimate fatalities caused by medium ‐voltage overhead lines and PV at Topaz because these facilities overlapped to a degree that I could not determine which caused the fatalities Fˆ/MW/year (95% CI)
Project Taxa Feature a Reportedb This study
CVSR Birds Gen ‐tie 0.38 (0.32 –0.48) 1.27 (0.89 –1.93) CVSR Birds MVOH 0.10 (0.08 –0.12) 0.35 (0.26 –0.48)
CVSR Birds Fence 0.72 (0.60 –0.89) 1.15 (0.85 –1.64)
CVSR Birds PV 8.97 (7.33 –11.49) 16.17 (11.98 –22.05)
CVSR Birds Project 10.07 (8.20 –12.92) 18.98 (14.00 –26.16) Topaz Birds MVOH 0.0546 –0.2159c
Topaz Birds PV 0.0491 –0.1909c
Topaz Birds Project 0.1037 –0.4068c 7.48 (6.12 –11.22)
Campo Verde Birds Fence 0.10 (0.08 −0.14)
Campo Verde Birds Gen ‐tie 0.50 (0.39 −0.78)
Campo Verde Birds PV 20.49 (16.52 −31.07)
Campo Verde Birds Project 19.65 21.09 (16.99 −32.00)
Centinela Birds Fence 1.82 (1.49 –2.77)
Centinela Birds Gen ‐tie 1.68 c 2.31 (1.7 –3.82)
Centinela Birds PV 12.93 22.92 (18.52 –34.71)
Centinela Birds Project 28.64 (23.09 –43.76)
Imperial Birds Project 9.70 (8.22 –14.65)
Calipatria Birds PV 2.1 22.09
Calipatria Birds Project 2.1 22.09
Midway Birds Project 3.22
Desert Sunlight Birds Fence 0.015 (0.005 –0.042) 0.03 (0.02 –0.03) Desert Sunlight Birds Gen ‐tie 1.86 (0.87 –4.99) 2.58 (1.68 –4.32) Desert Sunlight Birds PV 1.05 (0.88 –1.56) 1.68 (1.24 –3.19) Desert Sunlight Birds Project 2.93 (2.03 –6.67) 4.39 (3.01 –7.70) Stateline Birds Gen ‐tie 0.457 0.19
Stateline Birds PV 0.387 1.94
Stateline Birds Project 2.13
McCoy
d Birds Fence 0.096
(0.016–0.20/8)

0.25 (0.17 –0.36)
McCoyd Birds Gen ‐tie 3.124 (2.36 –4.04) 11.68 (7.19 –21.13) McCoyd Birds PV 0.23 (0.04 –0.49) 3.88 (2.58 –6.04)
McCoyd Birds Project 3.45 (2.66 –4.39) 15.83 (9.98 –27.60) (Continues)
SOLAR IMPACTS | 13 of 29
TABLE 3 (Continued)
Fˆ/MW/year (95% CI)
Project Taxa Feature a Reportedb This study
McCoyd Bats Project 0.46 (0.01 –0.73)
Blythed Birds Fence 0.09 (0.03 –0.17) 0.38 (0.28 –0.55) Blythed Birds Gen ‐tie 3.32 (2.51 –4.30) 16.35 (10.07 –28.96) Blythed Birds PV 0.18 (0.009 –0.455) 6.67 (4.52 –5.75)
Blythec Birds Project 3.59 (2.76 –4.60) 23.65 (15.04 –35.62) Genesis Bats Fence 0.012 (0.002 –0.030) 1.67 (0.11 –4.27) Genesis Bats Ponds 0.036 (0.008 –0.078) 4.82 (0.13 –7.90) Genesis Bats PBlocks 0.002 (0.002 –0.006) 2.40 (0.09 –4.77) Genesis Bats SCAs 0.052 (0.020 –0.092) 0.56 (0.09 –3.53) Genesis Bats Project 0.106 (0.032 –0.198) 10.92 (0.48 –22.98) Genesis Birds Fence 0.39 (0.25 –0.56) 5.63 (4.21 –7.70) Genesis Birds Gen ‐tie 1.50 (0.66 –2.73) 17.62 (11.26 –31.09) Genesis Birds Ponds 0.39 (0.14 –0.72) 5.57 (3.98 –8.09) Genesis Birds PBlocks 0.21 (0.13 –0.29) 4.93 (3.51 –7.06) Genesis Birds SCAs 2.41 (1.90 –3.10) 12.93 (9.77 –17.86) Genesis Birds Project 4.56 (3.44 –5.84) 47.83 (33.53 –73.59) Solar One Birds Heliostats 26.20 (18.70 –58.94)
Solar One Birds Tower area 16.24 (9.24 –92.42)
Solar One Birds Project 9.88 –11.44 42.44 (27.94 –151.36) Ivanpah Birds Fence 0.19 (0.13 –0.28) 0.58 (0.41 –0.83) Ivanpah Birds Gen ‐tie 0.005 0.14 (0.09 –0.25)
Ivanpah Birds Others 0.82 (0.45 –5.23)
Ivanpah Birds Heliostats 6.41 (4.66 –10.38) 63.91 (40.45 –141.33) Ivanpah Birds Tower area 2.69 (1.82 –4.46) 10.30 (5.79 –52.95) Ivanpah Birds Project 75.74 (47.19 –200.59)
Ivanpah Bats Tower area 1.79 (0.08 –3.70)
aProject = all features of the project, Fence = perimeter security fence, Ponds = solar evaporation ponds, Gen ‐ tie = generation lead or transmission lines, MVOH = medium ‐voltage overhead collector lines, PV = solar photo ‐voltaic panels; and in solar thermal projects, Heliostats = heliostat mirrors, SCAs = solar collector arrays, Tower area = bare ground between SCAs and the power tower, PBlocks = power blocks, and Others = ancillary structures such as operations and maintenance buildings.
b90% CI at California Valley Solar Ranch (CVSR) and Genesis. cI calculated the value based on the reported estimate in terms of a different metric. dI assumed methods remained unchanged in year 2, but the second ‐year report was unavailable. Martinson et al. ( 2018a,b) reported results only for year 1, whereas I report results for both years 1 and 2. 14 of 29 | SMALLWOOD
Bat mortality was highest at Genesis, followed by Ivanpah and McCoy. At Genesis, the most dangerous project feature to bats was the solar evaporation ponds, followed by power blocks, fences, and SCAs (Table 3). At Ivanpah, bat fatality estimates were possible only in the tower area, where mortality declined over the 5 years of monitoring (Figure 4). Species exhibiting the largest declines were canyon bat and California myotis ( Myotis californicus ). An ultrasonic bat deterrence method was implemented at unit 1 on 10 September 2014, and at units 2 and 3 on 23 April 2015.

California‐wide impacts of operative utility ‐scale solar projects According to the California Energy Commission, 1,948.8 MW of solar thermal projects operated in California in 2020, and 12,220 MW of PV had been installed by the end of 2019. On average, utility ‐scale solar energy projects required 3.48 ha/MW for solar thermal and 2.67 ha/MW for PV. Therefore, utility ‐scale solar thermal projects covered 6,782 ha, and utility ‐scale solar PV covered 32,627 ha. Projected to California's installed capacity of all utility‐scale solar projects in 2020, mean annual fatalities at monitored projects predicted 267,732 (95% CI = 186,071–495,391) bird fatalities and 11,418 (95% CI = 601 –23,926) bat fatalities (Table S3). Only 8 studies re- ported project ‐wide bird fatality estimates, and only 2 of these reported a confidence interval. Only 2 studies at solar thermal projects reported project ‐wide bat fatality estimates. Projected to California's installed capacity of all utility‐scale solar projects in 2020, originally reported mean annual fatalities would predict 82,471 birds, but original reporting was insufficient for including a 95% confidence interval or for predicting statewide bat fatalities.

At utility ‐scale solar PV projects, my estimates of F/MW/year averaged 11.61 (95% CI = 8.37 –17.56) birds and 0.06 (95% CI = 0.01 –0.10) bats (Table S4, available in Supporting Information). Projected to FIGURE 4 Estimates of annual bat fatalities in the tower area of Ivanpah Solar Electric Generating System, San Bernardino County, California, USA, 2013 ‒2018, spanning the cleared ground between the power tower and the mirror arrays and including all bats as a group, Mexican free ‐tailed bats ( Tadarida brasiliensis ), canyon bats, and California myotis. Arrows point to years following installations of ultrasonic deterrents at power towers, 1 installation after the first year, and 2 more during the second year. Declines in bat mortality preceded installations of deterrents
SOLAR IMPACTS | 15 of 29
California's installed capacity of 12,220 MW of PV in 2019, my fatality rates pred icted annual fatalities of 141,811 (95% CI = 102,227 –214,593) birds and 716 (95% CI = 124 –1,221) bats. At solar thermal projects my estimates of F/MW/year averaged 64.61 (95% CI = 41.74 –149.95) birds and 5.49 (95% CI = 0.25 –11.65) bats (Table S4). Projected to California's installed capacity of 1,948.8 MW of solar thermal in 2020, my fatality rates predicted annual fatalities of 125,921 (95% CI = 81,346 –292,225) birds and 10,701 (95% CI = 477–22,705) bats.

Along perimeter fences, I estimated F /km/year averaged 14.44 (95% CI = 10.88 –20.34) birds and 2.56 (95% CI = 0.17–6.54) bats (Table S4). Projected to California's installed 900 km of fencing around solar projects in 2020, these rates predicted annual fatalities of 13,005 (95% CI = 9,802 –18,324) birds and 2,304 (95% CI = 156 –5,892) bats. I estimated fatalities along fences averaged almost 5% of birds and 20% of bats killed at solar projects.

Along gen‐ties, I estimated F/km/year averaged 113.16 (95% CI = 71.78 –198.42) birds and no bats (Table S4).

Projected to California's installed 461 km of gen ‐ties at solar projects in 2020, these rates predicted annual fatalities of 52,162 (95% CI = 33,087 –91,463) birds. I estimated fatalities along gen ‐ties averaged 19% of birds and 0% of bats killed at solar projects.

Projected to the 2019 installed capacity, utility ‐scale solar PV annually killed more mourning doves ( Zenaida macroura) than any other species at 17,043, followed by 10,082 horned larks ( Eremophila alpestris ), and 7,628 western meadowlarks ( Sturnella neglecta ), and thousands of other grassland birds (Table S4). The PV panels killed 5,362 American coots ( Fulica americana ), 4,755 soras ( Porzana carolina ) among many other rails, grebes, and wa- terfowl, and 2,224 burrowing owls ( Athene cunicularia ), 958 American kestrels ( Falco sparverius ), and hundreds of other raptors. They also killed 1,486 common yellowthroats ( Geothlypis trichas ), 1,298 yellow warblers ( Setophaga petechia), and thousands of additional likely migrants.

Projected to the 2020 installed capacity, utility ‐scale solar thermal projects annually killed more unidentified small birds than any individual species, at 10,716. Solar thermal killed 8,118 canyon bats and members of ≥8 other bat species (Table S4). It killed likely migrants, including 7,043 yellow ‐rumped warblers ( Setophaga coronata), 5,582 yellow warblers, 4,470 Costa's hummingbirds ( Calypte costae ), and thousands of birds of other species. It killed many grassland and desert scrubland birds, including 3,678 white ‐crowned sparrows ( Zonotrichia leucophrys), 3,506 American pipits ( Anthus rubescens), 3,116 western meadowlarks, and 1,498 loggerhead shrikes ( Lanius ludovicianus).

I estimated that in 2020 fences surrounding solar projects killed 1,783 canyon bats, 1,385 western mea- dowlarks, 1,174 greater roadrunners ( Geococcyx californianus ), and 476 northern flickers ( Colaptes auratus ; Table S4). Fences also killed an estimated 226 burrowing owls, 172 yellow ‐headed blackbirds ( Chrysomus icter- ocephalus), and 108 northern harriers ( Circus hudsonius ).

I estimated that in 2020 gen ‐ties of solar projects killed 8,425 Wilson's warblers ( Cardellina pusilla ), 7,144 Brewer's sparrows ( Spizella breweri ), 3,364 common yellowthroats, 2,03 1 yellow warblers, 899 loggerhead shrikes ( Lanius ludovicianus ), 241 American kestrels, and 195 red ‐tailed hawks ( Buteo jamaicensis ; Table S4).

Fatalities at operational projects represented ≥8b a ts p e c i e sa n d≥192 bird species. Of the 10 species of bats represented by fatalities at so lar projects, 3 were California spe cies of special concern. Western Bat Working Group ranked 4 species as high and 2 as mode rate conservation priority. Of species of birds represented by fatalities at solar projects, 2 were lis ted as threatened or endangered under the federal or California Endangered Species Acts. California Fish and Game Code 3503.5 (i.e., birds of prey) protected ≥20 species, and additional statutes protected 8 of these species. Two bird species were California fully pro- tected, and another 36 species were listed as a United States Fish and Wildlife Service bird species of conservation concern or California species of conserva tion concern, or listed on California's Watch List. Of species listed as California species of special concern ,8w e r ep r i o r i t yl e v e l3 ,9w e r ep r i o r i t yl e v e l2 ,a n d2 were priority level 1 (2 species were assigned no priority level). Including raptors, 45 special ‐status species of vertebrate wildlife were documented as fatalities of utility ‐scaled solar projects, or 23% of all species re- presented by fatalities.
16 of 29 | SMALLWOOD
Additional sources of error and bias
The PV solar projects were searched by foot ( n = 6) and car ( n = 5), whereas SCAs at Genesis were searched by car, and power tower projects were searched by foot. Estimated fatalities differed between foot and car searches at PV projects ( t9 = 2.44, P = 0.037), with respective means of 15.14 F/MW/year and 5.08 F/MW/year.

To indicate whether and to what degree wildlife species were represented in fatality monitoring results at solar projects, I fit nonlinear models to the cumulative number of species represented by fatalities with increasing number of months into the study at 5 projects that were monitored more extensively in space and time (Table 4).

The monitoring data from Ivanpah served to exemplify not only the asymptotic nature of species representation with monitoring duration, but also the effect of a change in search coverage (Figure 5A). Monitoring effort sub- stantially lessened after 24 months when the monitor terminated fatality searches along the gen ‐tie and fence and reduced search coverage among heliostat arrays from 24.1% to 8% (0% in the fifth year). Subsequent to the reduced search effort, the increase in number of additional species represented by fatalities was slower than predicted by the pattern in the data through 24 months (Figure 5A). Where fatality monitoring lasted long enough, the pattern in the data revealed how much longer it would have taken at sustained search efforts to achieve asymptotes in species representation by fatalities (Figure 5B). Fatality monitoring at 5 projects was too brief to reveal anything other than a linear increase of species with increasing monitoring duration (Figure 5C). The re- gression slopes of Blythe and McCoy were nearly identical (Figure 5C). These 2 projects were adjacent to each other, and monitored by the same consulting firm. But even after 2 years of monitoring and documentation of >50 species as fatalities, the pattern of increasing cumulative species represented by fatalities hinted of no imminent arrival to an asymptote. I could not fit a model to the cumulative number of species represented by fatalities with increasing number of months into the study from 4 PV projects. The MW ‐years sampled among these 4 projects averaged only 11.4% of MW ‐years at the other 7 PV projects, a significant difference ( F 1,9 = 22.76, P = 0.001).

Species represented by fatalities were often vulnerable to biased fatality estimates because of mis- representation. Monitors did not identify nearly 22% of fatalities to species, and identified them instead to coarser levels such as genus, subfamily, small, medium, or large bird, or as unknown bird or bat (Table 5). The coarser the reported level of identification, the more I applied detection probabilities derived from lumped representations of body mass. For example, the degraded carcass of a Pacific ‐slope flycatcher ( Empidonax difficilis) found by a weekly TABLE 4 Best‐fit models of cumulative vertebrate species detected as fatalities at California, USA, utility ‐scale solar projects relative to the number of months since the fatality monitoring began, where model 1 was of the form Y abX=+× , and model 2 was of the form Y = abc
1 +× log X10
Project Model
Coefficients
r2 RMSEaab c
California Valley Solar Ranch 2 0.0157 0.4970 0.0490 0.99 51.34 Topaz 2 0.0032 0.2229 0.2548 0.96 266.20
Campo Verde 2 0.0096 0.2435 0.1401 0.97 81.30
Desert Sunlight 1 –2.5714 3.3571 0.96 1.44
Stateline 1 0.8 0.55 0.92 0.67
McCoy 1 3.2029 2.0804 0.96 1.44
Blythe 1 2.0000 2.1700 0.98 1.02
Genesis 2 0.0097 0.2305 0.0017 0.98 311.57
Ivanpah 2 0.0061 0.1762 0.0406 0.99 300.49
aRMSE = root mean square error.

SOLAR IMPACTS | 17 of 29
(A)
(B)
(C)
FIGURE 5 Cumulative number of vertebrate species detected as fatalities, California, USA, 2012 ‒2018: A) during the first 2 years of monitoring (open circles) and last 3 years of reduced monitoring effort (filled circles) at the Ivanpah Solar Electric Generating System, San Bernardino County, California, along with best ‐fit model to the first 2 years of data and projected to the last 3 years of monitoring; B) among solar projects where fatality monitoring lasted long enough to predict an asymptote of species represented by fatalities (vertical bars indicate duration of monitoring used to project the nonlinear pattern in the data); and C) among solar projects where fatality monitoring had not lasted long enough to predict an asymptote, and therefore insufficiently represented species affected.

Survey effort largely determined the number of species represented by found fatalities 18 of 29 | SMALLWOOD
search on open ground might have been reported as a small bird. I assigned D = 0.197 to small birds, which was the value predicted for house sparrows –the smallest species placed in small ‐bird detection trials among solar projects.

Had the Pacific ‐slope flycatcher been accurately identified to species, it would have been adjusted by D = 0.10, resulting in twice the estimated fatality rate.

The same bias applies to bats. As one of the most often ‐found bats among solar projects, canyon bats likely contributed most often to bats found and identified as unknown bat. At wind projects, an unknown bat would typically be adjusted using an overall detection probability quantified from the placements of all bats in carcass detection trials. I therefore used the mean from such trials to represent unknown bats with D = 0.1008. Overall detection for canyon bat was D = 0.0485. Again, had searchers been able to identify canyon bats to species instead of unknown bats, and assuming most of the unknown bats were canyon bats, each of these bats' contribution to bat fatality estimates would have doubled.

At Centinela, 47% of fatalities were not identified to species (Table 5). Other projects where monitors could not identify many fatalities to species included Solar One and Ivanpah at 28.8% and 27.7%, respectively. Projects where all fatalities were identified to species were surveyed briefly by car, including at Imperial where fatalities of only 4 large‐sized bird species were found, and at Stateline where fatalities of only 9 species were detected.

DISCUSSION
I largely achieved my study objectives. Original reports of fatality monitoring did not always provide fatality estimates for project features and whole projects, but of those that did, I compared my independent estimates to theirs. From my estimates of project ‐level fatalities, I estimated statewide fatalities based on the installed capacity TABLE 5 Species of wildlife identified as fatalities among California's utility ‐scale solar energy projects, USA, 1982‒2018, and the percentages of estimated fatalities ( F) that could and could not be identified to species Solar project
Number of species
represented by fatalities
Percent of F/MW/year of F
Identified Not identified
California Valley Solar Ranch 43 96.8 3.2
Topaz Solar Farm 17 95.3 4.7
Campo Verde Solar Project 45 81.4 18.6
Centinela Solar Energy Project 24 53.0 47.0
Imperial Solar Energy Center South 4 100.0 0.0
Calipatria Solar Project 13 77.8 22.2
Midway Solar Farm II 4 76.1 23.9
Desert Sunlight Solar Farm 53 83.3 16.7
Stateline Solar Farm 9 100.0 0.0
McCoy Solar Energy Project 49 79.9 20.1
Blythe Solar Power Project 54 82.6 17.4
Genesis Solar Energy Project 97 81.3 18.7
Solar One 25 71.2 28.8
Ivanpah Solar Electric Generating System 147 72.3 27.7
All projects 200 78.3 21.7
SOLAR IMPACTS | 19 of 29
of utility ‐scale solar projects in 2020, and I did so per each project feature. An exception was for medium ‐voltage overhead lines, which deposited bird carcasses too close to PV solar arrays to discern whether the fatalities were caused by collisions with the overhead lines or the PV solar arrays. I compared fatality rates among species of birds and bats, and among those with special status. I also identified sources of error and bias to alert investigators to research directions needed to improve accuracy of fatality estimation.

To achieve my study objectives, I assumed that the fa tality data and carcass detection trial data I ob- tained from natural resource agencies were accura tely reported. I assumed that I accurately interpreted reported study methods and descriptions of solar projec ts. Data inaccuracies or misinterpretation of methods would have diminished the accuracy of my independent f atality estimates. For biases and large sources of error that I understood from previous research on fatali ty estimation at wind energy projects, I endeavored to mitigate them by applying adjustment factors. For meth ods I believed were inappropriate, such as searches for fatalities from cars, I examined the data for evidenc e of bias and for opportunities to mitigate the effects of bias. I also applied a bridge from my study of the ef fects of variation in body mass on overall carcass detection probability to account for the disparity I fo und between carcasses placed in detection trials and those found as fatalities at solar projects. My estimates of fatalities at solar projects were as accurate as I could achieve but undoubtedly inaccurate to some unkn own degree. Potential for inaccuracy was greatest for the smallest birds and bats. On the whole, I suspect I more often underestimated rather than overestimated fatality rates. My fatality estimates nevertheless indicate the magnitude of impacts caused by utility ‐scale solar projects in California, and they can serve as a coll ective starting point for improving methods to increase accuracy of fata lity estimates.

I estimated that by 2020 annual fatalities exceeded a quarter million birds and 11,000 bats at California solar projects. My statewide fatality estimates were based on the available, nonrandom pool of monitored projects, which means they were vulnerable to geographic biases of unknown direction. At the same time, the fatality monitoring performed to date has been prone to substantial biases that largely resulted in underestimation of fatalities. Earlier estimates of regional fatalities were much lower than mine. At PV and solar thermal projects in California, Walston et al. ( 2016) estimated a weighted mean 9.9 bird fatalities/MW/year, whereas I estimated a weighted mean 18.9 bird fatalities/MW/year. At solar PV projects, Kosciuch et al. ( 2020) estimated 2.49 bird fatalities/MW/year, whereas I estimated a mean 11.6 bird fatalities/MW/year without adding fatalities caused by perimeter fences and gen ‐ties. The earlier regional estimates used reported estimates without sufficiently ac- counting for the effect of body mass on carcass detection probability. They also relied on project ‐level estimates exclusive of the many species represented by F < 5, and because many of these excluded species were small ‐ bodied, under ‐estimation of
F had been exacerbated. For these reasons, my fatality estimates were larger, but other biases indicate fatality estimates could be even larger. As examples, the monitors' use of cars for fatality searches and insufficient survey effort both appear to have negatively biased F . Omitting the 7 fatality monitoring studies that were either based on car surveys or cursory effort, the weighted mean increases 1.25 ‐fold from 11.605 to 14.51 bird fatalities/MW/year among PV projects.

Background mortality (i.e., mortality caused by natural processes) was often suggested as the source of fatalities for which cause of death could not be assigned (Kosciuch et al. 2020), but no reports raised the possibility of crippling bias, the rate of mortally injured animals leaving the search area undetected (Smallwood 2007). Some unknown number of animals mortally injured by exposure to the zone of solar flux, electrocution, or collision likely died, undiscovered, far from study areas. The very background mortality studies discussed by Kosciuch et al. ( 2020) and reported in (2013) and Althouse and Meade ( 2012, 2014) likely counted fatalities representative of both crippling bias and background mortality because the background mortality monitoring plots were adjacent to the solar energy projects. These studies could have lessened confounding had they begun monitoring ≥1 year in advance of project construction, and had they located study plots farther from the project to minimize counting of remains of birds that had been displaced by construction and rendered more vulnerable to predation and competition.
20 of 29 | SMALLWOOD
Estimates of solar energy's impacts could be more complete by monitoring for fatalities along transmission lines constructed to accommodate the projects. Transmission lines kill many birds, mostly from collisions but also electrocutions. Loss et al. (2014) estimated 29.6 bird collision fatalities/km/year (95% CI = 9.3 –66.4) in the United States based on a mix of studies with and without adjustment by ρ. Rioux et al. ( 2013) estimated 110.4 bird collision fatalities/km/year (95% CI = 43.5 –177.2) across Canada after adjusting by ρ. Avian fatality rates along transmission lines have been rarely studied in California. The only study of which I was aware was at Mare Island in Solano County (Hartman et al. 1992). Weighted mean fatality rates along gen ‐ties servicing solar projects (Table S4) could serve to estimate annual fatality rates along transmission lines that deliver power from solar projects, so long as one knows the length of transmission lines that do so.

Impacts of habitat loss
Estimates of solar energy's impacts could be more complete by also measuring numerical losses of wildlife popu- lations caused by habitat loss. Project sites should be studied prior to construction to estimate densities in breeding and non‐breeding seasons, and to measure migratory activity levels, which based on isotopes in feathers collected from dead birds found at solar projects, appears high (Conkling et al. 2020). Wildlife use of the site could be compared before and after project construction and concurrent with studies of wildlife at reference sites.

By 2020 the installed capacity of solar energy projects covered 39,409 ha in California, including 25,453 ha (64.6%) in the Mojave and Sonoran deserts. Construction grading for these projects removed all native vegetation, which continued to be suppressed through project operations. Utility ‐scale solar projects destroyed habitats of many wildlife species by eliminating breeding sites and forage. No studies were performed to estimate breeding and nonbreeding densities in advance of construction grading at any of the projects I reviewed. No efforts were made to estimate potential reductions in reproductive and numerical capacities of wildlife species.

Estimates of total bird nesting density are available, along with species ‐specific density estimates, for indicator ‐ level analysis of impacts. Bird nesting densities of 13.27 and 14.19 nest sites/ha were estimated at study sites composed of grassland, wetland, and woodland (Young 1948, Yahner 1982). Assuming 6.94 nest sites/ha, or half the average of Young's ( 1948) and Yahner's ( 1982) densities, would more realistically represent California's annual grasslands, this density multiplied against California's 12,414 ha of utility ‐scale solar in non ‐desert environments would predict a loss of 86,153 bird nest sites.

Franzeb (1978) provided a basis for applying bird breeding density to areas lost to solar projects in California's deserts. Franzeb ( 1978) estimated 0.366 breeding birds/ha at a Sonoran Desert site. Projected to 25,453 ha of California's desert environments, this density would estimate 9,316 breeding birds, or 4,658 nest sites, were lost to construction grading for solar projects. Added to the estimate above for non ‐desert environments, I estimate that California solar projects displaced 90,811 bird nests by 2020. Assuming 25 years of operational impacts, and assuming an average fledging of 2.9 birds/nest/year (Young 1948) and a generation time of 5 years, the lost capacity of breeders and annual chick production would be 7,491,908 birds: {[nests/year × chicks/nest × number of years] + [2 adults/nest × nests/year × (number of years ÷ years/generation)]}. Averaged over 25 years of project operations, this indicator ‐level approach would estimate an annual loss of 299,676 birds due to habitat destruction to accommodate California's 2020 installed capacity of utility ‐scale solar.

Burrowing owls, in particular, face significant impacts from habitat loss caused by development of utility ‐scale solar. Burrowing owls happen to select terrain conditions for breeding and foraging that are also sought by developers of solar projects: flat to gentle southwest ‐ or south ‐facing slopes. In the 1990s, an estimated 71% of California's entire burrowing owl population resided within the Imperial Valley, after the species had declined throughout the rest of its range in California (DeSante et al. 2007). The average density of breeding attempts was

8.13/km
2 (0.081/ha). With 1,488.5 MW of solar PV installed in Imperial County as of 2019, and with an average

2.67 ha/MW, utility ‐scale solar in Imperial County destroyed 3,974.3 ha of burrowing owl habitat. Assuming the SOLAR IMPACTS | 21 of 29
estimate of burrowing owl breeding density from DeSante et al. ( 2007) still applies, I estimate this level of habitat loss reduced the capacity for breeding attempts by 322 and the number of breeding adults by ≥644. Assuming an average generation time of 8 years and an average 3 chicks produced/breeding attempt, the equation in the preceding paragraph would estimate an annual loss of 1,073 burrowing owls due to habitat loss.

Another form of habitat loss, or habitat degradation, was project ‐generated wildfires. In June 2019, a bird landed on a power pole at CVSR, sparking a 486 ‐ha fire that removed 84% of the project's generating capacity. The fire reportedly did not damage solar panels, but it likely destroyed many bird nests on the grasslands that were burned next to the project.

Combined impacts to birds from collision mortality and habitat loss Assuming utility‐scale solar projects eliminate breeding habitat within the project footprint, then post ‐construction collision mortality is largely to migrants or residents of adjacent habitat. If true, then I estimated the installed capacity of utility ‐scale solar in 2020 removes from California >500,000 birds/year because of combined effects of collision mortality and loss of birds to habitat destruction. Over 25 years of operations of this installed capacity, I estimated California's loss of birds will be >12.5 million. I estimated just over half of this mortality was attributable to habitat destruction (53%), and the rest to operative solar energy projects, but source ‐specific magnitudes of impacts likely would vary interspecifically.

For burrowing owls in particular, the estimated mean 0.182 collision fatalities/MW/year at solar PV projects applied to 1,488.5 MW of solar PV installed in Imperial County as of 2019 would estimate 271 burrowing owl fatalities/year in Imperial County, where most of California's burrowing owls live. Combined with the estimated impact of habitat loss, I estimate utility ‐scale solar in Imperial County removed 1,344 burrowing owls annually as of 2020. With an estimated 5,600 burrowing owls in the Imperial Valley in the 1990s, my estimated loss could prove devastating for burrowing owls in Imperial Valley.

Avian mortality at utility ‐scale solar projects might contribute cumulatively to an ongoing substantial decline of birds in North America (Rosenberg et al. 2019). Bat mortality at solar projects contribute cumulatively to in- creasingly large nationwide impacts to bats caused by wind turbines (Smallwood 2020). That many of the birds and bats killed at solar projects were special ‐status species should heighten concern over the ecological and economic impacts of solar energy ‐caused mortality. A suite of errors and biases suppress understanding of the ecological impacts of solar projects and how to mitigate them.

Additional sources of error and bias in collision fatality estimates Fatality monitoring efforts not only varied greatly among solar projects, but they introduced biases that resulted in inaccurate fatality estimates. Monitoring that lasted <1 year or covered small portions of a project likely generated incomplete lists of affected species and hence inaccurate fatality estimates. Searches of PV projects by car yielded a third of the fatality estimates of searches of PV projects by foot. No studies performed car surveys simultaneously with pedestrian surveys. Perhaps the most substantial bias was the use of carcasses in detection trials that were larger than many of the birds and bats found as fatalities. I endeavored to offset this latter bias, whereas I identified most others as likely. I also identified additional sources of error and bias that cannot be quantified without further effort on the parts of monitors going forward.

Monitors recorded fatalities of nonvolant wildlife at only 1 project. Either nonvolant animals died only at the 1 project or monitors ignored them at other projects. In another example of potential bias, monitors removed all fatality finds to prevent double counting of fatalities. If this practice altered forage availability and hence scavenger removal rates of carcasses placed in detection trials, then bias was possible. Omissions of fatalities found 22 of 29 | SMALLWOOD
incidentally to routine searches introduced potential bias when those discoveries occurred on search plots. Also, where monitors searched daily for 5 to 7 consecutive days at beginning of monthly intervals, difficult decisions were necessary over how to treat fatalities found on the first day of each daily search sequence. Another potential bias was the practice of placing feather piles in bird carcass detection trials because doing so increased carcass persistence by offering scavengers inedible body parts that normally would not occur without flesh and bone upon a bird's death.

Monitors could not identify a fifth of all fatalities to species, which means they likely misidentified fatalities such as Allen's hummingbird ( Selasphorus sasin ) and rufous hummingbird ( Selasphorus rufus ). Such errors affect 2 species at once, the species inaccurately assigned the fatality and the species inaccurately denied it. Factors that diminish one's ability to identify fatalities to species include inexperience, insufficient motivation to examine re- mains carefully enough to identify them, and too much time between fatality searches. As search interval lengthens, more carcasses degrade to the point of becoming less recognizable to species. A related factor is lower detection probabilities of certain search methods. Human searchers miss more carcasses than do scent ‐detection dogs (Smallwood et al. 2020), and judging from the 3 ‐fold difference in fatality rates between car and pedestrian surveys, searchers in cars likely miss more carcasses than do searchers on foot. Missed carcasses can later be found, but the delay in detection allows carcasses to be scavenged and decay, thereby increasing the likelihood they will not be identified to species. Species identification errors could not be fully quantified without viewing photographic evidence or preserved remains of all fatality finds. Such evidence was not forthcoming in responses to PRA and FOIA requests to state and federal agencies.

Mitigation measures and recommendations
Other than measures implemented to minimize hazards along powerlines, little testing was performed of candidate mitigation measures to minimize or reduce collision fatalities with solar collectors. Nearly all projects implemented mitigation measures to minimize bird collision mortality with gen ‐ties and medium ‐voltage overhead lines, and 1 implemented a measure to minimize avian collisions with solar PV panels (Dudek 2018b). At CVSR, for example, the gen‐tie and medium ‐voltage overhead lines were fitted with Swan ‐flight Diverters SFD1520 during fatality mon- itoring (H. T. Harvey & Associates 2013). An ultrasonic acoustic deterrent was also implemented at Ivanpah's Unit 1 on 10 September 2014 and on 23 April 2015 at both Units 2 and 3 to reduce bat exposure to the zone of solar flux.

The deterrent, however, was installed without experimental design elements that would have supported a con- clusive hypothesis test of whether the deterrent reduced fatalities. Although bat fatalities declined throughout the study (Figure 4), the largest declines preceded installations of deterrents, which left the possibility that bat numbers were initially depleted by project ‐caused mortality. Although guidance had been prepared on how to perform experiments to test mitigation efficacy (Sinclair and DeGeorge 2016), none of the mitigation measures were implemented along with sufficient experimental design to convincingly test the measure's efficacy.

Considering the objectives of fatality monitoring and mitigation measures implemented at utility ‐scale solar projects, and considering what needs to be known of solar energy impacts to wildlife, I recommend the following changes and additions to wildlife studies: 1) perform baseline studies beginning ≥1 year prior to project con- struction to quantify relative abundance, densities in both breeding and nonbreeding seasons, and behavior pat- terns of resident and migratory wildlife species; 2) repeat fatality monitoring at projects where it was performed inadequately; 3) either monitor all new utility ‐scale solar projects for fatalities, or sample projects from the available pool of projects to obtain region ‐wide fatality estimates; 4) improve scientific access, transparency, and data sharing; 5) introduce peer review as a required standard of report preparation; 6) increase methodological stan- dardization where it is most needed; 7) minimize fatality search intervals or use scent ‐detection dogs (Paula et al. 2011, Mathews et al. 2013, Smallwood et al. 2020); 8) treat incidental fatality finds just like those found in scheduled searches, so long as they were located on sampling plots; 9) whenever trying a new search method, SOLAR IMPACTS | 23 of 29
compare it to the conventional method through concurrent implementation of methods on, minimally, a suitable subsample of the search area; 10) integrate carcass detection trials into routine fatality monitoring, using appro- priate species to quantify overall detection probability (Smallwood et al. 2018); 11) quantify and report observer error rates associated with monitoring, such as carcass detection rates and species misidentifications; 12) report species‐specific use rates and fatality rates in addition to use rates and fatality rates of all birds and all bats; and 13) implement tenets of experimental design to test hypothesized causal factors and the efficacy of mitigation mea- sures (Sinclair and DeGeorge 2016, Smallwood and Bell 2020 a).

Based on the impacts I quantified in this review, my first recommendation on performing baseline studies is of high priority. The reports I reviewed were overly focused on collision mortality and paid no attention to impacts of habitat loss. Whereas collision mortality is substantial, more animals are lost to habitat destruction caused by the development of utility ‐scale solar projects. Resource agencies and permitting authorities need to view the impacts and how to mitigate them more comprehensively. I suggest directing 2 to 3 times more effort into measuring and mitigating habitat impacts than collision impacts. I suggest implementing macro ‐siting strategies, and I note that I saw amid the data and reports I received via PRA and FOIA requests no evidence of siting to minimize impacts to wildlife. One strategy would be to site projects to avoid migration clusters or locations where rare or declining populations of particular species occur (Ruegg et al. 2020). Another would be to site projects to minimize collision mortality of all birds and bats, and yet another would be to site to minimize loss of breeding sites.

I suggest the formulation of a research fund to assess impacts of habitat loss, explore methodological im- provements to more accurately predict and estimate impacts, quantify the effects of causal factors, and test the efficacy of mitigation measures. An example of research that would vastly improve understanding of causal factors of collision mortality would be surveys performed by behavioral ecologists. As in wind energy projects, the most effective means to learn of the circumstances of fatalities is to study animal behavior in the context of the project specific to birds (Hoover and Morrison 2005, Smallwood et al. 2009a,b; Hull and Muir 2013) and bats (Horn et al. 2008, Cryan et al. 2014). In my experience, careful observation of living animals can determine collision factors more effectively than can collections of dead animals (Smallwood et al. 2017, Smallwood 2017b, Smallwood and Bell 2020a,b). Lastly, I suggest follow ‐up measures to reduce and compensate for impacts revealed by monitoring.

MANAGEMENT IMPLICATIONS
In light of the magnitudes of solar energy impacts to wildlife, it would be prudent to slow the development of utility ‐ scale solar projects long enough to formulate more accurate impacts assessments and more effective project decision‐making, mitigation, and oversight. More than sufficient capacities exist for distributed generation and energy conservation to compensate for the time needed to develop policies and methods for safer development of utility‐scale solar energy. Time is needed to develop decision ‐making frameworks that compare costs and benefits of solar technology, such as solar thermal versus PV, and alternative project locations. Time is needed to improve fatality monitoring and measurement of impacts, including routine implementation of baseline studies. Time is also needed to develop a more transparent oversight process, along with consequences for exceedance of mortality thresholds or for failing to meet standards of baseline studies and fatality monitoring.

ACKNOWLEDGMENTS
I thank N. L. Smallwood for organizing information from numerous reports and data provided in response to PRA and FOIA requests, and L. Neher for help with a geographic information system. I thank all those biologists who painstakingly searched for fatalities at solar projects. The paper benefited from the help of 2 anonymous reviewers.

CONFLICT OF INTEREST
The authors declare that there are no conflicts of interest. 24 of 29 | SMALLWOOD

DATA AVAILABILITY STATEMENT
Data available in article supplementary material and in cited reports. Cited reports of fatality monitoring are also available from the author on request.

REFERENCES
Althouse and Meade. 2012. Topaz Solar Farm avian and bat protection plan and bird monitoring and avoidance plan fourth quarterly/first annual report for 2012. Report to Topaz Solar Farms, Santa Margarita, California, USA.

Althouse and Meade. 2014. Topaz Solar Farms 2013 fourth quarter/second annual report for avian and bat protection plan and bird monitoring and avoidance plan. Report to Topaz Solar Farms, Santa Margarita, California, USA.

Beston, J. A., J. E. Diffendorfer, and S. R. Loss. 2015. Insufficient sampling to identify species affected by turbine collisions.

Journal of Wildlife Management 79:513 –517.

Brown, K., K. S. Smallwood, J. Szewczak, and B. Karas. 2016. Final 2012 –2015 report avian and bat monitoring project Vasco Winds. Prepared for NextEra Energy Resources, Livermore, California, USA.

Calvert, A. M., C. A. Bishop, R. D. Elliot, E. A. Krebs, T. M. Kydd, C. S. Machtans, and G. J. Robertson. 2013. A synthesis of human‐related avian mortality in Canada. Avian Conservation and Ecology 8:11.

California Energy Commission, California Department of Fish and Wildlife, U.S. Bureau of Land Management, U.S.

Fish and Wildlife Service. 2014. Draft Desert Renewable Energy Conservation Plan (DRECP) and Environmental Impact Report/Environmental Impact Statement, SCH No. 2011071092, BLM/CA/PL ‐2014/025+1793, FWS–R8–ES–2014–N165. California Energy Commission, Sacramento, USA.

California Public Utilities Commission. 2008. 33% renewables portfolio standard for California by 2020 implementation analysis workplan. California Public Utilities Commission Energy Division, San Francisco, USA.

Chambers Group. 2016. Centinela Solar Project bird and bat conservation strategy post ‐construction monitoring: annual report for the period of October 2014 –December 2015. Report to San Diego Gas & Electric Company, San Diego, California, USA.

Conkling, T. J., H. B. Vander Zanden, S. A. Poessel, S. R. Loss, T. D. Allison, J. E. Diffendorfer, A. E. Duerr, D. M. Nelson, J. Yee, and T. E. Katzner. 2020. Learning from real ‐world experience to understand renewable energy impacts to wildlife. California Energy Commission. CEC ‐500‐2020‐012, Sacramento, USA.

Cooper, D. S. 2016. Industrial ‐scale solar projects and birds in the California desert: assessing impacts and developing mitigation. Report prepared for Sonoran Joint Venture, Tucson, Arizona, USA.

Corcoran, A. J., and T. J. Weller. 2018. Inconspicuous echolocation in hoary bats ( Lasiurus cinereus ). Proceedings of the Royal Society of London B 285:20180441.

Cryan, P. M., P. M. Gorresen, C. D. Hein, M. R. Schirmacher, R. H. Diehl, M. H. Huso, D. T. S. Hayman, P. D. Fricker, F. J. Bonaccorso, D. H. Johnson, et al. 2014. Behavior of bats at wind turbines. Proceedings National Academy of Science 111:15126 –15131.

DeSante, D. F., E. D.Ruhlen, and R. Scalf. 2007. The distribution and relative abundance of burrowing owls in California during 1991–1993: evidence for a declining population and thoughts on its conservation. Pages 1 –41 in J. H. Barclay, K. W. Hunting, J. L. Lincer, J. Linthicum, and T. A. Roberts, editors. Proceedings of the California Burrowing Owl Symposium, November 2003. Bird Populations Monographs No. 1. The Institute for Bird Populations and Albion Environmental, Point Reyes Station, California, USA.

Doering, B., and V. Santistevan. 2013. Blythe Mesa Solar Project bird and bat conservation strategy. Report to Renewable Resources Group and Solar Star Blythe Mesa 1, Los Angeles, California, USA.

Dudek. 2018 a. Bird and bat conservation strategy (BBCS) annual report: June 2017 through May 2018, Stateline Solar Farm, San Bernardino County, California. Report to Southern Power Company, Atlanta, Georgia, USA.

Dudek. 2018 b. Tranquility Solar – avian deterrent study. Report to Southern Power Company, Atlanta, Georgia, USA.

Dunn, E. H. 1993. Bird mortality from striking residential windows in winter. Journal of Field Ornithology 64:302 –309.

Erickson, W. P., M. M. Wolfe, K. J. Bay, D. H. Johnson, and J. L. Gehring. 2014. A comprehensive analysis of small ‐passerine fatalities from collision with turbines at wind energy facilities. PLoS One 9(9):e107491.

Franzeb, K. E. 1978. Breeding bird densities, species composition, and bird species diversity of the Algodones Dunes.

Western Birds 9:9 –20.

G o r r e s e n ,P .M . ,P .M .C r y a n ,K .M o n t o y a‐Aiona, and F. J. Bonaccorso. 2017. Do you hear what I see? Vocalization relative to visual detection rates of Hawaiian hoary bats ( Lasiurus cinereus semotus ). Ecology and Evolution 7:
6669
–6679.

Greif, S. and B. M. Siemers. 2010. Innate recognition of water bodies in echolocating bats. Nature Communications 1:107.

Hager, S. B, and M. E. Craig. 2014. Bird ‐window collisions in the summer breeding season. PeerJ 2:e460.

Harrison, C., H. Lloyd, and C. Field. 2016. Evidence review of the impact of solar farms on birds, bats and general ecology.

Natural England, York, England.

SOLAR IMPACTS | 25 of 29
Hartman, P. A., S. Byrne, and M. F. Dedon. 1992. Bird mortality in relation to the Mare Island 115 ‐kV transmission line: final report, 1988 –1991. Pages 14.1 –14.17 in Proceedings of the International Workshop on Avian Interactions with Utility Structures. Electrical Power Research Institute and Avian Power Line Interactions Committee, Palo Alto, California, USA.

Hayes, M. A. 2013. Bats killed in large numbers at United States wind energy facilities. BioScience 63:975 –979.

Heritage Environmental Consultants. 2014. Post ‐construction avian mortality monitoring report for the Campo Verde Solar Project: third quarter 2014 (July, August, September) and year 1 annual report (October 2013 –September 2014).

Heritage Environmental Consultants, Denver, Colorado, USA.

Heritage Environmental Consultants. 2015 a. Post ‐construction avian mortality monitoring report for the Campo Verde Solar Project: year 2 annual report (October 2014 –September 2015). Heritage Environmental Consultants, Denver, Colorado, USA.

Heritage Environmental Consultants. 2015 b. Post ‐construction avian mortality monitoring report Centinela Solar Energy Project: fourth quarterly report (May –July 2015) and annual summary report. Heritage Environmental Consultants, Denver, Colorado, USA.

Heritage Environmental Consultants. 2016. Post ‐construction avian mortality monitoring report for the Campo Verde Solar Project: year 3 annual report (October 2015 –September 2016). Heritage Environmental Consultants, Denver, Colorado, USA.

Heritage Environmental Consultants. 2017 a. Post ‐construction avian mortality monitoring report Calipatria Solar Project:
year 1 semi ‐annual report (June 2016 –October 2016). Heritage Environmental Consultants, Denver, Colorado, USA.

Heritage Environmental Consultants. 2017 b. Post ‐construction avian mortality monitoring report Calipatria Solar Project:
year 1 semi ‐annual report #2 (November 2016 –May 2017). Heritage Environmental Consultants, Denver, Colorado, USA.

Heritage Environmental Consultants. 2017 c. Post ‐construction avian mortality monitoring report Calipatria Solar Project:
year 2 semi ‐annual report #3 (June 2017 –Oct 2017). Heritage Environmental Consultants, Denver, Colorado, USA.

Hoover, S. L., and M. L. Morrison. 2005. Behavior of red ‐tailed hawks in a wind turbine development. Journal of Wildlife Management 69:150 –159.

Horn, J. W., E. B. Arnett, and T. H. Kunz. 2008. Behavioral responses of bats to operating wind turbines. Journal of Wildlife Management 72:123 –132.

Horváth, G., M. Blahó, Á. Egri, G. Kriska, I. Seres, and B. Robertson. 2010. Reducing the maladaptive attractiveness of solar panels to polarotactic insects. Conservation Biology 24:1644 –1653.

Horváth, G., G. Kriska, P. Malik, and B. Robertson. 2009. Polarized light pollution: a new kind of ecological photopollution.

Frontiers in Ecology and the Environment 7:317 –325.

H.T. Harvey & Associates. 2013. California Valley Solar Ranch Project avian fatality during construction first annual report September 2011 –September 2012. Report for HPR III, LLC, Carlsbad, California, USA.

H.T. Harvey & Associates. 2015 a
. California Valley Solar Ranch Project avian and bat protection plan final postconstruction fatality report. Report to HPR II, LLC, Santa Margarita, California, USA.

H.T. Harvey & Associates. 2015 b. Ivanpah Solar Electric Generating System avian & bat monitoring plan 2013 –2014 annual report (revised) (29 October 2013 –20 October 2014). Report to Solar Partners I, II, and VIII, Nipton, California, USA.

Hull, C. L., and S. Muir. 2013. Behavior and turbine avoidance rates of eagles at two wind farms in Tasmania, Australia.

Wildlife Society Bulletin 37:49 –58.

Huso, M. M. P., and D. Dalthorp. 2014. A comment on “bats killed in large numbers at United States wind energy facilities. ” BioScience 64:546 –547.

Huso, M., T. Dietsch, and C. Nicolai. 2016. Mortality monitoring design for utility ‐scale solar power facilities. U.S. Geological Survey Open ‐File Report 2016 –1087, Reston, Virginia, USA.

Johnson, D. H., S. R. Loss, K. S. Smallwood, and W. P. Erickson. 2016. Avian fatalities at wind energy facilities in North America: a comparison of recent approaches. Human –Wildlife Interactions 10:7 –18.

Kagan, R. A., T. C. Viner, P. W. Trail, and E. O. Espinoza. 2014. Avian mortality at solar energy facilities in Southern California: a preliminary analysis. National Fish and Wildlife Forensics Laboratory, Ashland, Oregon, USA.

Kahle, L. Q., M. E. Flannery, and J. P. Dumbacher. 2016. Bird ‐window collisions at a west ‐coast urban park museum:
analyses of bird biology and window attributes from Golden Gate Park, San Francisco. PLoS ONE 11(1):e144600.

Kitano, M., M. Ino, K. S. Smallwood, and S. Shiraki. 2020. Seasonal difference in carcass persistence rates at wind farms with snow, Hokkaido, Japan. Ornithological Science 19:63 –71.

Kitano, M., and S. Shiraki. 2013. Estimation of bird fatalities at wind farms with complex topography and vegetation in Hokkaido, Japan. Wildlife Society Bulletin 37:41 –48.

Korner‐Nievergelt, F., P. Korner ‐Nievergelt, O. Behr, I. Niermann, R. Brinkmann, and B. Hellriegel. 2011. A new method to determine bird and bat fatality at wind energy turbines from carcass searches. Wildlife Biology 17:350 –363. 26 of 29 | SMALLWOOD
Kosciuch, K., D. Riser ‐Espinoza, M. Gerringer, and W. Erickson. 2020. A summary of bird mortality at photovoltaic utility scale solar facilities in the Southwestern U.S. PLoS ONE 15(4):e0232034.

Loss, S. R., T. Will, and P. P. Marra. 2013 a. Estimates of bird collision mortality at wind facilities in the contiguous United States. Biological Conservation 168:201 –209.

Loss, S. R., T. Will, and P. P. Marra. 2013 b. The impact of free ‐ranging domestic cats on wildlife of the United States. Nature Communications 4:1396.

Loss, S. R., T. Will, and P. P. Marra. 2014 a. Estimation of bird ‐vehicle collision mortality on U.S. roads. Journal of Wildlife Management 78:763 –771.

Loss, S. R., T. Will, and P. P. Marra. 2014 b. Refining estimates of bird collision and electrocution mortality at power lines in the United States. PLoS ONE 9(7):e101565.

Loss, S. R., T. Will, S. S. Loss, and P. P. Marra. 2014 c. Bird –building collisions in the United States: estimates of annual mortality and species vulnerability. Condor: Ornithological Applications 116:8 –23.

Machtans, C. S., C. H. R. Wedeles, and E. M. Bayne. 2013. A first estimate for Canada of the number of birds killed by colliding with building windows. Avian Conservation and Ecology 8:6.

Martinson, L., M. Gerringer, and J. Lombardi. 2018 a. Post ‐construction monitoring at the Blythe Solar Power Project Riverside County, California. Report to Blythe Solar Energy Center, LLC, Juno Beach, Florida, USA.

Martinson, L., M. Gerringer, and J. Lombardi. 2018 b. Post ‐construction monitoring at the McCoy Solar Energy Project Riverside County, California first annual report draft 2016 –2017. Report to McCoy Solar, LLC, Juno Beach, Florida, USA.

Mathews, F., M. Swindells, R. Goodhead, T. A. August, P. Hardman, D. M. Linton, and D. J. Hosken. 2013. Effectiveness of search dogs compared with human observers in locating bat carcasses at wind ‐turbine sites: a blinded randomized trial. Wildlife Society Bulletin 37:34 –40.

May, R., A. B. Gill, J. Köppel, R. H. W. Langston, M. Reichenbach, M. Scheidat, S. Smallwood, C. C. Voigt, O. Hüppop, and M. Portman. 2017. Future research directions to reconcile wind turbine –wildlife interactions. Pages 255 –276 in J.

Köppel, editor. Wind energy and wildlife impacts: proceedings from the CWW2015 conference. Springer, Cham, Switzerland.

May, R., O. Reitan, K. Bevanger, S. ‐H. Lorensten, and T. Nygård. 2015. Mitigating wind ‐turbine induced avian mortality:
sensory, aerodynamic and cognitive constraints and options. Renewable and Sustainable Energy Reviews 42:
170–181.

McCrary, M. D., R. L. McKernan, R. W. Schreiber, W. D. Wagner, and T. C. Sciarrotta. 1986. Avian mortality at a solar energy plant. Journal of Field Ornithology 57:135 –141.

Miller, D. L. 2016. Distance: distance sampling detection function and abundance estimation. R package version 0.9.6. https://CRAN.R-project.org/package=Distance
Multiagency Avian ‐Solar Collaborative Working Group. 2016. Avian ‐solar science coordination plan. https://blmsolar.anl. gov/program/avian-solar/. Accessed 25 Aug 2021.

Nicolai, C., S. Abele, H. Beeler, R. Doster, E. Kershner, and T. McCabe. 2011. Monitoring migratory bird take at solar power facilities: an experimental approach. U.S. Fish and Wildlife Service –Pacific Southwest Region, Sacramento, California, USA.

Paula, J., M. C. Leal, M. J. Silva, R. Mascarenhas, H. Costa, and M. Mascarenhas. 2011. Dogs as a tool to improve bird ‐strike mortality estimates at wind farms. Journal for Nature Conservation 19:202 –208.

Péron, G., J. E. Hines, J. D. Nichols, W. L. Kendall, K. A. Peters, and D. S. Mizrahi. 2013. Estimation of bird and bat mortality at wind ‐power farms with superpopulation models. Journal of Applied Ecology 50:902 –911.

Reyes, G. A., M. J. Rodriguez, K. T. Lindke, K. L. Ayres, M. D. Halterman, B. R. Boroski, and D. S. Johnston. 2016.

Searcher efficiency and survey coverage affect precisio n of fatality estimates. Journal of Wildlife Management 80:1488 –1496.

Rioux, S., J. ‐P. L. Savard, and A. A. Gerick. 2013. Avian mortalities due to transmission line collisions: a review of current estimates and field methods with an emphasis on applications to the Canadian electric network. Avian Conservation and Ecology 8(2):7.

Rosenberg, K. V., A. M. Dokter, P. J. Blancher, J. R. Sauer, A. C. Smith, P. A. Smith, J. C. Stanton, A. Panjabi, L. Helft, M. Parr, and P. P. Marra. 2019. Decline of the North American avifauna. Science 366:120 –124.

Ruegg, K. C., C. M. B. Bossu, J. Rajbhandary, T. Fuller, R. Harrigan, and T. Smith. 2020. A genoscape framework for assessing the population ‐level impacts of renewable energy development on migratory bird species in California. California Energy Commission report CEC ‐500‐2020‐036, Sacramento, USA.

Shoener and Barrett's Biological Surveys. 2018. 2018 Spring/summer bat and bird carcass monitoring survey report Midway Solar Farm II. Barrett's Biological Surveys, El Centro, California, USA.

SOLAR IMPACTS | 27 of 29
Sinclair, K., and E. DeGeorge. 2016. Framework for testing the effectiveness of bat and eagle impact ‐reduction strategies at wind energy projects. Technical Report NREL/TP ‐5000‐65624. National Renewable Energy Laboratory, Golden, Colorado, USA.

Smallwood, K. S. 2007. Estimating wind turbine ‐caused bird mortality. Journal of Wildlife Management 71:2781 –2791.

Smallwood, K. S. 2013. Comparing bird and bat fatality ‐rate estimates among North American wind ‐energy projects.

Wildlife Society Bulletin 37:19 –33.

Smallwood, K. S. 2017 a. Long search intervals under ‐estimate bird and bat fatalities caused by wind turbines. Wildlife Society Bulletin 41:224 –230.

Smallwood, K. S. 2017 b. Monitoring birds. Chapter 1 in M. Perrow, editor. Wildlife and wind farms –conflicts and solutions.

Volume 2. Onshore: monitoring and mitigation. Pelagic Publishing, Exeter, United Kingdom.

Smallwood, K. S. 2020. USA wind energy ‐caused bat fatalities increase with shorter fatality search intervals. Diversity 12(3):

98. Smallwood, K. S., and D. A. Bell. 2020 a. Effects of wind turbine curtailment on bird and bat fatalities. Journal of Wildlife Management 84:684 –696.

Smallwood, K. S., and D. A. Bell. 2020 b. Relating bat passage rates to wind turbine fatalities. Diversity 12(2):84.

Smallwood, K. S., D. A. Bell, and S. Standish. 2020. Dogs detect larger wind energy impacts on bats and birds. Journal of Wildlife Management 84:852 –864.

Smallwood, K. S., D. A. Bell, E. L. Walther, E. Leyvas, S. Standish, J. Mount, B. Karas. 2018. Estimating wind turbine fatalities using integrated detection trials. Journal of Wildlife Management 82:1169 –1184.

Smallwood, K. S., L. Neher, and D. A. Bell. 2017. Turbine siting for raptors: an example from repowering of the Altamont Pass Wind Resource Area. Chapter 7 in M. Perrow, editor. Wildlife and wind farms – conflicts and solutions. Volume 2.

Onshore: monitoring and mitigation. Pelagic Publishing, Exeter, United Kingdom.

Smallwood, K. S., L. Neher, D. Bell, J. DiDonato, B. Karas, S. Snyder, and S. Lopez. 2009 a. Range management practices to reduce wind turbine impacts on burrowing owls and other raptors in the Altamont Pass Wind Resource Area, California. Final Report to the California Energy Commission, Public Interest Energy Research –Environmental Area, Contract No. CEC ‐500‐2008‐080, Sacramento, California, USA.

Smallwood, K. S., L. Rugge, and M. L. Morrison. 2009 b. Influence of behavior on bird mortality in wind energy developments: the Altamont Pass Wind Resource Area, California. Journal of Wildlife Management 73:1082 –1098.

UltraSystems. 2014 a. Imperial Solar Energy Center South post ‐construction avian monitoring biological monitoring report November 2013. Report to Csolar IV South, LLC. UltraSystems, Irvine, California, USA.

UltraSystems. 2014 b. Imperial Solar Energy Center South post ‐construction avian monitoring biological monitoring report December 2013. Report to Csolar IV South, LLC. UltraSystems, Irvine, California, USA.

UltraSystems. 2014c. Imperial Solar Energy Center South post ‐construction avian monitoring biological monitoring January 2014. Report to Csolar IV South, LLC. UltraSystems, Irvine, California, USA.

UltraSystems. 2014 d. Imperial Solar Energy Center South post ‐construction avian monitoring biological monitoring February 2014. Report to Csolar IV South, LLC. UltraSystems, Irvine, California, USA.

UltraSystems. 2014 e. Imperial Solar Energy Center South post ‐construction avian monitoring biological monitoring March 2014. Report to Csolar IV South, LLC. UltraSystems, Irvine, California, USA.

Walston, L. J., Jr., K. E. Rollins, K. E. LaGory, K. P. Smith, and S. A. Meyers. 2016. A preliminary assessment of avian mortality at utility ‐scale solar energy facilities in the United States. Renewable Energy 92:405 –414.

Warren‐Hicks, W., J. Newman, R. Wolpert, B. Karas, and L. Tran. 2013. Improving methods for estimating fatality of birds and bats at wind energy facilities. California Energy Commission, Sacramento, USA.

Western EcoSystems Technology. 2016. Post ‐construction monitoring at the Genesis Solar Energy Project Riverside County, California, 2015 –2016 First Annual Report. Report to Genesis Solar LLC, Juno Beach, Florida, USA.

Western EcoSystems Technology. 2017a. Post‐construction monitoring at the Desert Sunlight Solar Farm Project Riverside County, California, 2015 –2016 first annual report. Report to Desert Sunlight 250, LLC and Desert Sunlight 300, LLC, Juno Beach, Florida, USA.

Western EcoSystems Technology. 2017 b. Ivanpah Solar Electric Generating System avian & bat monitoring plan 2015–2016 annual report. Report to Solar Partners I, II, and VIII, Nipton, California, USA.

Western EcoSystems Technology. 2017 c. Post ‐construction monitoring at the Genesis Solar Energy Project Riverside County, California, second annual report 2016 –2017. Report to Genesis Solar LLC, Juno Beach, Florida, USA.

Western EcoSystems Technology. 2018a. Post‐construction monitoring at the Desert Sunlight Solar Farm Project Riverside County, California, second annual report 2016 –2017. Report to Desert Sunlight 250, LLC and Desert Sunlight 300, LLC, Juno Beach, Florida, USA.

Western EcoSystems Technology. 2018 b. Ivanpah Solar Electric Generating System avian & bat monitoring plan 2016–2017 annual report. Report to Solar Partners I, II, and VIII, Nipton, California, USA. 28 of 29 | SMALLWOOD
Western EcoSystems Technology. 2019. Ivanpah Solar Electric Generating System avian & bat monitoring plan 2017 –2018 annual report. Report to Solar Partners I, II, and VIII, Nipton, California, USA.

Yahner, R. H. 1982. Avian nest densities and nest ‐site selection in farmstead shelterbelts. Wilson Bulletin 94:156 –175.

Young, H. 1948. A comparative study of nesting birds in a five ‐acre park. Wilson Bulletin 61:36 –47.

Zimmerling, J. R., and C. M. Francis. 2016. Bat mortality due to wind turbines in Canada. Journal of Wildlife Management 80:1360–1369.

Zimmerling, J. R., A. C. Pomeroy, M. V. d'Entremont, and C. M. Francis. 2013. Canadian estimate of bird mortality due to collisions and direct habitat loss associated with wind turbine developments. Avian Conservation and Ecology 8:10.

Associate Editor: Courtney Duchardt .

SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher ’s website.

How to cite this article: Smallwood, K. S. 2022. Utility ‐scale solar impacts to volant wildlife. Journal of Wildlife Management e22216. https://doi.org/10.1002/jwmg.22216 SOLAR IMPACTS | 29 of 29
Citation: Smallwood, K.S.;
Smallwood, N.L. Measured Effects of
Anthropogenic Development on
Vertebrate Wildlife Diversity.

Diversity 2023, 15, 1037.
https://doi.org/10.3390/d15101037
Academic Editor: Xueyou Li
Received: 26 July 2023
Revised: 18 September 2023
Accepted: 21 September 2023
Published: 27 September 2023
Copyright: © 2023 by the authors.

Licensee MDPI, Basel, Switzerland.

This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/

4.0/).
diversity
Article
Measured Effects of Anthropogenic Development on Vertebrate Wildlife Diversity
K. Shawn Smallwood *
and Noriko L. Smallwood
Independent Researcher, 3108 Finch Street, Davis, CA 95616, USA; [email protected] * Correspondence: [email protected]
Abstract: A major driver of the declining biodiversity is landcover change leading to loss of habitat.

Many studies have estimated large-scale declines in biodiversity, but loss of biodiversity at a local scale due to the immediate effects of development has been poorly studied. California, in particular, is a biodiversity hotspot and has rapidly developed; thus, it is important to understand the effects of development on wildlife in the State. Here, we conducted reconnaissance surveys—a type of survey often used by consulting biologists in support of environmental review of proposed projects—to measure changes in the relative abundance and richness of vertebrate species in response to urban development. We completed 2 reconnaissance surveys at each of 52 control sites that remained undeveloped at the times of both surveys, and at each of 26 impact sites that had been developed by the time of the second survey. We completed the surveys as part of a before–after, control–impact (BACI) experimental design. Our main interest was the interaction effect between the before–after phases and the control–impact treatment levels, or the impact of development. After controlling for survey duration, we also tested for the effects of the number of years intervening the surveys in the before and after phases, project area size, latitude, degree of connectedness to adjacent open space, and whether the site was a redevelopment site, infill, or not infill. After development, the average number of vertebrate wildlife species we detected declined by 48% within the project area, and by 66% within the bounds of the project sites. Further, the average number of vertebrate animals we counted declined by 90% within the project area, and 89% within the bounds of the project sites. Development impacts measured by the mean number of species detected per survey were greatest for amphibians (−100%), followed by mammals (−86%), grassland birds (−75%), raptors (−53%), special-status species (−49%), all birds as a group (−48%), non-native birds (−44%), and synanthropic birds (−28%). Our results indicated that urban development substantially reduced vertebrate species richness and numerical abundance, even after richness and abundance had likely already been depleted by the cumulative effects of loss, fragmentation, and degradation of habitat in the urbanizing environment. Monitoring is needed in and around urbanizing areas to measure the cumulative effects of urbanization, and so are conservation measures to mitigate the effects of urbanization.

Keywords: BACI experiment; birds; California; development; reconnaissance survey; species richness; urbanization; vertebrate wildlife

1. Introduction
Urbanization has been defined as “the process of human settlement that gradually transforms uninhabited wildlands into lands including some degree of relatively perma- nent human presence” [1]. Urban growth profoundly affects the availability and condition of natural resources, and within its immediate area it often fragments and degrades habitat and simplifies biological species composition [2], as well as homogenizes species composi- tion of plants [3], arthropods [4–6], birds [7–11], and land-cover composition, landscape structure, and ecosystem functions [ 12]. Urban areas also reduce avian taxonomic di- Diversity 2023, 15, 1037. https://doi.org/10.3390/d15101037 https://www.mdpi.com/journal/diversity Diversity 2023, 15, 1037 2 of 39
versity [13]. Biodiversity is on the decline [ 14,15]. A major driver of declining trends in biodiversity within metropolitan areas is the extent of landcover that serves as habitat [11].

In the context of a city or metropolitan area, where habitat is lost to impervious sur- faces, and where habitat is degraded by noise, light, and air pollution and by sources of mortality [16], habitat loss is likely exacerbated by habitat fragmentation, which results in a cumulative net loss of species’ productive capacity that exceeds that of habitat loss alone [17]. Since habitat loss and habitat fragmentation have rapidly progressed around the world, the cumulative effects of these processes on wildlife are also rapidly advanc- ing [18]. Already, there have been documented genetic effects [19], and shifts in community composition and in morphologies and behaviors of species remaining within the areas of urbanization [16,20].

Many species of vertebrate wildlife have been in numerical decline across North America [21]. These declines have been attributed to multiple causal factors, but habitat loss and habitat fragmentation have usually been hypothesized as the leading causes of declines [11,22]. Habitat loss is readily believable because we can see and measure the extent to which we have been clearing natural vegetation to make way for agricultural, industrial, commercial, and residential uses and all of their connecting roads and highways, pipelines, and electrical transmission lines. Less measured, however, has been the actual changes in wildlife species composition and numerical abundance on sites where natural or managed vegetation has been removed to accommodate anthropogenic structures [1,22,23].

The effects of habitat loss due to development have more often been assumed or inferred from gradient experiments. To indicate the effects of habitat loss due to urban- ization, correlational analysis has been performed on bird species richness with variables intended to measure urbanization and degrees of departure from natural conditions [23–28].

Investigators in one study estimated the relative species richness of birds as an indicator of the effects of urbanization by comparing sampled species richness to a specified reference community or to the regional pool of species that should have existed prior to develop- ment [29]. The reference community would indicate a baseline ecological integrity, or the biological species assemblage during pristine conditions [ 30]. This approach, however, directly measured the effects of urbanization only if its pristine reference community was accurately specified. Direct measurements of the effects of habitat loss have been conducted in short-term studies lacking key elements of experimental design, and thus prone to finding equivocal to no effects of development on bird communities [ 31,32]. Long-term studies or experimental studies including controls to more directly test for the effects of urbanization are rare [1].

Urban development presents opportunities for experiments to measure the effects of urbanization on wildlife [1,31]. Realistically, however, these opportunities must make use of baseline environmental settings that are highly disturbed or consist of habitat fragments in an urbanizing landscape [31,32]. What can be measured in such experiments are only the later-stage, onsite effects of urbanization on biota. We had the opportunity to measure the effects of urbanization on vertebrate wildlife because we often survey for wildlife at sites proposed for development. The California Environmental Quality Act (CEQA) requires the characterization of the existing environmental setting. This characterization informs the public and decision-makers of what is at stake, and it serves as the baseline from which to opine on or predict project-caused impacts and to formulate appropriate mitigation measures. To this end, consulting biologists usually perform what are referred to as reconnaissance surveys, otherwise known as general biological surveys. These surveys typically include one or more biologists walking over the project site or scanning the site from vantage points. The surveys vary in duration without any clear stopping rules, but typically last one to several hours. Following the consultants’ surveys, we were often hired by parties other than the project applicants or the permitting agencies to also survey the project sites, and sometimes to survey project sites that had not been surveyed previously.

We managed our surveys of project sites in the framework of a before–after, control– impact experiment to measure average project impacts to wildlife. Our control sites Diversity 2023, 15, 1037 3 of 39
were those project sites that had not been developed prior to our second survey, which represented the after phase of the experiment, and the impact sites were those sites that had been developed by the time of our second survey. Our primary objective was to measure changes in the local vertebrate wildlife community caused by development, based on the following metrics: (1) the number of vertebrate species detected, (2) the number of species uniquely detected at a site in one phase relative to the species detected in the other phase, (3) counts of live animals, and (4) the percentage of project sites within each BACI treatment group where we detected each species. Our secondary objective was to explore whether other measured variables might explain the residual variation from the analysis of variance (ANOVA) model used to test the BACI hypotheses for main and interaction effects. We additionally tested for effects of survey duration, years intervening the surveys in the before and after phases, project area, latitude, level of site disturbance, degree of connectedness to adjacent open space, and whether the site was a redevelopment site, infill, or not infill.

2. Materials and Methods

2.1. Study Area
Our study sites were clustered in the Sacramento and Central Valleys, the San Francisco Bay Area, and the coastal region of southern California (Figure 1). Each of these sites was selected because applications had been submitted for development. We later added follow- up surveys where practical and when we could closely match the date and start time of the initial surveys. Twenty-six of the sites had been developed by the time of our follow-up surveys, whereas fifty-two sites remained undeveloped. Developed sites were those for which the intended structures of the project were completely or nearly completely built, but they did not have to be occupied (some structures remained vacant for extended periods) (Figure 2). Thirty-five of the sites were within or peripheral to existing urban, commercial, or industrial areas, but thirty were located on agricultural or desert landscapes (Table A1).

The condition of most of our study sites was poor to moderate at the times of our initial surveys, as most sites had been disturbed by mechanical clearing of vegetation (Figure 3), frequent fires, off-road vehicle use, invasive plant species composition (Figure 4), or by other forms of pollution, e.g., dumping of waste materials and mowing for weed abatement.

Only four of the sites were surrounded on all sides by open space. Study sites ranged from

0.526 ha to 1,549 ha (mean = 91.26 ha, SD = 255.44 ha), with the two largest study sites consisting of natural reserves, which we used as control sites.

2.2. Reconnaissance Surveys
We performed what are referred to in California as reconnaissance surveys, also known as general biological surveys. We intentionally implemented the same methodology as used by environmental consultants when they perform reconnaissance surveys. In these surveys, all species are recorded if detected by visual or auditory means or by signs such as burrows, tracks, or scat. Wildlife recorded included birds, mammals, amphibians, and reptiles. We surveyed by walking the perimeter of the site, or by a stationary vantage point, where we scanned for wildlife with the use of binoculars. We recorded those animals that were onsite, i.e., within the boundary of the project site, and which we refer to hereafter as onsite. We also recorded animals in the project area, which included those onsite and those we judged were close enough to the project site to readily make use of it, which was usually ≤100 m from the site. In most surveys, we recorded the time within the survey when a new species was detected, whether the species was on the project site or in the surrounding area, and the approximate abundance of that species. We recorded temperature, wind, and sky conditions at the start and end of most surveys, and we recorded ground conditions at all sites at the time of each survey.

Diversity 2023, 15, 1037 4 of 39
Diversity 2023, 15, x FOR PEER REVIEW 4 of 43

Figure 1. Locations of 78 project sites in ( a) northern California and ( b) southern California where we completed reconnaissance surveys used in a before–after, control–impact (BACI) experiment of the effects of development (habitat loss) on species of vertebrate wildlife. Numbers refer to before and after pairs of surveys, which are described in Table A1.

Figure 1. Locations of 78 project sites in (a) northern California and (b) southern California where we completed reconnaissance surveys used in a before–after, control–impact (BACI) experiment of the effects of development (habitat loss) on species of vertebrate wildlife. Numbers refer to before and after pairs of surveys, which are described in Table A1.

Diversity 2023, 15, 1037 5 of 39
Diversity 2023, 15, x FOR PEER REVIEW 5 of 43

(a)

(b)
Figure 2. Cont.

Diversity 2023, 15, 1037 6 of 39
Diversity 2023, 15, x FOR PEER REVIEW 6 of 43

(c)
Figure 2. (a) GLP Warehouse project site on 12 February 2020 (a, top) and 18 February 2022 (a, bot- tom). (b) First Industrial Warehouse project site on 28 February 2020 ( b, top) and 5 February 2023 (b, bottom). (c) Winters Highlands residential project site on 18 May 2004 (c, top) and 11 June 2021 (c, bo ttom). A pair of burrowing owls are visible in the foreground, center-left aspect of the top photo.

Figure 2. (a) GLP Warehouse project site on 12 February 2020 ( a, top) and 18 February 2022 (a, bottom). ( b) First Industrial Warehouse project site on 28 February 2020 ( b, top) and 5 February 2023 (b, bottom). (c) Winters Highlands residential project site on 18 May 2004 (c, top) and 11 June 2021 ( c, bottom). A pair of burrowing owls are visible in the foreground, center-left aspect of the top photo.

Diversity 2023, 15, 1037 7 of 39
Diversity 2023, 15, x FOR PEER REVIEW 7 of 43

Figure 3. Western meadowlark on the Mango Avenue Warehouse project site in Fontana, despite the ground disturbance caused by human activities.

Figure 4. Despite the dominant tree cover of non-native blue gum eucalyptus ( Eucalyptus globulus) on Mount Sutro Open Space Reserve, which was one of our control sites, Pacific wren continued to thrive onsite, along with numerous other species of vertebrate wildlife.

2.2. Reconnaissance Surveys
We performed what are referred to in Cali fornia as reconnaissance surveys, also known as general biological surveys. We intentionally implemented the same methodol- ogy as used by environmental consultants when they perform reconnaissance surveys. In Figure 3. Western meadowlark on the Mango Avenue Warehouse project site in Fontana, despite the ground disturbance caused by human activities.

Diversity 2023, 15, x FOR PEER REVIEW 7 of 43

Figure 3. Western meadowlark on the Mango Avenue Warehouse project site in Fontana, despite the ground disturbance caused by human activities.

Figure 4. Despite the dominant tree cover of non-native blue gum eucalyptus ( Eucalyptus globulus) on Mount Sutro Open Space Reserve, which was one of our control sites, Pacific wren continued to thrive onsite, along with numerous other species of vertebrate wildlife.

2.2. Reconnaissance Surveys
We performed what are referred to in Cali fornia as reconnaissance surveys, also known as general biological surveys. We intentionally implemented the same methodol- ogy as used by environmental consultants when they perform reconnaissance surveys. In Figure 4. Despite the dominant tree cover of non-native blue gum eucalyptus (Eucalyptus globulus) on Mount Sutro Open Space Reserve, which was one of our control sites, Pacific wren continued to thrive onsite, along with numerous other species of vertebrate wildlife.

Reconnaissance surveys are unbound by time, but typically last between one and several hours. We stopped our initial survey at each project site once our species detection rate declined to about one new species per 20 or 30 min, similar to the rule advocated by Diversity 2023, 15, 1037 8 of 39
Watson [33]. These lower detection rates typically coincided with the increasing heat of the day or oncoming darkness of the evening. In the cases of our second surveys to represent the after phase of the experiment, we stopped each survey at the time it took to record all of the species that had been recorded in each initial survey.

Beginning in January 2020, we began to resurvey sites of proposed building projects that we had originally surveyed during the same time of year and about the same time of day between 1 and 19 years earlier. In addition to starting the repeat survey as close to the original start time as possible, we surveyed for the same duration and using the same person, or both of us if we had originally surveyed together. Of the sites we resurveyed, 52 had remained undeveloped and 26 had been developed since our initial survey. Upon each repeat survey, we assigned sites that remained undeveloped to the control group, and sites that were since developed to the impact group. We applied the same survey standards between control and impact sites.

2.3. BACI Tests
We compared our survey outcomes in a BACI experimental design. One metric of the survey outcome was the total number of vertebrate species seen during the survey, including species seen in the project area but off the project site. A second metric was the total number of species seen only on the project site. A third metric was the number of vertebrate species detected solely offsite. A fourth metric was the number of species uniquely detected at a site in one phase relative to the other phase. A fifth metric was the total number of live animals counted during the survey (excluding fossorial mammals indicated by signs of burrow activity). A sixth metric was the number of sites at which a species or group of species was detected. For each metric, we quantified the expected outcome at impact sites (EIA) relative to the before–after change in outcomes at the control site, and the effect of the development on the impact treatment level:

EIA =
( CA
CB
)
× IB
E f f ect = (IA − EIA )
EIA
× 100%,
where treatment levels were CB = control-before, CA = control-after, IB = impact-before, and IA = impact-after.

We compared rates of species detections with increasing time into the survey by the following experimental treatment groups: control-before (CB), control-after (CA), impact- before (IB), and impact-after (IA). To arrive at these comparisons, we fit a nonlinear least- squares regression model to the cumulative number of species detected, Y, as a function of the number of minutes, X, into each survey. The form of the model was Y = 1 1 a +b×(X+1)−c ,
where X represents minutes into the survey, anda, b, and c were best-fit coefficients. Since the surveys varied in duration, this modeling approach was also useful for minimizing the effect of survey duration on the metric, number of species detected. We did this by using each model to predict the number of species that would be detected after one hour. We chose 1 h because it was within the data range of all but 4 of the surveys we completed. We also projected the model to 5 h for comparison.

We used 2-factor analysis of variance with interest mostly in the significance of the interaction effect between the before–after time period (BA) and the control–impact treat- ment (CI) of each BACI experiment, as implemented by others [ 34,35]. We performed the tests on data collected from the project areas as well as those strictly from onsite. Ex- cept for our model-predicted number of species detected after one hour of surveying, we log10-transformed count variables, such as the number of species detected and number of animals counted. We visually examined normal probability plots, and we performed Hartley’s F-max test for homogeneity of variance to determine whether our tests met the Diversity 2023, 15, 1037 9 of 39
assumptions of ANOVA. The assumptions were met in nearly every test. To further assess the 2-factor ANOVA interaction effects, we calculated their statistical power.

2.4. Effects of Other Factors
From the BACI test of the number of vertebrate species detected, we saved ANOVA model residuals for exploring whether additional variation in the data could be explained by factors represented by other measured variables. We did the same for the BACI test of the model-predicted number of vertebrate species detected at 1 h. We regressed both sets of residuals on survey duration (minutes of survey common to both the before and after phases) to compare the degrees to which the effect of survey duration had been reduced, and hence to decide which set of residuals to use in exploration of the effects of other factors, such as landscape and site attributes. Our objective was to maximally control for the effect of survey duration when testing whether and how the number of vertebrate species detected related to project size (ha), latitude, number of years between the surveys in the before and after phases, the similarity index [ 36], and as described in the following paragraph, the intensity of pre-survey actions that would have suppressed wildlife, landscape settings such as whether the site was infill, redevelopment, or surrounded by open space, and the site disturbance rating. We note that whereas the similarity index was intended to measure the similarity of community composition of constituent species, its true measurement must also be of species detection probabilities attributable to the surveys.

We categorized an urban setting index for each site as 0 = largely non-urbanized, 1 = urban infill, and 2 = redevelopment. We rated the connectivity of project site borders to adjacent open space (including agriculture) as 0%, 10%, 25%, 50%, 75%, and 100%. We categorized sites as having undisturbed vegetation; evidence of disturbance over the last 5 years or so; ruderal; mowed; neglected by accumulation of trash, construction debris, waste soil or machine parts; neglected by cessation of irrigation; burned; disked; graded; construction ongoing, or constructed, as well as combinations of the foregoing categories.

From these categories, we rated sites for the level of disturbance as: 1 = natural and biolog- ically intact, with no more than small patches of non-native vegetation; 2 = mostly intact, with some native and some non-native vegetation, or all native with some past ground disturbance; 3 = modified (disked or highly disturbed) in the past but with a substantial ex- tent of vegetation, such as patches of shrubs or scattered trees; 4 = landscaped parks or golf courses; 5 = agriculture, including orchards and vineyards; 6 = agriculture, including row crops; 7 = parking lot with mature shrubs or trees, and where buildings do not cover the entire site; 8 = highly modified with little vegetation remaining; 9 = compacted, pervious ground with no vegetation remaining; 10 = impervious ground with no vegetation remain- ing; 11 = constructed buildings. We further categorized site conditions to represent the intensity of actions that likely would have suppressed wildlife as: 0 = none evident,1 = low (ruderal, cleared fire break), 2 = routine disturbance, 3 = moderate (mowed, neglected), 4 = earlier intense (near-recent grading, regrowth after disking), 5 = intense (cleared, disked, disked and neglected), and 6 = very intense (converted to crop, graded, constructed).

2.5. Species Characteristics
We compared the species detected among surveys to identify the frequency that each was found in the before and after phases and between the control and impact treatment levels. We further grouped species into classes, including amphibians, reptiles, mammals, birds, grassland birds, raptors, synanthropic birds, non-native species, and special-status species. The latter class was informed by legal protections afforded species by state and federal statutes and by designations assigned to species by state and federal wildlife agencies (species names, species groupings, and special-status species are listed in Table A2).

We measured development impacts to these classes by the mean number of species within each that was detected per survey.

Diversity 2023, 15, 1037 10 of 39

3. Results
Impact sites differed from control sites in several ways, including their average smaller size, lower elevation, and 94 km more northerly locations (Table 1). On average, impact sites were half to less than half connected to open space, as compared to control sites.

Impact sites also ranked higher on the urban setting index, which meant they were more likely to be infill or redevelopment projects. Furthermore, impact sites rated higher for the level of disturbance, even prior to development, and they ranked higher on the intensity of actions resulting in suppression of wildlife occurrences, even prior to development (Table 1). On average, the survey duration was briefer on impact sites by nearly half an hour, and the time between the first and second surveys was longer by 1.3 years, but the average difference in start times was insubstantial.

Table 1. Summary of survey, site, and landscape attributes of the 78 project sites we surveyed in California, in 2002–2023, where the treatment levels were: CB = control-before, CA = control-after, IB = impact-before, and IA = impact-after.

Metric
CB (n = 52) CA (n = 52) IB (n = 26) IA (n = 26)
/macron.ts1
X SD /macron.ts1
X SD /macron.ts1
X SD /macron.ts1
X SD
Size of project site (Hectares) 131 310 16.25 30.22
Elevation (m) 175 214 138 243
Northing (m) 3,994,623 241,749 4,088,256 213,973
Urban setting 0.55 0.61 0.77 0.51
Connectivity (%) 35.0 30.8 35.0 30.8 17.3 19.7 15.4 17.4 Project site disturbance 4.12 2.70 4.13 2.70 5.20 2.69 10.58 1.65 Rating of suppressive actions 1.24 1.92 1.51 2.24 2.58 2.47 5.50 1.63 Survey duration (minutes) 128 47 125 46 96 38 96 38
Years since first survey 2.7 3.9 4.0 4.2
Start time difference (minutes) −3.6 38.0 −2.9 34.4

3.1. BACI Experiment
As part of our experiment, we completed 78 pairs of before and after surveys, or 156 surveys. Our cumulative number of species detections increased with the increasing survey duration, but the rates of these increases differed between sites in the control and impact treatment levels, and the mean rate was slowest among sites in the impact-after group, i.e., the sites that had been developed (Figure 5a). The model-predicted number of species detected by 1 h into a survey averaged about 10.4 in the IA group, as compared to 20.8 to 21.8 in the CB and CA groups. By 5 h, the disparity increased to 12.7 species detected in the IA group, as compared to 37.6 in the CB and CA groups (Figure 5b). At 1 h, the model-predicted number of species detected was 51% lower in the IA group, but at 5 h it was 66% lower.

We observed large changes in species composition and relative abundance among the project sites that were developed before our second survey. Some of the species we detected in the before phase were relatively abundant, but their abundance sharply declined after development. For example, at the CenterPoint Warehouse Project site in Manteca, our before and after counts changed from 300 to 9 American crows, from 40 to 3 mourning doves, from 400 to 0 western meadowlarks, and from 30 to 0 house finches. On average, we counted 88% fewer vertebrate animals, including 85% fewer animals of special-status species, on impact sites after development, and we detected 44% fewer vertebrate species on impact sites after development, and 62% fewer vertebrate species on the project footprint (Table 2).

Diversity 2023, 15, 1037 11 of 39
Diversity 2023, 15, x FOR PEER REVIEW 11 of 43

the model-predicted number of species detected was 51% lower in the IA group, but at 5 h it was 66% lower.

Figure 5. Mean and 95% confidence intervals (CI) of the model-predicted number of vertebrate wild- life species detected by minute into the reconnaissance survey, and extended to 5 h (a) and only 1 h (b), where for each survey the model fit to the cumulative number of species detected, Y, was of the form: Y=
ଵ
ଵ ୟൗ ାୠ×(ଡ଼ାଵ)షౙ, where X represents minutes into the survey, and a, b, and c are the best- fit coefficients. The coefficient of determination, r2, averaged among the models fit to the data.

We observed large changes in species composition and relative abundance among the project sites that were developed before our second survey. Some of the species we detected in the before phase were relatively abundant, but their abundance sharply de- clined after development. For example, at the CenterPoint Warehouse Project site in Man- teca, our before and after counts changed from 300 to 9 American crows, from 40 to 3 mourning doves, from 400 to 0 western meadowlarks, and from 30 to 0 house finches. On average, we counted 88% fewer vertebrate animals, including 85% fewer animals of spe- cial-status species, on impact sites after de velopment, and we detected 44% fewer verte- brate species on impact sites after development, and 62% fewer vertebrate species on the project footprint (Table 2).

Table 2. Mean and standard deviation of the number of species detected and the number of animals counted in the reconnaissance surveys, and the numb er of sites (N) within the BACI experimental treatment levels: control-before (CB), control-afte r (CA), impact-before (IB) , and impact-after (IA), where impact sites were those at which a proposed project had been developed prior to the survey that was completed in the after phase of the study. All project sites were in California. Unique spe- cies per survey refers to the number of uniquely de tected species at a site in one phase relative to the species detected in the other phase.

Metric CB CA IB IA Effect 𝑿 SD N 𝑿 SD N 𝑿 SD N 𝑿 SD N
Number of species
All vertebrates 26.4 11.1 51 28.2 10.7 51 19.1 5.9 26 10.7 3.7 26 −48 Onsite vertebrates 22.8 12.9 30 25.5 12.6 30 17.4 6.6 20 6.6 3.3 20 −66 Offsite vertebrates 5.0 8.85 30 3.55 3.50 30 1.25 1.58 20 3.85 2.89 20 334 0 50 100 150 200 250 300
Minutes into survey
0 5
10 15
20 25
30 35
40 45
Control Before
Control After
Impact Before
Impact After
Mean & 95% CI
0 1 02 03 04 05 06 0
Minutes into survey
0 2
4 6
8 10
12 14
16 18
20 22
24 Mean and 95% CI of cumulative number
of wildlife species detected
ab
Figure 5. Mean and 95% confidence intervals (CI) of the model-predicted number of vertebrate wildlife species detected by minute into the reconnaissance survey, and extended to 5 h (a) and only 1 h (b), where for each survey the model fit to the cumulative number of species detected, Y, was of the form: Y = 1
1/a+b×(X+1)−c , where X represents minutes into the survey, anda, b, and c are the best-fit coefficients. The coefficient of determination, r 2, averaged among the models fit to the data.

We also observed changes in species composition and relative abundance among the project sites that did not undergo development and which we treated as our control sites. For example, at the Operon HKI Project site in Perris, our before and after counts shifted from 10 to 20 savannah sparrows (Figure 6), 0 to 20 western meadowlarks, and 0 to 20 horned larks. On average, we detected about 2–3 additional species in most groups of species during our second surveys among the control sites, and we counted about 26% more birds, but notably we counted 56% fewer species, and 44% fewer animals of special-status species vertebrate wildlife (Table 2).

In the before phase, the number of species detected averaged fewer at the impact sites compared to the controls, whereas the number of animals counted averaged more at the impact sites compared to the controls (Table 2). Consequently, the control–impact main effects were significant for all but one of the metrics consisting of the number of species detected, whereas they were not significant for any of the metrics of the number of animals counted (Table 3). The before–after main effects were significant for the number of vertebrate species detected and the number of birds detected, but not for the numbers of species detected of mammals or reptiles and amphibians. The before–after main effects were significant for all the metrics of the number of animals counted. However, whereas these main effects point towards potential biases, our main interest was in the interaction effect, which informs of the impact of the action (development), and which presumably would have been largely controlled for in the experiment.

Diversity 2023, 15, 1037 12 of 39
Table 2. Mean and standard deviation of the number of species detected and the number of animals counted in the reconnaissance surveys, and the number of sites (N) within the BACI experimental treatment levels: control-before (CB), control-after (CA), impact-before (IB), and impact-after (IA), where impact sites were those at which a proposed project had been developed prior to the survey that was completed in the after phase of the study. All project sites were in California. Unique species per survey refers to the number of uniquely detected species at a site in one phase relative to the species detected in the other phase.

Metric

CB CA IB IA
Effect/macron.ts1
X SD N /macron.ts1
X SD N /macron.ts1
X SD N /macron.ts1
X SD N
Number of species
All vertebrates 26.4 11.1 51 28.2 10.7 51 19.1 5.9 26 10.7 3.7 26 −48 Onsite vertebrates 22.8 12.9 30 25.5 12.6 30 17.4 6.6 20 6.6 3.3 20 −66 Offsite vertebrates 5.0 8.85 30 3.55 3.50 30 1.25 1.58 20 3.85 2.89 20 334 All birds 23.5 9.6 51 25.2 9.4 51 17.6 5.9 26 10.3 3.6 26 −45 Onsite birds 19.9 11.3 30 22.5 11.1 30 16.0 6.4 20 6.5 3.3 20 −64 All mammals 2.5 2.3 51 2.4 1.9 51 1.3 0.9 26 0.3 0.6 26 −79 Onsite mammals 2.5 2.4 30 2.3 2.0 30 1.3 1.0 20 0.1 0.3 20 −92 All herps 0.4 0.7 51 0.5 0.8 51 0.2 0.4 26 0.1 0.3 26 −47 Onsite herps 0.5 0.9 27 0.6 0.7 27 0.1 0.3 20 0.0 0.0 20 −100 All non-birds 2.9 2.6 51 3.0 2.1 51 1.5 1.0 26 0.4 0.8 26 −75 Onsite non-birds 2.9 2.9 30 2.8 2.3 30 1.4 1.0 20 0.1 0.2 20 −96 All special-status species 5.2 2.8 51 4.9 2.3 51 3.7 2.1 26 1.8 1.0 25 −49 Onsite special-status species 4.3 3.1 30 4.3 2.5 30 3.0 1.8 20 1.3 1.3 19 −58 Model-predicted at 1 h 20.0 6.4 42 21.3 5.2 51 16.9 4.5 14 11.3 4.8 25 −37 Unique species per survey 9.9 5.2 51 11.6 4.9 51 12.6 4.8 27 3.9 2.2 27 −74 Animals counted
All vertebrates 155.4 117.0 19 187.2 157.4 19 358.7 259.0 15 42.9 23.7 15 −90 Onsite vertebrates 107.7 144.3 14 99.0 80.9 14 335.4 284.4 14 34.9 18.2 14 −89 All birds 135.6 102.6 19 183.3 154.3 19 354.0 255.3 15 42.5 23.5 15 −91 Onsite birds 98.6 130.4 14 96.7 80.4 14 330.5 280.3 14 34.8 18.1 14 −89 All special-status species 30.9 67.5 19 17.3 22.2 19 31.8 51.0 15 5.4 6.6 14 −70 Onsite special-status species 11.8 16.0 14 9.9 12.5 14 35.1 54.1 13 5.3 7.4 12 −82 Diversity 2023, 15, x FOR PEER REVIEW 12 of 43

All birds 23.5 9.6 51 25.2 9.4 51 17.6 5.9 26 10.3 3.6 26 −45 Onsite birds 19.9 11.3 30 22.5 11.1 30 16.0 6.4 20 6.5 3.3 20 −64 All mammals 2.5 2.3 51 2.4 1.9 51 1.3 0.9 26 0.3 0.6 26 −79 Onsite mammals 2.5 2.4 30 2.3 2.0 30 1.3 1.0 20 0.1 0.3 20 −92 All herps 0.4 0.7 51 0.5 0.8 51 0.2 0.4 26 0.1 0.3 26 −47 Onsite herps 0.5 0.9 27 0.6 0.7 27 0.1 0.3 20 0.0 0.0 20 −100 All non-birds 2.9 2.6 51 3.0 2.1 51 1.5 1.0 26 0.4 0.8 26 −75 Onsite non-birds 2.9 2.9 30 2.8 2.3 30 1.4 1.0 20 0.1 0.2 20 −96 All special-status species 5.2 2.8 51 4.9 2.3 51 3.7 2.1 26 1.8 1.0 25 −49 Onsite special-status species 4.3 3.1 30 4.3 2.5 30 3.0 1.8 20 1.3 1.3 19 −58 Model-predicted at 1 h 20.0 6.4 42 21.3 5.2 51 16.9 4.5 14 11.3 4.8 25 −37 Unique species per survey 9.9 5.2 51 11.6 4.9 51 12.6 4.8 27 3.9 2.2 27 −74 Animals counted
All vertebrates 155.4 117.0 19 187.2 157.4 19 358.7 259.0 15 42.9 23.7 15 −90 Onsite vertebrates 107.7 144.3 14 99.0 80.9 14 335.4 284.4 14 34.9 18.2 14 −89 All birds 135.6 102.6 19 183.3 154.3 19 354.0 255.3 15 42.5 23.5 15 −91 Onsite birds 98.6 130.4 14 96.7 80.4 14 330.5 280.3 14 34.8 18.1 14 −89 All special-status species 30.9 67.5 19 17.3 22.2 19 31.8 51.0 15 5.4 6.6 14 −70 Onsite special-status species 11.8 16.0 14 9.9 12.5 14 35.1 54.1 13 5.3 7.4 12 −82 We also observed changes in species composition and relative abundance among the project sites that did not undergo development and which we treated as our control sites.

For example, at the Operon HKI Project site in Perris, our before and after counts shifted from 10 to 20 savannah sparrows (Figure 6), 0 to 20 western meadowlarks, and 0 to 20 horned larks. On average, we detected about 2–3 additional species in most groups of species during our second surveys among th e control sites, and we counted about 26% more birds, but notably we counted 56% fewer species, and 44% fewer animals of special- status species vertebrate wildlife (Table 2).

Figure 6. A savannah sparrow at the Operon HKI project site in Perris on 21 November 2021, which was the date of the first survey in the before phase. Twice as many savannah sparrows were counted in the second survey at this site, which was not developed. Where projects were developed, devel- opment impacts reduced savannah sparrow counts by 63% on average.

In the before phase, the number of species detected averaged fewer at the impact sites compared to the controls, whereas the number of animals counted averaged more at the impact sites compared to the controls (Table 2). Consequently, the control–impact main Figure 6. A savannah sparrow at the Operon HKI project site in Perris on 21 November 2021, which was the date of the first survey in the before phase. Twice as many savannah sparrows were counted in the second survey at this site, which was not developed. Where projects were developed, development impacts reduced savannah sparrow counts by 63% on average.

Diversity 2023, 15, 1037 13 of 39
Table 3. Before–after, control–impact (BACI) comparisons of log10 number of vertebrate species detected and log10 number of animals counted in the reconnaissance surveys at sites of proposed projects in California, where the CI × BA interaction effect is the principal effect of interest, but tests for main effects are also reported.

Also reported are determinations of whether the data were normally distributed based on visual examination of normal probability plots, the p-value of Hartley’s F-max test for homogeneity of variance, and statistical power (1 − β, where β is the probability of a Type II error), estimated for the interaction effect. Unique species per survey refers to the number of uniquely detected species at a site in one phase relative to the species detected in the other phase.

Metric Normally Distributed? Hartley’s F-Maxp-Value
Control–Impact Main Effect Before–After Main Effect CI × BA Interaction Effect F p F p F p 1 − β
Number of species
All vertebrates Yes 0.6161 99.34 0.0000 16.31 0.0001 28.23 0.0000 1.00 Onsite vertebrates Yes 0.7228 65.44 0.0000 21.31 0.0000 30.54 0.0000 1.00 Offsite vertebrates Yes 0.0002 5.98 0.0163 16.71 0.0001 19.72 0.0000 0.99 All birds Yes 0.7747 84.00 0.0000 12.52 0.0005 23.90 0.0000 1.00 Onsite birds Yes 0.6205 49.67 0.0000 15.93 0.0001 26.42 0.0000 1.00 All mammals Yes 0.0600 8.52 0.0043 0.60 0.4408 0.03 0.8560 0.05 Onsite mammals Yes 0.1827 5.78 0.0192 1.20 0.2783 0.60 0.4402 0.12 All herps Yes 1.0000 3.07 0.0888 0.01 0.9104 0.01 0.9104 0.05 Onsite herps
All non-birds Yes 0.0622 14.50 0.0002 1.19 0.2770 0.56 0.4576 0.11 Onsite non-birds Yes 0.0837 4.24 0.0436 0.71 0.4029 0.47 0.4947 0.10 All special-status species Yes 0.6230 46.00 0.0000 12.18 0.0006 12.18 0.0006 0.93 Onsite special-status species Yes 0.2806 12.05 0.0008 5.26 0.0242 4.71 0.0327 0.57 Model-predicted at 1 h Yes 0.2336 37.04 0.0000 4.06 0.0459 10.32 0.0017 0.89 Unique species per survey Yes 0.7910 24.05 0.0000 36.01 0.0000 71.50 0.0000 1.00 Animals counted
All vertebrates Yes 0.2890 2.44 0.1235 17.80 0.0001 32.11 0.0000 1.00 Onsite vertebrates Yes 0.0035 0.02 0.8842 7.40 0.0088 11.77 0.0012 0.92 All birds Yes 0.2916 1.62 0.2076 15.63 0.0002 34.79 0.0000 1.00 Onsite birds Yes 0.0040 0.00 0.9964 6.68 0.0126 11.86 0.0011 0.92 All special-status species Yes 0.8945 1.96 0.1672 10.11 0.0023 4.76 0.0331 0.57 Onsite special-status species Yes 0.8629 0.84 0.3647 9.87 0.0030 6.15 0.0171 0.68 Diversity 2023, 15, 1037 14 of 39
After development, the average number of vertebrate wildlife species we detected declined by 48% within the project area, and by 66% within the bounds of the project sites (Table 2, Figure 7a,b). These declines were significant (Table 3). At the same time, the number of vertebrate wildlife species we detected solely offsite increased by 334% (Table 2, Figure 7c), which was significant (Table 3).

Diversity 2023, 15, x FOR PEER REVIEW 14 of 43

Figure 7. BACI tests revealed that development reduced the number of species detected by 48% (a) and the number of species detected within the bounds of the study site by 66% ( b) and increased the number of species detected solely o ffsite by 334% ( c). The red dashed arrow points to the ex- pected value where no development impact occurred.

The average number of bird species declined by 45% within the project area, and by 64% within the bounds of the project sites (Table 2). These declines were also signi ficant (Table 3). Although not significant due to insufficient statistical power (Table 3), the aver- age number of mammal species declined by 79% across the entire viewshed and by 92% within the bounds of the project sites, and the average number of amphibians and reptiles (“herps”) declined by 47% in the project area, and by 100% within the bounds of the pro- ject sites (Table 2).

After development, the average number of vertebrate animals we counted declined by 90% within the project area (Figure 8a), and by 89% within the bounds of the project

0.9 1.0

1.1 1.2

1.3 1.4

1.5log10 Number of vertebrate species detected
F1, 154 = 28.23, p = 0.0000
Before After
, 95% CI
Control
Impact 0.6

0.7 0.8

0.9 1.0

1.1 1.2

1.3 1.4

1.5 log10 Number of vertebrate species detected on site Before After
, 95% CI
Control
Impact
F1, 100 = 30.54, p = 0.0000
ab

0.00 0.05

0.10 0.15

0.20 0.25

0.30 log10 Number of vertebrate species detected off site F1, 148 =19.72, p = 0.0000
, 95% CI
Control
Impact
Before After
c
Figure 7. BACI tests revealed that development reduced the number of species detected by 48% (a) and the number of species detected within the bounds of the study site by 66% (b) and increased the number of species detected solely offsite by 334% (c). The red dashed arrow points to the expected value had no development impact occurred.

The average number of bird species declined by 45% within the project area, and by 64% within the bounds of the project sites (Table 2). These declines were also significant (Table 3). Although not significant due to insufficient statistical power (Table 3), the average Diversity 2023, 15, 1037 15 of 39
number of mammal species declined by 79% across the entire viewshed and by 92% within the bounds of the project sites, and the average number of amphibians and reptiles (“herps”) declined by 47% in the project area, and by 100% within the bounds of the project sites (Table 2).

After development, the average number of vertebrate animals we counted declined by 90% within the project area (Figure 8a), and by 89% within the bounds of the project sites (Table 2). These declines were significant (Table 3). The average number of birds we counted declined by 91% within the project area (Table 2, Figure 8b), and by 89% within the bounds of the project sites (Table 2), both of which were significant (Table 3).

Diversity 2023, 15, x FOR PEER REVIEW 15 of 43

sites (Table 2). These declines were significant (Table 3). The average number of birds we counted declined by 91% within the project area (Table 2, Figure 8b), and by 89% within the bounds of the project sites (Table 2), both of which were significant (Table 3).

Figure 8. BACI tests revealed that development reduced the counts of observed animals by 90% (a) and the counts of birds by 91% ( b). The red dashed arrow points to the expected value where no development impact occurred.

After development, the average number of special-status species declined by 49% within the project area, and by 58% within the bounds of the project sites (Table 2, Figure 9a). The average number of vertebrate animals of special-status species that we counted declined by 70% within the project area, and by 82% within the bounds of the project sites (Table 2, Figure 9b). All of these declines were significant (Table 3).

1.2 1.4

1.6 1.8

2.0 2.2

2.4 2.6

2.8 F1, 68 = 32.11, p = 0.0000
Before After
Log10 Counts of vertebrate animals
, 95% CI
Control
Impact
ab

1.2 1.4

1.6 1.8

2.0 2.2

2.4 2.6

2.8 log10 Counts of birds
Before After
, 95% CI
Control
Impact
F1, 68 = 34.79, p = 0.0000
Figure 8. BACI tests revealed that development reduced the counts of observed animals by 90% (a) and the counts of birds by 91% (b). The red dashed arrow points to the expected value had no development impact occurred.

After development, the average number of special-status species declined by 49% within the project area, and by 58% within the bounds of the project sites (Table 2, Figure 9a).

The average number of vertebrate animals of special-status species that we counted declined by 70% within the project area, and by 82% within the bounds of the project sites (Table 2, Figure 9b). All of these declines were significant (Table 3).

After development, the average model-predicted number of vertebrate species de- tected in one hour of surveying declined by 37% within the project area (Table 2, Figure 10a), which was significant (Table 3). The number of vertebrate species uniquely detected at a site in one phase relative to the other phase declined by 74% (Table 2, Figure 10b), which was also significant (Table 3).

Diversity 2023, 15, 1037 16 of 39
Diversity 2023, 15, x FOR PEER REVIEW 16 of 43

Figure 9. BACI tests revealed that development reduce d the number of special-status species de- tected by 49% (a) and the number of special-status species detected within the bounds of the study site by 58% (b). The red dashed arrow points to the ex pected value where no development impact occurred.

After development, the average model-predicted number of vertebrate species de- tected in one hour of surveying declined by 37% within the project area (Table 2, Figure 10a), which was significant (Table 3). The number of vertebrate species uniquely detected at a site in one phase relative to the other phase declined by 74% (Table 2, Figure 10b), which was also significant (Table 3).

Figure 10. BACI tests revealed that development redu ced the mean model-predicted number of species detected after one hour of surveying by 37% ( a), and the number of uniquely detected

0.1 0.2

0.3 0.4

0.5 0.6

0.7 0.8
log10 Number of special-status species detected
Before After
, 95% CI
Control
Impact
F1, 153 = 12.18, p = 0.0006

0.0 0.1

0.2 0.3

0.4 0.5

0.6 0.7
log10 Number of special-status species detected on site Before After
, 95% CI
Control
Impact
F1, 99 = 4.71, p = 0.0327
ab
8 10
12 14
16 18
20 22
24 Model-predicted number of vertebrate species
detected after 1 hour
Before After
, 95% CI
Control
Impact
F1, 132 = 10.32, p = 0.0017

0.5 0.6

0.7 0.8

0.9 1.0

1.1 1.2
log10 Number of vertebrate species uniquely
detected per survey
Before After
, 95% CI
Control
Impact
F1, 156 = 71.50, p = 0.0000
ab
Figure 9. BACI tests revealed that development reduced the number of special-status species detected by 49% (a) and the number of special-status species detected within the bounds of the study site by 58% (b). The red dashed arrow points to the expected value had no development impact occurred.

Diversity 2023, 15, x FOR PEER REVIEW 16 of 43

Figure 9. BACI tests revealed that development reduce d the number of special-status species de- tected by 49% (a) and the number of special-status species detected within the bounds of the study site by 58% (b). The red dashed arrow points to the ex pected value where no development impact occurred.

After development, the average model-predicted number of vertebrate species de- tected in one hour of surveying declined by 37% within the project area (Table 2, Figure 10a), which was significant (Table 3). The number of vertebrate species uniquely detected at a site in one phase relative to the other phase declined by 74% (Table 2, Figure 10b), which was also significant (Table 3).

Figure 10. BACI tests revealed that development redu ced the mean model-predicted number of species detected after one hour of surveying by 37% ( a), and the number of uniquely detected

0.1 0.2

0.3 0.4

0.5 0.6

0.7 0.8
log10 Number of special-status species detected
Before After
, 95% CI
Control
Impact
F1, 153 = 12.18, p = 0.0006

0.0 0.1

0.2 0.3

0.4 0.5

0.6 0.7
log10 Number of special-status species detected on site Before After
, 95% CI
Control
Impact
F1, 99 = 4.71, p = 0.0327
ab
8 10
12 14
16 18
20 22
24 Model-predicted number of vertebrate species
detected after 1 hour
Before After
, 95% CI
Control
Impact
F1, 132 = 10.32, p = 0.0017

0.5 0.6

0.7 0.8

0.9 1.0

1.1 1.2
log10 Number of vertebrate species uniquely
detected per survey
Before After
, 95% CI
Control
Impact
F1, 156 = 71.50, p = 0.0000
ab
Figure 10. BACI tests revealed that development reduced the mean model-predicted number of species detected after one hour of surveying by 37% (a), and the number of uniquely detected species at a site in one phase relative to the species detected in the other phase declined by 74% (b). The red dashed arrow points to the expected value had no development impact occurred.

3.2. Effects of Other Factors
We found that the ANOVA model residuals significantly increased with the increasing survey duration (Figure 11a), which should have had no effect on our BACI tests, but which would likely confound our tests for the effects of other factors. Therefore, we used Diversity 2023, 15, 1037 17 of 39
the ANOVA residuals from the BACI experiment involving model-predicted numbers of vertebrate species detected after one hour of surveying, assuming the residuals from this test would most effectively minimize any residual variation of survey duration. The model- derived residuals continued to increase with the increasing survey duration (Figure 11b), but with a much smaller r 2, a smaller standardized slope coefficient, β, and a slightly larger root-mean squared error (RMSE), all of which indicated a reduced residual effect of survey duration.

Diversity 2023, 15, x FOR PEER REVIEW 17 of 43

species at a site in one phase relative to the species detected in the other phase declined by 74% (b).

The red dashed arrow points to the expected value where no development impact occurred.

3.2. Effects of Other Factors
We found that the ANOVA model residuals significantly increased with the increas- ing survey duration (Figure 11a), which should have had no effect on our BACI tests, but which would likely confound our tests for the effects of other factors. Therefore, we used the ANOVA residuals from the BACI experiment involving model-predicted numbers of vertebrate species detected after one hour of surveying, assuming the residuals from this test would most e ffectively minimize any residual va riation of survey duration. The model-derived residuals continued to increase with the increasing survey duration (Fig- ure 11b), but with a much smaller r 2, a smaller standardized slope coe fficient, β, and a slightly larger root-mean squared error (RMSE), all of which indicated a reduced residual effect of survey duration.

Figure 11. ANOVA residuals of: (a) number of vertebrate species detected and (b) model-predicted number of vertebrate species detected at one hour regressed on survey duration (minutes).

Model-adjusted residuals related only weakly with multiple variables, including with the intensity of pre-survey actions that would have suppressed wildlife (F6,125 = 1.14, p = 0.3437), project size (ha), latitude, the number of years between the surveys in the be- fore and after phases, the site disturbance rating, and the similarity index measured be- tween the before and after surveys at the same site. However, they signi ficantly differed among groups of sites located in open space or in an in fill setting, or as redevelopment within developed areas such as cities (Figur e 12a). Mean residuals were positive within open areas, and negative in areas of infill or redevelopment. Model-adjusted residuals also significantly differed by levels of connectedness to open space (Figure 12b). Mean residu- als were positive among sites with >50% connectivity to open space, and negative among sites with <50% connectivity to open space.

Y = -0.2138 + 0.0018X
β = 0.53, r2 = 0.28, RMSE = 38.9
Compared minutes
Residuals of 2-factor ANOVA model of the
log10 number of vertebrate species detected
Y = -4.7831 + 0.0410X
β = 0.35, r2 = 0.11, RMSE = 43.4
Compared minutes
Residuals of 2-factor ANOVA model of the nonlinear
model-predicted numbers of vertebrate species
detected at 1 hour of survey
ab
40 80 120 160 200 240 280 320
-0.4
-0.3
-0.2
-0.1

0.0 0.1

0.2 0.3

0.4 40 80 120 160 200 240 280 320
-12
-8
-4
0 4
8 12
Figure 11. ANOVA residuals of: (a) number of vertebrate species detected and (b) model-predicted number of vertebrate species detected at one hour regressed on survey duration (minutes).

Model-adjusted residuals related only weakly with multiple variables, including with the intensity of pre-survey actions that would have suppressed wildlife (F 6,125 = 1.14, p = 0.3437), project size (ha), latitude, the number of years between the surveys in the before and after phases, the site disturbance rating, and the similarity index measured between the before and after surveys at the same site. However, they significantly differed among groups of sites located in open space or in an infill setting, or as redevelopment within developed areas such as cities (Figure 12a). Mean residuals were positive within open areas, and negative in areas of infill or redevelopment. Model-adjusted residuals also significantly differed by levels of connectedness to open space (Figure 12b). Mean residuals were positive among sites with >50% connectivity to open space, and negative among sites with <50% connectivity to open space.

3.3. Species Characteristics
A few species of wildlife increased in the frequency of detection among project sites that were developed, including, in order of increase: Cooper’s hawk, ruby-crowned kinglet, yellow-rumped warbler, California gull, black phoebe, house cat, and Anna’s hummingbird (Table A2). Many more species, however, decreased in the frequency of detection, including, in order of decrease: California ground squirrel, Botta’s pocket gopher, burrowing owl, California quail, California vole, Cassin’s kingbird, cedar waxwing, coyote, great-tailed grackle, killdeer, loggerhead shrike, northern rough-winged swallow, oak titmouse, orange- crowned warbler, Sierran treefrog, white-tailed kite, white-throated swift, and yellow-billed magpie, followed by: western meadowlark, red-winged blackbird, western fence lizard, great egret, American robin, eastern gray squirrel, mallard, American kestrel, red-tailed hawk, white-crowned sparrow, black-tailed jackrabbit, dark-eyed junco, western gull, Diversity 2023, 15, 1037 18 of 39
savannah sparrow, European starling, California towhee, bushtit, lesser goldfinch, Brewer’s blackbird, Canada goose, northern flicker, turkey vulture, Nuttall’s woodpecker, barn swallow, western kingbird, Swainson’s hawk, rock pigeon, mourning dove, red-shouldered hawk, double-crested cormorant, house finch, Eurasian collared-dove, California scrub-jay, cliff swallow, house sparrow, desert cottontail, northern mockingbird, common raven, American goldfinch, Say’s phoebe, and American crow (Table A2).

Groups of wildlife that declined the most following development included, in the following order: amphibians, mammals, grassland birds, raptors, special-status species, all birds as a group, non-native birds, and synanthropic birds (Table 4).

Diversity 2023, 15, x FOR PEER REVIEW 18 of 43

Figure 12. ANOVA residuals of model-predicted number of vertebrate species detected at one hour, compared by (a) whether the project site was si tuated in open space or as in fill or redevelopment within developed areas, and by (b) the percentage of the project boundary adjacent to open space.

3.3. Species Characteristics
A few species of wildlife increased in the frequency of detection among project sites that were developed, including, in order of increase: Coopers hawk, ruby-crowned king- let, yellow-rumped warbler, California gull, black phoebe, house cat, and Anna s hum- mingbird (Table A2). Many more species, however, decreased in the frequency of detec- tion, including, in order of decrease: California ground squirrel, Bo ttas pocket gopher, burrowing owl, California quail, California vole, Cassins kingbird, cedar waxwing, coy- ote, great-tailed grackle, killdeer, loggerhead shrike, northern rough-winged swallow, oak titmouse, orange-crowned warbler, Sierran tr eefrog, white-tailed kite, white-throated swift, and yellow-billed magpie, followed by: western meadowlark, red-winged black- bird, western fence lizard, great egret, Americ an robin, eastern gray squirrel, mallard, American kestrel, red-tailed hawk, white-crowned sparrow, black-tailed jackrabbit, dark- eyed junco, western gull, savannah sparrow, European starling, California towhee, bush- tit, lesser gold finch, Brewers blackbird, Canada goose, northern flicker, turkey vulture, Nuttalls woodpecker, barn swallow, western kingbird, Swainson s hawk, rock pigeon, mourning dove, red-shouldered hawk, double-crested cormorant, house finch, Eurasian collared-dove, California scrub-jay, cliff swallow, house sparrow, desert cottontail, north- ern mockingbird, common raven, American goldfinch, Says phoebe, and American crow (Table A2).

Groups of wildlife that declined the most following development included, in the following order: amphibians, mammals, grasslan d birds, raptors, special-status species, all birds as a group, non-native birds, and synanthropic birds (Table 4).

F2, 129 = 18.76, p = 0.0000
Adjacent to
open space
Infill Redevelopment
In urban/industrial
Residuals of 2-factor ANOVA model of the number
of species predicted by nonlinear models to be detected after 1 hour of survey
F5, 126 = 11.71, p = 0.0000
Percent of project boundary adjacent to open space
ab
-8
-6
-4
-2
2 4
0 1 0 0 7 55 02 51 0 0
-6
-4
-2
2 4
6 8
10 0
Figure 12. ANOVA residuals of model-predicted number of vertebrate species detected at one hour, compared by (a) whether the project site was situated in open space or as infill or redevelopment within developed areas, and by (b) the percentage of the project boundary adjacent to open space.

Table 4. Mean number of species detected per survey in each identified species’ group among the surveys in the experimental treatment groups of control-before, control-after, impact-before, and impact-after. Measures of the percentage effect of development appear in the right column.

Group
Control (n = 52) Impact (n = 26)
Effect (%)Before After Before After
Birds 23.1 25.73 17.73 10.19 −48
Mammals 2.52 2.42 1.38 0.19 −86
Reptiles 0.33 0.33 0.08 0.08 0
Amphibians 0.06 0.4 0.04 0 −100
Special-status species 5.11 4.88 3.69 1.81 −49
Non-native birds 1.98 2.37 2.69 1.81 −44
Synanthropic birds 7.19 7.67 7.42 5.73 −28
Raptors 2.77 2.61 2.58 1.15 −53
Grassland birds 1.9 1.81 1.5 0.36 −75
Diversity 2023, 15, 1037 19 of 39

4. Discussion

4.1. Effects of Development on Vertebrate Wildlife
Assuming our sampling design sufficiently controlled for differences in size, condition, and setting between control and impact sites, and for differences in survey duration, our experiment revealed substantial reductions in vertebrate species richness and numerical abundance caused by development. Although our surveys likely failed to detect all the species or to count all the animals available at the times of our surveys, we believe it is unlikely that underlying survey biases could have substantially confounded the magnitudes of development impacts we measured. We suggest, for example, that survey bias cannot explain the 74% decline we measured in the number of vertebrate species that we uniquely detected at a site in one phase relative to the species we detected in the other phase. The magnitude of this effect was too large to be explained by anything other than a profound shift in the species composition of project sites following development. Site-specific project impacts are generally devastating to wildlife.

Immediately offsite, we detected a >3-fold increase in the number of vertebrate species that were solely offsite. This increase likely reflected a spatial shift by a few species in response to development, but the numbers of species we detected solely offsite were small regardless of the treatment group. Many of the species we detected on project sites were also detected offsite, but we did not record which species were both on- and off-site until the last few surveys.

Only seven species of wildlife increased in the frequency of detection among surveys at sites where development preceded our surveys in the after phase of our experiment. Of these seven species, two were generalists—California gull and yellow-rumped warbler— consistent with the finding that generalist species of birds were most often the species that adapt to urbanized landscapes [37]. Black phoebe, house cat, and Anna’s hummingbird were three other species that increased in the frequency of occurrence, but their increases were small. Ruby-crowned kinglet’s increase remains questionable, considering the small sample sizes, but Cooper’s hawk is a specialized forager that appears to capitalize on urbanization. Otherwise, majority of wildlife species with sufficient sample sizes declined in their numbers of detection among our surveys at sites where buildings were constructed in the period between our before and after surveys. We suggest that the categorization of wildlife as urban avoiders, urban adapters, and urban exploiters [38] provides a useful framework for understanding how wildlife respond to urbanization, but we also suggest that most of the urban adapters and urban exploiters can be vulnerable to the final stage of development at a given site.

Whereas invasive and synanthropic species might fare better than native species in urban environments [39], we found that species in both these groups also declined after the development of project sites, similar to the finding of Scott [31]. The declines of species in these groups were not as great as for raptors and grassland bird species, but they were nevertheless substantial. Overall, the development projects reduced the species richness and wildlife abundance.

Terrestrial vertebrate species declined the most in our study, consistent with previous findings [ 40–42], but the declines we measured were not significant due to insufficient statistical power. Though not statistically significant, we suggest that our measured declines ought to be considered biologically significant. In the field, finding fewer or no terrestrial mammals and amphibians where we had seen them before was noticeable, and we assert that these declines resulted directly from development. Some of these terrestrial vertebrate species were ecological keystone species, such as the Botta’s pocket gopher (Figure 13) and California ground squirrel. The California ground squirrel, in particular, has been found to limit the distribution of multiple special-status species, such as the burrowing owl [43] and loggerhead shrike [44]. Indeed, where development preceded our second surveys, California ground squirrels were not observed, and neither were any of the burrowing owls or loggerhead shrikes that we had seen at those sites prior to development (Figure 13).

Diversity 2023, 15, 1037 20 of 39
Diversity 2023, 15, x FOR PEER REVIEW 20 of 43

the development of project sites, similar to the finding of Scott [31]. The declines of species in these groups were not as great as for raptors and grassland bird species, but they were nevertheless substantial. Overall, the develop ment projects reduced the species richness and wildlife abundance.

Terrestrial vertebrate species declined the most in our study, consistent with previous findings [41–43], but the declines we measured were not signi ficant due to insu fficient statistical power. Though not statistically signi ficant, we suggest that our measured de- clines ought to be considered biologically significant. In the field, finding fewer or no ter- restrial mammals and amphibians where we had seen them before was noticeable, and we assert that these declines resulted directly from development. Some of these terrestrial vertebrate species were ecological keystone species, such as the Bo ttas pocket gopher (Figure 13) and California ground squirrel. The California ground squirrel, in particular, has been found to limit the distribution of multiple special-status species, such as the bur- rowing owl [44] and loggerhead shrike [45]. Indeed, where development preceded our second surveys, California ground squirrels were not observed, and neither were any of the burrowing owls or loggerhead shrikes that we had seen at those sites prior to devel- opment (Figure 13).

Figure 13. A Bottas pocket gopher (left) peers from its burrow system on the Brokaw Campus pro- ject site in San Jose on 16 November 2018. This and any other pocket gopher stood no chance of survival following the development of the project, and no evidence of this species was seen during the second survey of 30 October 2021. Although the loggerhead shrike ( right) was detected on 23 June 2019 at the Monte Vista Warehouse project site in Vacaville, it was not detected in the follow- up survey on 16 June 2021 after the project was built.

4.2. Landscape Effects
Andrén [46] predicted that “landscapes with highly fragmented habitat, patch size and isolation will complement the e ffect of habitat loss and the loss of species or decline in population size will be greater than ex pected from habitat loss alone”. Our results tended to support this prediction. Our mean ANOVA residuals of the number of verte- brate species detected at one hour of surv eying was negative among sites in urban in fill and redevelopment settings, and positive among sites surrounded by open space (Figure 12), meaning there were relatively fewer species in urban se ttings and relatively more in settings of open space. This result resembles that of another study that found that bird Figure 13. A Botta’s pocket gopher (left) peers from its burrow system on the Brokaw Campus project site in San Jose on 16 November 2018. This and any other pocket gopher stood no chance of survival following the development of the project, and no evidence of this species was seen during the second survey of 30 October 2021. Although the loggerhead shrike (right) was detected on 23 June 2019 at the Monte Vista Warehouse project site in Vacaville, it was not detected in the follow-up survey on 16 June 2021 after the project was built.

4.2. Landscape Effects
Andrén [45] predicted that “landscapes with highly fragmented habitat, patch size and isolation will complement the effect of habitat loss and the loss of species or decline in population size will be greater than expected from habitat loss alone”. Our results tended to support this prediction. Our mean ANOVA residuals of the number of vertebrate species detected at one hour of surveying was negative among sites in urban infill and redevelopment settings, and positive among sites surrounded by open space (Figure 12), meaning there were relatively fewer species in urban settings and relatively more in settings of open space. This result resembles that of another study that found that bird species richness in urban settings correlated positively with bird species richness in adjacent landscapes composed of managed or natural vegetation [46].

We note, also, that we detected more species composed of smaller average counts of individuals in the before phase of control sites, as compared to the before phase of impact sites—the sites that were to be developed later during our study; alternatively, we found smaller numbers of species of larger average counts at impact sites even in the before phase, which was a pattern previously noted [31]. By the time we initiated our first surveys at the impact sites, they were already different in species composition. In fact, the impact sites differed from control sites with their average smaller size and lower elevation, but perhaps more importantly, with their lower connectivity to open space, their higher average rank on the urban setting index, their higher average ratings for the level of disturbance, and their average higher intensity of actions resulting in suppression of wildlife occurrences. We also surveyed impact sites more briefly than we surveyed control sites, but our briefer surveys probably reflected the smaller average size of impact sites. Earlier in our study, we could not have predicted which sites would be developed sooner than other sites, but now it appears that smaller infill sites tend to be managed more aggressively to suppress wildlife, tend to support fewer species, and are more likely to be approved for development.

Diversity 2023, 15, 1037 21 of 39
Numerous species of vertebrate wildlife were found only at control sites, further revealing the potential wildlife community difference that already existed by the time of our first surveys at project sites, but which also prevented species-specific measures of development effects. Such species included Allen’s hummingbird and the black-chinned hummingbird, American coot, black-necked stilt, blue-gray gnatcatcher and California gnatcatcher, bobcat, California thrasher, common yellowthroat, great horned owl, hairy woodpecker, hooded oriole, horned lark, marsh wren, mule deer, olive-sided flycatcher, ring-billed gull, song sparrow, striped skunk, western bluebird, white-breasted nuthatch, and Wilson’s warbler. A mitigating factor in our findings of these species only at con- trol sites was the fact that we surveyed twice the number of control sites compared to impact sites.

4.3. Evidence of Ongoing Cumulative Effects of Urbanization Before- and after-phase surveys at control sites revealed a trend that was likely more indicative of cumulative effects at landscape scales, as these surveys were of equal number and free of onsite development. Despite our average detections of about 2–3 additional species after our second surveys at control sites, and despite our average counts of 26% more birds in the absence of site-specific development, in our follow-up surveys we detected 56% fewer special-status species, and we counted 44% fewer animals of special-status species.

During our study, special-status species of vertebrate wildlife appeared to have been on the decline within the regions of our study. These declines indicate that project-specific mitigation measures have been failing to avoid cumulative impacts.

4.4. Potential Biases
We endeavored to design and implement our study to minimize the effects of bias and error by standardizing site-specific survey dates, start times, survey duration, and survey methods. Where we walked a transect along the perimeter of a site during the first survey, we tried to repeat the walk of the same transect during the second survey. We surveyed most sites the second time with the same investigator, or both of us, as we had surveyed the first time. However, there was variation in the survey methods between sites, most notably the survey duration. We attempted to adjust our survey findings for variation in the survey duration by best-fitting a nonlinear model to each survey’s increase in the cumulative number of detected species with the increasing time into the survey. From each model, we predicted the number of species detected after one hour of surveying, which was enough time to predict a substantial number of species but also well within the range of survey durations that we completed among the project sites. Nevertheless, the ANOVA model’s residuals derived from model-predicted numbers of species detected after 1 h continued to weakly increase with the increasing survey duration among project sites (Figure 11b). We did not eliminate the effects of survey duration. We do not believe that the remaining effect of survey duration significantly affected our study results, but we note this effect for designing future studies of the effects of urbanization on wildlife. The effect was stronger without the model-based adjustments, but the model-based adjustments appear to have been sample-dependent. The duration of the survey affects the pattern of the increasing cumulative number of species with the increasing survey duration, and hence affects the nonlinear model fit to the pattern. It might be possible to standardize the pattern in the cumulative number of species detected by standardizing Watson’s [33] results-based stopping rule, or by standardizing the survey duration [47]. If the latter, then we recommend a relatively long survey duration, such as ≥2 h.

Another potential bias is the change in detection rates of wildlife species after the project sites were developed. Our probability of detecting the average bird was likely higher after the available perches transitioned from trees and shrubs to light standards on parking lots and the rooflines of warehouses. The detection likelihood might have increased after opportunities to view flying birds transitioned from views of complex environmental backgrounds to the walls of warehouses, although the environmental backgrounds of most Diversity 2023, 15, 1037 22 of 39
of our project sites were rather simple. At some sites, the landscaping around warehouses and other new structures comprising the project might have been simpler than the pre- construction environments, thus better enabling us to detect an animal on those portions of the project site had the animals been present. Although we acknowledge this potential bias, the amount of survey time we committed to each site gave us ample opportunity to detect the species of wildlife that were truly there at the times of both of our surveys.

At developed sites, our rates of detection of wildlife species were much slower (Figure 8) and the total numbers of species detected were fewer (Tables 2 and 3), but we believe that these differences more likely reflected biological conditions than they did our survey detection rates.

Another potential bias was the differential detection rates among species of wildlife.

We likely disproportionately detected the most readily detectable species, while failing to detect those species that are smallest in body size, nocturnal, fossorial, and more cryp- tic. Furthermore, the numerical abundances we attributed to the species we detected likely often differed from the true numerical abundances, even within the spatial and temporal scopes of our surveys. Whereas our counts of animals might have more of- ten approached the true numbers of the largest-bodied species, such as red-tailed hawks (Buteo jamaicensis) and mule deer (Odocoileus hemionus), they likely under-represented the smaller-bodied species, such as sparrows, warblers, western fence lizards, and Belding’s orange-throated whiptails.

Our baseline settings in the before-phase surveys were far from pristine at most project sites, but they nevertheless served as baselines for measuring changes brought about by construction of buildings to the number of vertebrate species detected and to our counts of live animals, which we intended to indicate, respectively, as species richness and relative abundance. We use the term “indicate” here because we recognize that we did not truly measure species richness nor true abundance, as multiple potential biases and errors prevented such measurement [48]. On the other hand, our surveys were of sufficient duration to detect most of the diurnal bird species that would have been available to us at each site at the times of our surveys [47].

5. Conclusions
By use of an experiment including control sites, we found that development projects directly and substantially reduced vertebrate wildlife species richness and wildlife abun- dance. Vertebrate wildlife species most affected by development in California were terres- trial species, as well as grassland birds, raptors, and special-status species. We also found that special-status species declined on control sites even in the absence of site-specific devel- opment, which indicates widespread ineffectiveness of project-level mitigation measures, and hence cumulative impacts from regionwide urbanization. More needs to be learned as soon as possible about the impacts of urbanization on wildlife. Experiments need to be designed with the use of control sites to more directly measure project-level effects, and with sampling plots to monitor regional effects of urbanization.

To follow-up on Marzluff’s [18] recommendation, and to take advantage of the op- portunities to measure the effects of development projects on wildlife, the California Environmental Quality Act should be amended to require that reconnaissance surveys be repeated in each season of the year preceding the public circulation of an environmental re- view document. The required mitigation plan should include funding for post-construction reconnaissance surveys of the same methods, number, and seasonal spacing to more ro- bustly represent the wildlife community before and after development. The CEQA should be further amended to require a sufficient funding allocation from each project applicant that would be directed to control sites, which should also be integrated into cumulative effects monitoring. Whereas the CEQA requires a cumulative impacts analysis, data needed to analyze the cumulative impacts to wildlife usually do not exist, and thus consultants’ analyses of cumulative impacts are speculative. Since long-term monitoring is often not re- quired, and thus not performed, the consultants’ conclusions about the cumulative impacts Diversity 2023, 15, 1037 23 of 39
cannot be confirmed nor denied. Long-term monitoring would give all parties involved a better understanding of how to analyze the cumulative impacts, because we would have a better understanding of how development truly affects each species. A cumulative impacts fund should be administered by a trusted party to ensure that unbiased, qualified biologists implement long-term monitoring of wildlife within a spatial area that can meaningfully inform of cumulative effects.

To soften the impacts of urbanization on wildlife, the CEQA should be amended to require the use of native and xeric-adapted plants in landscaping, i.e., chaparral, grassland, and locally appropriate scrub plants, as opposed to landscaping with lawn and exotic shrubs. Native plants offer more structure, cover, food resources, and breeding substrates for wildlife than landscaping with lawn [49,50] and increase the abundance and diversity of birds, especially native birds [ 51–54]. Landscaping with native plants is a way to interconnect patches of habitat for wildlife [55,56].

The CEQA should also be amended to require project applicants to contribute funding to wildlife rehabilitation facilities. As projects are built, and wildlife are subsequently injured by the windows of buildings, project-generated traffic, and free-ranging house cats of new residents, wildlife rehabilitation facilities should be provided the resources they need to attempt to rectify these types of project impacts.

Author Contributions: Conceptualization, K.S.S.; methodology, K.S.S. and N.L.S.; formal analysis, K.S.S.; investigation, K.S.S. and N.L.S.; writing—original draft, K.S.S. and N.L.S.; writing—review and editing, N.L.S.; supervision, K.S.S. All authors have read and agreed to the published version of the manuscript.

Funding: The research specific to this study received no external funding, but all the surveys in the before phase of the experiment were funded by the authors’ clients who had interests in the outcomes of the proposed projects. All but one of the surveys in the after phase were self-funded.

Data Availability Statement: Data supporting the reported results can be requested from the authors.

Acknowledgments: We thank our clients for funding the initial survey at each site of the proposed development projects. We also thank the two anonymous peer reviewers for their very helpful reviews of earlier drafts of this manuscript.

Conflicts of Interest: We declare no conflict of interest.

Diversity 2023, 15, 1037 24 of 39
Appendix A
Table A1. Project site and survey attributes.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
1 Control Before 64 11th Street Development Upland Warehouse 8 November 2020 6:40 1.98 Ruderal scrub around old cement pad 1 Control After 64 11th Street Development Upland Warehouse 24 November 2021 6:43 1.98 Same as above 2 Control Before 135 4150 Point Eden Way
Industrial Development Hayward Warehouse 11 May 2021 6:40 4.37 Grassland bounded by salt ponds,
including those of Eden Landing Reserve,
CA Highway 92, and
industrial warehouses
2 Control After 135 4150 Point Eden Way
Industrial Development Hayward Warehouse 10 May 2022 7:12 4.37 Same as above 3 Control Before 135 Airport Business Centre Manteca Warehouse 28 April 2021 16:17 9.51 Mowed hay bordered on the north by warehouses
3 Control After 135 Airport Business Centre Manteca Warehouse 28 March 2022 16:31 9.51 Unmowed hay bordered on the north and west by warehouses
4 Impact Before 50 Almond Street Warehouse Fontana Warehouse 27 April 2019 9:25 4.05 Former parking lot with ornamental trees 4 Impact After 50 Almond Street Warehouse Fontana Warehouse 25 April 2022 8:50 4.05 Warehouse 5 Control Before 105 Alta Cuvee Rancho
Cucamonga Residential 4 September 2021 6:54 2.55 Highly disturbed dirt field with low shrubs and non-native grass
5 Control After 105 Alta Cuvee Rancho
Cucamonga Residential 30 August 2022 7:04 2.55 Same as above 6 Control Before 163 Amare Apartments Martinez Residential 4 June 2018 17:17 2.45 Disked woodland savannah 6 Control After 163 Amare Apartments Martinez Residential 19 July 2021 17:07 2.45 Same as above 7 Control Before 130 Antonio Mountain Ranch Placer County Residential 18 November 2002 14:30 359.00 Grassland/vernal pool complex with riparian
7 Control After 130 Antonio Mountain Ranch Placer County Residential 16 November 2021 14:45 359.00 Same as above 8 Impact Before 135 Brokaw Campus San Jose Corporate
Campus 16 November 2018 12:45 6.78 Disked with ruderal cover 8 Impact After 135 Brokaw Campus San Jose Corporate
Campus 30 October 2021 12:42 6.78 Four tall buildings
9 Control Before 80 Casmalia and Linden Rialto Warehouse 21 June 2020 6:10 2.77 Grassland and shrubs 9 a Control After 80 Casmalia and Linden Rialto Warehouse 31 July 2021 6:12 2.77 Grassland and shrubs 10 Impact Before 94 CenterPoint Manteca Warehouse 31 October 2018 16:15 9.12 Ruderal vegetation subsequent to grading
10 Impact After 94 CenterPoint Manteca Warehouse 11 November 2021 15:26 9.12 Warehouse with parking lot 11 Impact Before 100 Cessna and
Aviator Warehouse Vacaville Warehouse 12 August 2018 18:00 5.21 Disked annual grassland 11 Impact After 100 Cessna and
Aviator Warehouse Vacaville Warehouse 31 August 2022 18:07 5.21 Warehouse and parking lot Diversity 2023, 15, 1037 25 of 39
Table A1. Cont.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
12 Control Before 95 Cordelia Industrial Cordelia Warehouse 16 October 2019 15:54 13.11 Disked annual grassland with some regrowth next to riparian
12 Control After 95 Cordelia Industrial Cordelia Warehouse 7 October 2021 12:38 13.11 Disked annual grassland next to riparian; new houses on west side
13 Control Before 165 Davidon Homes Petaluma Residential 11 February 2021 7:41 23.74 Grassland and riparian oak woodland 13 Control After 165 Davidon Homes Petaluma Residential 1 March 2022 7:33 23.74 Same as above 14 Control Before 54 Aggie Research Campus Davis Residential 13 April 2020 18:39 74.90 Planted sugarbeets, wheat, almonds 14 Control After 54 Aggie Research Campus Davis Residential 2 April 2022 18:33 74.90 Wheat, dirt furrows, planted sugarbeets 15 Control Before 93 Del Rey Pointe
Residential Project Playa Del Rey Residential 31 October 2019 14:07 1.16 Ruderal vegetation bordered by Eucalyptus and 3 concrete-lined streams
15 Control After 93 Del Rey Pointe
Residential Project Playa Del Rey Residential 18 October 2021 13:54 1.16 Ruderal vegetation undergoing clearing
by tractor; bordered by Eucalyptus and
3 concrete-lined streams
16 Impact Before 122 GLP Store Mather Warehouse 12 February 2020 6:56 3.76 Annual grassland 16 Impact After 122 GLP Store Mather Warehouse 18 February 2022 7:09 3.76 Warehouse 17 Impact Before 90 Green Valley II Fairfield Residential 18 November 2019 9:00 5.39 Disked grassland with 1 oak and bordered by shrubs
17 Impact After 90 Green Valley II Fairfield Residential 7 December 2021 9:47 5.39 Nearly built warehouse and apartments 18 Control Before 120 Hillcrest LRDP Bachman
Canyon None 9 November 2019 7:00 12.95 Diegan coastal sage scrub 18 Control After 120 Hillcrest LRDP Bachman
Canyon None 11 December 2021 7:11 12.95 Same as above
19 Control Before 90 IKEA Outlet Dublin Warehouse retail 26 March 2018 11:15 11.11 Ruderal and annual grassland 19 Control After 90 IKEA Outlet Dublin Warehouse retail 25 March 2022 9:53 11.11 Ruderal and annual grassland; tractors
and trucks onsite, and about 15% to 20%
is graded
20 Control Before 110 Jersey Industrial Complex Rancho
Cucamonga Warehouse 16 June 2021 6:26 2.99
Ruderal vegetation on disturbed soil,
surrounded by warehouses and major
roads and railroad tracks
20 Control After 110 Jersey Industrial Complex Rancho
Cucamonga Warehouse 11 July 2022 6:30 2.99
Previously disked, non-native grass
present, surrounded by warehouses and
major roads and railroad tracks
21 Control Before 120 Johnson Drive
Economic Zone Pleasanton Warehouse
retail, hotel 29 July 2019 17:38 16.19 Mix of developed structures, vacant lots, and grassland
21 Control After 120 Johnson Drive
Economic Zone Pleasanton Warehouse
retail, hotel 26 July 2021 17:46 16.19 Same as above
22 Control Before 195 Kassis Rancho Cordova Residential 3 December 2020 7:47 16.51 Disked grassland and abandoned walnuts
22 Control After 195 Kassis Rancho Cordova Residential 2 November 2021 7:39 16.51 Same as above 24 Control Before 70 Lake Home Lodi Residential 13 March 2019 8:28 3.56 Abandoned orchard Diversity 2023, 15, 1037 26 of 39
Table A1. Cont.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
24 Control After 70 Lake Home Lodi Residential 25 March 2022 7:26 3.56 Same as above 25 Impact Before 92 LDK Warehouse Vacaville Warehouse 10 November 2018 7:50 27.88 Disked annual grassland, riparian 25 Impact After 92 LDK Warehouse Vacaville Warehouse 13 November 2021 7:53 27.88 Operational warehouse and nearly completed empty warehouse
26 Control Before 127 Legacy Highlands Beaumont,
upper Residential 4 May 2021 17:39 647.50 Sage scrub
26 Control After 127 Legacy Highlands Beaumont,
upper Residential 24 April 2022 17:34 647.50 Sage scrub 27 Control Before 189 Legacy Highlands Beaumont,
lower Residential 5 May 2021 6:02 647.50 Riparian, grassland, sage scrub 27 Control After 189 Legacy Highlands Beaumont,
lower Residential 26 April 2022 6:02 647.50 Riparian, grassland, sage scrub 28 Impact Before 90 Logisticenter at Vacaville Vacaville Warehouse 1 September 2018 7:50 5.68 Disked annual grassland 28 Impact After 90 Logisticenter at Vacaville Vacaville Warehouse 5 September 2021 7:48 5.68 Warehouse surrounded by warehouses on 3 sides, disked on 4th side
29 Control Before 100 Mango Avenue Fontana Warehouse 24 January 2021 7:33 2.35 Ruderal grassland 29 Control After 100 Mango Avenue Fontana Warehouse 13 February 2022 7:25 2.35 Ruderal grassland 30 Control Before 112 Vista Mar Pacifica Residential 20 August 2020 6:59 0.53 Trees, shrubs, grassland 30 Control After 112 Vista Mar Pacifica Residential 15 September 2021 7:17 0.53 Trees, shrubs, grassland 31 Control Before 75 Marriott Hotel
Harbor Bay
Parkway,
Alameda
Hotel 16 November 2018 15:18 2.23 Ruderal cover on disked field lined by trees
31 Control After 75 Marriott Hotel
Harbor Bay
Parkway,
Alameda
Hotel 30 October 2021 14:54 2.23 Same as above
32 Control Before 188 Mather South Masterplan Mather Residential 16 February 2019 8:02 343.17 Annual grassland, wetland, riparian 32 Control After 188 Mather South Masterplan Mather Residential 6 February 2022 7:20 343.17 Same as above 33 Impact Before 145 Monte Vista Warehouse Vacaville Warehouse 23 June 2019 6:48 4.67 Disked grassland with volunteer shrubs/trees
33 Impact After 145 Monte Vista Warehouse Vacaville Warehouse 16 June 2021 6:48 4.67 Nearly completely constructed
warehouse; field to west under
construction wtih pad
34 Impact Before 70 Morton Salt Plant Newark Warehouse 8 May 2018 16:58 12.10 Abandoned salt ponds 34 Impact After 70 Morton Salt Plant Newark Warehouse 3 June 2021 16:20 12.10 Warehouses next to row of Eucalyptus 35 Impact Before 75 Natomas Crossing Natomas Commercial 30 June 2018 19:00 27.60 Feral grassland on disked soil 35 Impact After 75 Natomas Crossing Natomas Commercial 9 June 2021 18:30 27.60 New buildings and parking lots; field to east was disked
36 Control Before 90 Nova Business Park Napa Warehouse 15 July 2018 18:50 9.39 Annual grassland and riparian forest Diversity 2023, 15, 1037 27 of 39
Table A1. Cont.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
36 Control After 90 Nova Business Park Napa Warehouse 14 July 2021 18:19 9.39 Annual grassland and riparian forest, but
early grading for project over past month
or two and lots of development in
surrounding area
37 Impact Before 70 Oakley Logistics Center Oakley Warehouse 22 November 2019 8:04 152.04 Marsh, grassland, riparian, disturbed 37 Impact After 70 Oakley Logistics Center Oakley Warehouse 7 December 2021 7:55 152.04 Warehouses and parking lots 38 Control Before 90 Olympic Holding
Inland Center San Bernardino Warehouse 1 December 2019 8:18 2.12 Barren ground and ruderal vegetation lined by trees
38 b Control After 90 Olympic Holding
Inland Center San Bernardino Warehouse 6 December 2021 8:24 2.12 Barren ground and ruderal vegetation lined by trees
39 Control Before 75 PARS Global Storage Murietta Warehouse 31 October 2019 10:06 1.28 Shrubs, grass, trees 39 Control After 75 PARS Global Storage Murietta Warehouse 19 October 2021 10:02 1.28 Shrubs, grass, trees 40 Control Before 177 Regional University Roseville University 15 January 2008 a 16:10 468.42 Annual grasslands, vernal pools 40 c Control After 165 Regional University Roseville University 12 January 2022 a 15:13 468.42 Annual grasslands, vernal pools 41 Impact Before 150 Rider Warehouse Perris Warehouse 22 July 2019 6:55 3.89 Disked annual grassland 41 d Impact After 150 Rider Warehouse Perris Warehouse 28 March 2021 6:50 3.89 Warehouse 42 Control Before 138 Clover Project Petaluma Residential 13 July 2020 17:24 1.36 Grassland with a few mature trees, next to Petaluma River
42 Control After 138 Clover Project Petaluma Residential 22 July 2021 17:30 1.36 Grassland with a few mature trees, next to Petaluma River
43 Control Before 135 Ruby Street Castro Valley Multi-family housing 17 October 2020 7:23 2.55
Grass meadow next to riparian forest of
San Lorenzo Creek and otherwise
surrounded by residential
43 Control After 135 Ruby Street Castro Valley Multi-family housing 10 September 2021 7:40 2.55
Grass meadow next to riparian forest of
San Lorenzo Creek and otherwise
surrounded by residential
44 Control Before 295 San Pedro Mountain Pacifica Residential 3 June 2021 6:00 9.45 Eucalyptus/Monterey Pine forest and Coyote bush scrub
44 Control After 295 San Pedro Mountain Pacifica Residential 6 June 2022 6:36 9.45 Eucalyptus/Monterey Pine forest and Coyote bush scrub
45 Control Before 60 Santa Maria Airport
Business Park Santa Maria Office complex 9 April 2021 7:14 11.33 Strawberries with Eucalyptus woodland border
45 Control After 60 Santa Maria Airport
Business Park Santa Maria Office complex 11 May 2022 8:03 11.33 Strawberries with Eucalyptus woodland border
46 Impact Before 60 Scheu Warehouse Rancho
Cucamongo Warehouse 31 October 2019 8:02 5.35 Mowed grassland 46 Impact After 60 Scheu Warehouse Rancho
Cucamonga Warehouse 19 October 2021 8:07 5.35 Warehouse with dirt mound on west side 47 Impact Before 61 Seefried Warehouse Lathrop Warehouse 20 November 2019 8:23 4.65 Disked grassland 47 Impact After 61 Seefried Warehouse Lathrop Warehouse 17 November 2021 9:44 4.65 Warehouse and parking lot Diversity 2023, 15, 1037 28 of 39
Table A1. Cont.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
48 Impact Before 100 Shoe Palace Morgan Hill Warehouse 16 November 2018 9:45 15.40 Annual grassland 48 Impact After 100 Shoe Palace Morgan Hill Warehouse 30 October 2021 10:15 15.40 Warehouse and parking lot with strip of planted shrubs
49 Impact Before 60 South Hayward Hayward Residential 14 April 2018 17:00 10.42 Annual grass either side of water channel 49 Impact After 60 South Hayward Hayward Residential 3 June 2021 18:06 10.42 Residential development either side of water channel
50 Control Before 165 Sun Lakes Village North Banning Warehouse 9 November 2020 7:15 19.03 Annual grassland with willow patch and buckwheat scrub
50 Control After 165 Sun Lakes Village North Banning Warehouse 23 November 2021 6:55 19.03 Annual grassland with willow patch and buckwheat scrub
51 Impact Before 90 The Promenade Carmichael Commercial 1 October 2002 9:25 4.13 Woodland savannah 51 Impact After 90 The Promenade Carmichael Commercial 13 October 2021 9:22 4.13 Commercial strip/parking lots 52 Control Before 210 UCSF Parnassus Campus
and Sutro Park San Francisco University
expansion 20 August 2020 8:17 67.99 Campus; forested
52 Control After 210 UCSF Parnassus Campus
and Sutro Park San Francisco University
expansion 16 July 2021 9:14 67.99 Campus; forested
53 Control Before 110 Veterans Affairs Clinic Bakersfield VA Clinic 20 January 2021 8:00 4.07 Recently burned annual grassland 53 Control After 110 Veterans Affairs Clinic Bakersfield VA Clinic 11 January 2022 7:21 4.07 Annual grassland 55 Impact Before 150 Winter’s Highlands and
Callahan Estates Winters Residential 18 May 2004 9:20 60.70 Annual grassland 55 Impact After 150 Winter’s Highlands and
Callahan Estates Winters Residential 11 June 2021 9:04 60.70 Residential 56 Control Before 75 Hayward Regional
Shoreline Hayward None 31 January 2018 14:45 734.50 Coastal scrub and wetlands 56 Control After 75 Hayward Regional
Shoreline Hayward None 23 January 2022 14:50 734.50 Coastal scrub and wetlands 57 Impact Before 75 Winton Ave
Industria Project Hayward Warehouse 31 January 2018 14:45 9.47 Vacant lot with old concrete pads surrounded by ruderal vegetation
57 Impact After 75 Winton Ave
Industrial Project Hayward Warehouse 23 January 2022 16:07 9.47 Warehouse 59 Control Before 107 Woodland Research Park South of
Woodland Residential 30 June 2021 18:04 156.61
Agricultural field crops and
woodland/savannah between residential
and Highway 113
59 Control After 107 Woodland Research Park South of
Woodland Residential 13 July 2022 17:56 156.61
Agricultural field crops and
woodland/savannah between residential
and Highway 113
60 Control Before 155 Yuba Highlands Spenceville
WMRA Residential 12 November 2006 13:00 1174.40 Annual grassland, oak woodland, riparian; east of proposed project
60 Control After 155 Yuba Highlands Spenceville
WMRA Residential 20 November 2021 13:55 1174.40 Annual grassland, oak woodland, riparian
Diversity 2023, 15, 1037 29 of 39
Table A1. Cont.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
61 Impact Before 60 Zeiss Innovation Center Dublin Office commercial 8 February 2018 10:50 4.60 Annual grassland
61 Impact After 60 Zeiss Innovation Center Dublin Office commercial 3 February 2021 10:38 4.60 Mid-rise buildings and parking lots nearly completed
62 Control Before 133 Fairway Business Park Lake Elsinore Warehouse 1 December 2021 7:12 3.56 Non-native grassland and ruderal shrubs 62 Control After 133 Fairway Business Park Lake Elsinore Warehouse 8 December 2022 7:15 3.56 Annual grass and shrubs, and mule fat and salt cedar
63 Impact Before 51 First Industrial Logistics
Center II Moreno Valley Warehouse 28 February 2020 13:25 3.93 Ruderal grassland with piles of dirt and debris from neighboring development
63 Impact After 51 First Industrial Logistics
Center II Moreno Valley Warehouse 5 February 2023 13:25 3.93 Warehouse landscaped with low shrubs and ornamental trees
64 Control Before 118 Hagemon Bakersfield Warehouse 9 January 2022 15:20 31.95 Annual grassland that had been disked within last few years
64 Control After 118 Hagemon Bakersfield Warehouse 4 February 2023 15:32 31.95 Annual grassland that had been disked again recently
65 Control Before 98 Operon HKI Perris Warehouse 21 November 2021 7:03 3.52 Mowed grassland surrounded by warehouses
65 Control After 98 Operon HKI Perris Warehouse 9 December 2022 7:22 3.52 Annual grass and prickly Russian thistle surrounded by warehouses
66 Control Before 162 Brawley Solar
Energy Facility Brawley Utility-scale
solar 4 February 2022 6:52 91.86 Alfalfa, ruderal, Atriplex, Tamarisk, and Sueda along railroad tracks
66 e Control After 162 Brawley Solar
Energy Facility Brawley Utility-scale
solar 22 February 2023 6:22 91.86 Alfalfa, ruderal, Atriplex, Tamarisk, and Sueda along railroad tracks
67 Control Before 210 Rio Del Oro Rancho Cordova Residential 25 May 2008 b 19:30 1549.14 Annual grassland, wetland, oak woodland
67 f Control After 208 Rio Del Oro Rancho Cordova Residential 13 June 2021 b 18:58 1549.14 Same as above, but bordered by new
grading to the west and houses to the
east and south
68 Impact Before 93 San Bernardino
Logistics Center San Bernardino Warehouse 25 January 2018 8:00 8.22 Feral grassland on disked soil 68 Impact After 93 San Bernardino
Logistics Center San Bernardino Warehouse 8 February 2023 7:12 8.22 Warehouse 70 Control Before 95 The Ranch at Eastvale Eastvale Warehouse 9 January 2020 9:22 7.08 Disked grassland bordered by irrigated, planted native plants, shrubs, trees
70 Control After 95 The Ranch at Eastvale Eastvale Warehouse 6 February 2023 8:58 7.08 Warehouses with landscaped shrubs and trees
71 Control Before 156 Alviso Hotel Alviso Hotel 1 April 2022 6:44 2.52 Wetland and ruderal vegetation 71 Control After 156 Alviso Hotel Alviso Hotel 6 April 2023 7:12 2.52 Wetland and ruderal vegetation Diversity 2023, 15, 1037 30 of 39
Table A1. Cont.

Pair Treatment
Level Phase
Survey
Minutes
Compared
Project Location Proposed Use Survey Date Start
Time Ha Conditions on the Ground
72 Control Before 146 Conejo Summit Thousand Oaks
Biotech
industrial
buildings
7 March 2022 15:18 20.17
Annual grassland, sage scrub dominated
by California buckwheat, California
sagebrush, coyote brush, deerweed
72 Control After 146 Conejo Summit Thousand Oaks
Biotech
industrial
buildings
2 April 2023 15:18 20.17
Annual grassland, sage scrub dominated
by California buckwheat, California
sagebrush, coyote brush, deerweed
73 Control Before 147 Gillespie Field El Cajon Warehouse 13 March 2021 6:48 12.83 Annual grassland, San Diegan sage scrub 73 Control After 147 Gillespie Field El Cajon Warehouse 29 March 2023 6:56 12.83 Annual grassland, San Diegan sage scrub 75 Control Before 130 Greentree Vacaville Residential 25 May 2022 5:31 76.65 Disked abandoned golf course with dead and living trees and dried wetlands
75 Control After 130 Greentree Vacaville Residential 18 May 2023 5:37 76.65 Freshly disked abandoned golf course with more dead trees, some removed
76 Control Before 60 Amazing 34 San Bernardino Warehouse 25 April 2022 10:18 1.55 Demolished buildings and annual
grassland and ornamental trees
around pads
76 Control After 60 Amazing 34 San Bernardino Warehouse 22 May 2023 10:08 1.55 Demolished buildings and annual
grassland and ornamental trees
around pads
77 Control Before 139 Haun and Holland Menifee Warehouse 6 June 2020 6:06 15.00 Annual grassland and ruderal vegetation 77 Control After 139 Haun and Holland Menifee Warehouse 22 May 2023 6:13 15.00 Annual grassland and ruderal vegetation 78 Impact Before 120 Hillcrest LRDP UCSD, Hillcroft
Campus
Campus
redevelopment 9 November 2019 7:00 12.95 Riparian, Eucalyptus 78 Impact After 120 Hillcrest LRDP UCSD Hillcroft
Campus
Campus
redevelopment 11 December 2021 7:11 12.95 New buildings and construction underway on campus
79 Control Before 195 Diamond Street
Warehouse San Marcos Warehouse 25 June 2021 5:59 9.31 Coastal sage scrub with grown-over disturbed area in central aspect
79 Control After 195 Diamond Street
Warehouse San Marcos Warehouse 14 June 2023 6:35 9.31 Coastal sage scrub with grown-over disturbed area in central aspect
80 Impact Before 80 Casmalia and Linden Rialto Warehouse 31 July 2021 6:12 2.77 Grassland and shrubs 80 Impact After 80 Casmalia and Linden Rialto Warehouse 8 July 2023 6:12 2.77 Warehouses with landscaping 81 Control Before 135 Fore Apartments Oxnard Residential 26 June 2022 6:40 1.71 Mowed annual grassland with a few peripheral trees
81 Control After 135 Fore Apartments Oxnard Residential 15 July 2023 6:54 1.71 Ruderal grassland around graded plots 82 Impact 148 Scannell Properties Richmond Warehouses 13 July 2021 17:50 11.90 Ruderal grassland around graded plots 82 Impact 148 Scannell Properties Richmond Warehouses 16 July 2023 17:45 11.90 Operational warehouse and nearly completed empty warehouse
a The site was used twice in the experiment, once as a control site and then as an impact site. N.L.S. surveyed this site twice before it was developed into a warehouse, so the first two surveys represented the control treatment. She completed a third survey after the site was developed, so based on the second and third surveys we also treated the site as an impact treatment. b Whereas both of us surveyed the site in 2019, N.L.S. surveyed it alone in 2021. c The survey in the before phase consisted of two surveys separated by 6 days. Since the species detected were lumped between the two surveys, we could not single out the first survey date for comparison. Therefore, we treated the survey in the after phase the same way by Diversity 2023, 15, 1037 31 of 39
completing a second survey 7 days after the first survey. Shown in the Appendix are only the dates and start times of the first survey in both phases. The second surveys in both phases were completed 6 and 7 days later, respectively, on 21 January 2008 and 12 January 2021, and the combined survey duration between the before and after phases differed by only 12 min. d We did not quantify our defined metrics from the unconstrained viewshed, because our survey extended too far beyond the Rider project footprint and, therefore, included too many animals that were less likely to have been directed affected by the project. We did, however, quantify our metrics for the onsite comparisons, because we had carefully noted which species were onsite. Additionally, one of us (N.L.S.) surveyed the site alone in the after phase, whereas both of us surveyed the site in the before phase. The season of the second survey did not match the season of the first (March instead of July), because the second survey was in response to a client request to survey the adjacent property and had to be completed in March 2021. e Whereas K.S.S. surveyed the site alone in 2022, both of us surveyed it in 2023. f The survey in the before phase consisted of two surveys separated by 3 days. Since the species detected were lumped between the two surveys, we could not single out the first survey date for comparison. Therefore, we treated the survey in the after phase similarly by completing a second survey 20 days after the first survey. Shown in the Appendix are only the dates and start times of the first survey in both phases. The second surveys in both phases were completed 3 and 20 days later, respectively, on 21 January 2008 and 3 July 2021, and the combined survey duration between the before and after phases differed by only 2 min.

Table A2. Frequency of detection of each species among the surveys in the experimental treatment groups of control-before, control-after, impact-before, and impact-after. Measures of the effect of development appear in the right column for those species with sufficient sample sizes or special status.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
Abert’s towhee Melozone aberti 1 1 0 0
Acorn woodpecker Melanerpes formicivorus 6 12 0 0
Alameda song sparrow Melospiza melodia pusillula BCC, SSC2 1 0 0 0 Allen’s hummingbird Selasphorus sasin BCC 9 10 0 0
American avocet Recurvirostra americanus BCC * 0 1 0 0
American beaver Castor canadensis 0 1 0 0
American bittern Botaurus lentiginosus 1 0 0 0
American coot Fulica americana 3 5 0 0
American crow Corvus brachyrhynchos S 41 38 24 21 −6
American goldfinch Spinus tristis 18 10 8 4 −10
American kestrel Falco sparverius R BOP 20 22 14 5 −68
American pipit Anthus rubescens G 4 9 1 0
American robin Turdus migratorius 15 12 5 1 −75
American wigeon Mareca americana 1 3 0 0
Anna’s hummingbird Calypte anna S 35 40 12 15 9
Ash-throated flycatcher Myiarchus cinerascens 5 5 0 1
Baja California treefrog Pseudacris hypochondriaca 1 2 0 0 Bald eagle Haliaeetus leucocephalus R CE, BGEPA, CFP 1 1 0 1 Band-tailed pigeon Patagioenas fasciata 1 1 0 0
Barn swallow Hirundo rustica 9 13 5 4 −45
Bat sp. 1 0 0 0
Belted kingfisher Ceryle alcyon 2 2 0 0
Bewick’s wren Thryomanes bewickii 10 7 1 0
Black-chinned hummingbird Sayornis nigricans 3 1 0 0
Black-crowned night-heron Nycticorax nycticorax 2 0 1 1 Diversity 2023, 15, 1037 32 of 39
Table A2. Cont.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
Black-headed grosbeak Archilochus alexandri 2 1 1 0
Black-necked stilt Nycticorax nycticorax 3 4 0 0
Black-tailed gnatcatcher Pheucticus melanocephalus TWL 1 1 0 0 Black-tailed jackrabbit Himantopus mexicanus 6 3 6 1 −67 Black-throated gray warbler Polioptila melanura 0 1 1 0 Black phoebe Lepus californicus S 30 22 11 9 12
Blue-gray gnatcatcher Passerina caerulea 3 4 0 0
Blue grosbeak Polioptila caerulea 0 1 2 0
Bobcat Felis rufus 4 2 0 0
Botta’s pocket gopher Thomomys bottae 24 31 8 0 −100
Brewer’s blackbird Euphagus cyanocephalus 9 13 6 4 −54
Brewer’s sparrow Spizella breweri 0 1 0 0
Broad-footed mole Scapanus latimanus 0 2 0 0
Brown-headed cowbird Molothrus ater 9 3 3 1 0
Bryant’s Savannah sparrow Passerculus sandwichensis alaudinus G SSC 3 1 0 0 0 Bryant’s woodrat Neotoma bryanti 1 2 0 0
Bufflehead Bucephala albeola 2 1 0 0
Bullock’s oriole Icterus bullockii BCC 3 2 1 0
Burrowing owl Athene cunicularia R, G BCC, SSC 2, BOP 3 1 1 0 −100 Bushtit Psaltriparus minimus 19 21 4 2 −55
Cactus wren Campylorhynchus brunneicapillus 1 1 0 0
California brown pelican Pelicanus occidentalis californicus CFP 1 2 0 0 California gnatcatcher Polioptila c. californica FT, SSC2 4 1 0 0 California ground squirrel Otospermophilus beecheyi 18 24 8 0 −100 California gull Larus californicus BCC, TWL 22 15 11 10 33 California horned lark Eremophila alpestris actia G TWL 1 3 0 0 California quail Callipepla californica 6 10 3 0 −100
California scrub-jay Aphelocoma californica 19 25 10 9 −32 California thrasher Toxostoma redivivum BCC 4 3 0 0
California towhee Melozone crissalis 19 18 5 2 −58
California vole Microtus californicus 5 5 2 0 −100
Calliope hummingbird Stellula calliope 1 0 0 0
Canada goose Branta canadensis 12 19 5 4 −50
Cassin’s kingbird Tyrannus vociferans 8 10 2 0 −100
Cattle egret Bubulcus ibis 1 1 0 0
Cedar waxwing Bombycilla cedrorum 3 4 3 0 −100
Chestnut-backed chickadee Poecile rufescens 6 6 0 1
Chipping sparrow Spizella passerina 1 1 1 0
Cinnamon teal Spatula cyanoptera 1 2 0 0
Cliff swallow Petrochelidon pyrrhonota 10 14 4 4 −29
Common goldeneye Bucephala clangula 1 0 0 0
Diversity 2023, 15, 1037 33 of 39
Table A2. Cont.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
Common ground dove Columbina passerina 1 0 0 0
Common merganser Mergus merganser 1 0 0 0
Common raven Corvus corax S 33 34 11 10 −12
Common yellowthroat Geothlypis trichas 2 5 0 0
Cooper’s hawk Accipiter cooperii R TWL, BOP 13 10 3 4 73 Costa’s hummingbird Calypte costae BCC 1 0 0 0
Coyote Canis latrans 15 11 2 0 −100
Dark-eyed junco Junco hyemalis 7 7 3 1 −67
Deer mouse Peromyscus maniculatus 0 1 0 0
Desert cottontail Sylvilagus audubonii 12 7 2 1 −14
Domestic dog Canis familiaris 2 0 0 0
Double-crested cormorant Nannopterum auritum TWL 13 10 4 2 −35 Downy woodpecker Dryobates pubescens 2 5 1 0
Eared grebe Podiceps nigricollis 0 1 0 0
Eastern fox squirrel Sciurus niger 2 3 0 0
Eastern gray squirrel Sciurus carolinensis 3 4 3 1 −75
Egyptian goose Alopochen aegyptiacus 0 0 0 1
Eurasian collared-dove Streptopelia decaocto S 18 23 15 13 −32 European starling Sturnus vulgaris S 28 35 20 10 −60
Evening grosbeak Coccothraustes vespertinus 0 1 1 0
Ferruginous hawk Buteo regalis R TWL, BOP 0 1 2 0
Forster’s tern Sterna forstreri 1 2 0 1
Fox sparrow Passerella iliaca 0 0 1 0
Gadwall Anas strepera 1 1 0 0
Gambel’s quail Callipepla gambelii 1 1 0 0
Glaucous-winged gull Larus glaucescens 1 1 1 0
Golden eagle Aquila chrysaetos R BGEPA, CFP ,
BOP , TWL 0 2 0 0
Golden-crowned sparrow Zonotrichia atricapilla 1 4 1 0
Gopher snake Pituophis melanoleucus 1 0 0 0
Granite spiny lizard Sceloporus orcutti 1 1 0 0
Grasshopper sparrow Ammodramus savannarum G SSC 2 1 1 0 0 Gray fox Urocyon cinereoargenteus 2 1 0 0
Great Basin fence lizard Sceloporus occidentalis longipes 2 2 0 0 Great blue heron Ardea herodias 9 7 1 0
Great egret Ardea alba 17 12 7 1 −80
Great horned owl Bubo virginianus R BOP 5 3 0 0
Greater roadrunner Geococcyx californianus 1 3 0 0
Greater white-fronted goose Anser albifrons 1 0 0 0
Greater yellowlegs Tringa melanoleuca 2 1 0 0
Diversity 2023, 15, 1037 34 of 39
Table A2. Cont.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
Great-tailed grackle Quiscalus mexicanus 3 2 3 0 −100
Green heron Butorides virescens 1 1 0 0
Hairy woodpecker Dryobates villosus 3 4 0 0
Harbor seal Phoca vitulina 1 1 0 0
Herring gull Larus argentatus 2 0 2 0
Hooded merganser Lophodytes cucullatus 0 1 0 0
Hooded oriole Icterus cucullatus 3 5 0 0
Horned grebe Podiceps auritus 2 0 0 0
Horned lark Eremophila alpestris G 4 7 0 0
House cat Felis catus 5 3 3 2 11
House finch Haemorphous mexicanus S 45 46 18 12 −35
House sparrow Passer domesticus S 10 9 7 5 −21
House wren Troglodytes aedon 1 2 0 0
Hutton’s vireo Vireo huttoni 3 1 1 0
Kangaroo rat Dipodomys sp. 4 4 0 0
Killdeer Charadrius vociferus G 16 13 9 0 −100
Large-billed savannah sparrow Passerculus sandwichensis rostratus SSC2 1 0 0 0 Lark sparrow Chondestes grammacus 2 3 0 1
Lawrence’s goldfinch Spinus lawrencei BCC 4 2 0 0
Lazuli bunting Passerina amoena 1 2 0 0
Least sandpiper Calidris minutilla 1 2 0 0
Lesser goldfinch Spinus psaltria S 22 32 6 4 −54
Lesser nighthawk Chordeiles acutipennis 1 2 0 0
Lesser scaup Aythya affinis 1 0 0 0
Lesser yellowlegs Tringa flaviceps 0 1 0 0
Lincoln’s sparrow Melospiza lincolnii 7 10 1 0
Loggerhead shrike Lanius ludovicianus G SSC 2 2 2 3 0 −100 Long-billed curlew Numenius americanus TWL, BCC * 1 0 0 0 Long-tailed weasel Mustella frenata 1 0 0 0
MacGillivray’s warbler Geothlypus tolmiei 1 2 0 0
Mallard Anas platyrhynchos 17 19 3 1 −70
Marsh wren Cistothorus palustris 3 2 0 0
Merlin Falco columbarius R TWL, BOP 4 1 1 0
Merriam’s chipmunk Neotamias merriami 1 0 0 0
Mew gull Larus canus 1 0 0 0
Mouse sp. 1 0 0 0
Mountain bluebird Sialia currucoides 1 0 0 0
Mountain chickadee Parus gambeli 1 1 0 0
Mountain lion Puma concolor CCT 1 1 0 0
Mourning dove Zenaida macroura S 43 43 21 12 −43
Mule deer Odocoileus hemionus 7 6 0 0
Diversity 2023, 15, 1037 35 of 39
Table A2. Cont.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
Muskrat Ondatra zibethicus 1 0 0 0
Mute swan Cygnus olor 1 1 1 0
Nashville warbler Vermivora ruficapilla 1 1 0 0
Northern flicker Colaptes auratus 10 13 6 4 −49
Northern harrier Circus hudsonius R, G BCC, SSC 3, BOP 8 7 0 1 Northern mockingbird Mimus polyglottos S 27 25 16 13 −12 Northern pintail Anas acuta 1 0 0 0
Northern rough-winged swallow Stelgidopteryx serripennis 15 9 4 0 −100 Nutmeg mannikin Lonchura punctulata 2 2 0 0
Nuttall’s woodpecker Dryobates nuttallii BCC 8 15 1 1 −47 Oak titmouse Baeolophus inornatus BCC 5 11 1 0
Olive-sided flycatcher Contopus cooperi BCC, SSC2 3 2 0 0 Orange-crowned warbler Leiothlypis celata 3 6 1 0
Osprey Pandion haliaetus R TWL, BOP 2 2 1 1
Pacific-slope flycatcher Empidonax difficilis 1 2 0 0
Pacific wren Troglodytes pacificus 1 2 0 0
Pelagic cormorant Phalacrocorax pelagicus 0 1 0 0
Peregrine falcon Falco peregrinus R CFP , BOP 5 3 1 1
Phainopepla Phainopepla nitens 2 3 0 0
Pied-billed grebe Podilymbus podiceps 2 1 0 0
Prairie falcon Falco mexicanus R, G TWL, BOP 1 0 1 0
Purple finch Haemorhous purpureus 2 2 1 0
Pygmy nuthatch Sitta pygmaea 1 2 0 0
Raccoon Procyon lotor 3 3 1 0
Red-breasted nuthatch Sitta canadensis 1 3 0 0
Red-breasted sapsucker Sphyrapicus ruber 1 0 0 0
Red-masked parakeet Psittacara erythrogenys 0 1 0 0
Red-necked phalarope Phalaropus lobatus 0 1 0 0
Red-shouldered hawk Buteo lineatus R BOP 7 3 4 1 −42
Red-tailed hawk Buteo jamaicensis R BOP 35 44 17 7 −67
Red-winged blackbird Agelaius phoeniceus 14 16 6 1 −85
Red fox Vulpes vulpes 1 1 0 0
Ring-billed gull Larus delawarensis 4 7 0 0
Ring-necked pheasant Phasianus colchicus 2 0 0 0
Rock pigeon Columba livia S Non-native 21 31 18 15 −44
Rose-ringed parakeet Psittacula krameri 0 2 0 0
Ruby-crowned kinglet Regulus calendula 6 12 1 3 50
Ruddy duck Oxyura jamaicensis 0 1 0 0
Rufous hummingbird Selasphorus rufus BCC 1 0
San Diegan tiger whiptail Aspidoscelis tigris stejnegeri SSC 1 0 0 0 San Francisco common yellowthroat Geothlypis trichas sinuosa BCC, SSC3 1 1 0 0 Diversity 2023, 15, 1037 36 of 39
Table A2. Cont.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
San Francisco dusky-footed woodrat Neotoma fuscipes annectens SSC 1 1 0 0 Savannah sparrow Passerculus sandwichensis G 13 14 5 2 −63 Say’s phoebe Sayornis saya G 21 15 6 4 −7
Sharp-shinned hawk Accipiter striatus R TWL, BOP 1 2 0 1 Short-billed dowitcher Limnodromus griseus BCC 1 0 0 0
Short-eared owl Asio flammeus R BCC, SSC 3, BOP 2 0 0 0
Side-blotched lizard Uta stansburiana 1 0 0 0
Sierran treefrog Pseudacris sierra 3 9 1 0 −100
Snow goose Chen caerulescens 1 1 0 0
Snowy egret Egretta thula 3 4 0 1
Song sparrow Melospiza melodia 13 15 0 0
Sora Porzana carolina 1 0 0 0
Southern alligator lizard Gerrhonotus multicarinatus 2 1 0 0 Southern California
rufous-crowned sparrow Aimophila ruficeps canescens TWL 2 0 0 0 Southern mule deer Odocoileus hemionus fuliginatus 1 1 0 0 Southern Pacific rattlesnake Crotalus oreganus helleri 1 1 0 0 Southern sagebrush lizard Sceloporus graciosus vandenburgianus 1 0 0 0 Spotted sandpiper Actitis macularius 1 1 0 0
Spotted towhee Pipilo maculatus 10 11 0 0
Steller’s jay Cyanocitta stelleri 4 3 0 0
Striped skunk Mephitis mephitis 3 3 0 0
Surf scoter Melanitta perspicillata 0 1 0 0
Swainson’s hawk Buteo swainsoni R CT, BOP 5 3 3 1 −44
Swainson’s thrush Catharus ustulatus 1 1 0 0
Townsend’s warbler Setophaga townsendi 1 2 1 0
Tree swallow Tachycineta bicolor 6 6 0 1
Tricolored blackbird Agelaius tricolor G CT, BCC, SSC 1 2 0 0 0 Tundra swan Cygnus columbianus 1 0 0 0
Turkey vulture Cathartes aura R BOP 17 21 11 7 −49
Vaux’s swift Chaetura vauxi SSC2, BCC 0 1 0 0
Verdin Auriparus flaviceps BCC 1 1 0 0
Vermilion flycatcher Pyrocephalus rubinus G SSC 2 1 0 0 0 Violet-green swallow Tachycineta thalassina 3 4 0 1
Virginia opossum Didelphis virginianus 2 0 0 0
Warbling vireo Vireo gilvus 2 1 0 0
Western bluebird Sialia mexicana 9 12 0 0
Western fence lizard Sceloporus occidentalis 2 6 2 1 −83 Western gray squirrel Sciurus griseus 3 5 0 1
Western gull Larus occidentalis BCC 2 6 1 1 −67
Diversity 2023, 15, 1037 37 of 39
Table A2. Cont.

Species Scientific Name Type 1 Status 2
Control (n = 52) Impact (n = 26)
Effect (%)
Before After Before After
Western kingbird Tyrannus verticalis 10 9 4 2 −44
Western meadowlark Sturnella neglecta G 18 23 11 2 −86
Western sandpiper Calidris mauri 0 1 0 0
Western screech-owl Megascops kennicottii R BOP 1 0 0 0 Western side-blotched lizard Uta stansburiana elegans 4 5 0 0 Western tanager Piranga ludoviciana 2 2 0 0
White-breasted nuthatch Sitta carolinensis 4 6 0 0
White-crowned sparrow Zonotrichia leucophrys 17 23 9 4 −67 White-faced ibis Plegadis chihi TWL 1 3 1 0
White-tailed kite Elanus leucurus R CFP , BOP 14 7 7 0 −100 White-throated swift Aeronautes saxatalis 5 3 2 0 −100
White-winged dove Zenaida asiatica 1 0 0 0
Wild turkey Meleagris gallopavo 6 7 2 0 −100
Willet Tringa semipalmata BCC 3 3 0 0
Willow flycatcher Empidonax traillii CE 0 1 1 0
Wilson’s snipe Gallinago delicata 0 2 1 0
Wilsons warbler Wilsonia pusilla 3 6 0 0
Wood duck Aix sponsa 1 0 0 0
Wrentit Chamaea fasciata BCC 5 6 0 0
Yellow-billed magpie Setophaga petechia BCC 2 1 3 0 −100 Yellow-headed blackbird Pica nuttalli G SSC 3 1 0 0 0
Yellow-rumped warbler Xanthocephalus xanthocephalus S 21 21 7 10 43 Yellow warbler Setophaga coronata SSC2 5 6 0 2
1 G = grassland bird, R = raptor, S = synanthrope. 2 FT or FE = federal threatened or endangered, FC = federal candidate for listing as threatened or endangered, BGEPA = Bald and Golden Eagle Protection Act, BCC = U.S. Fish and Wildlife Service’s Birds of Conservation Concern, CT or CE = California threatened or endangered, CCT or CCE = candidate California threatened or endangered, CFP = California Fully Protected (California Fish and Game Code 3511), SSC = California Species of Special Concern (not threatened with extinction, but rare, very restricted in range, declining throughout range, peripheral portion of species’ range, associated with habitat that is declining in extent), SSC1, SSC2, and SSC3 = California Bird Species of Special Concern priorities 1, 2, and 3, respectively [57], TWL = California Taxa to Watch List, and BOP = Birds of Prey (California Fish and Game Code 3503.5). * Uncertain of range of BCC status based on 2021 Bird of Conservation Concern list.

Diversity 2023, 15, 1037 38 of 39
References

1. Marzluff, J.M.; Bowman, R.; Donnelly, R. A historical perspective on urban bird research: Trends, terms, and approaches. InAvian Ecology and Conservation in an Urbanizing World; Springer Science & Business Media: New York, NY, USA, 2001; pp. 1–18.

2. Alberti, M. The effects of urban patterns on Ecosystem function. Int. Reg. Sci. Rev. 2005, 28, 168–192. [CrossRef]

3. McKinney, M.L. Urbanization as a major cause of biotic homogenization. Biol. Conserv. 2006, 127, 247–260. [CrossRef]

4. Fenoglio, M.S.; Rossetti, M.R.; Videla, M. Negative effects of urbanization on terrestrial arthropod communities: A meta-analysis.

Glob. Ecol. Biogeogr. 2020, 29, 1412–1429. [CrossRef]

5. Piano, E.; Souffreau, C.; Merckx, T.; Baardsen, L.F.; Backeljau, T.; Bonte, D.; Brans, K.I.; Cours, M.; Dahirel, M.;Debortoli, N.; et al.

Urbanization drives cross-taxon declines in abundance and diversity at multiple spatial scales. Glob. Chang. Biol. 2020, 26, 1196–1211. [CrossRef] [PubMed]

6. Chatelain, M.; Rüdisser, J.; Traugott, M. Urban-driven decrease in arthropod richness and diversity associated with group-specific changes in arthropod abundance. Front. Ecol. Evol. 2023, 11, 980387. [CrossRef]

7. Donnelly, R.; Marzluff, J.M. Importance of reserve size and landscape context to urban bird conservation. Conserv. Biol. 2003, 18, 733–745. [CrossRef]

8. Crooks, K.R.; Suarez, A.V .; Bolger, D.T. Avian assemblages along a gradient of urbanization in a highly fragmented landscape.

Biol. Conserv. 2004, 115, 451–462. [CrossRef]

9. Chace, J.F.; Walsh, J.J. Urban effects on native avifauna: A review. Landsc. Urban Plan. 2006, 74, 46–69. [CrossRef]

10. Devictor, V .; Julliard, R.; Couvet, D.; Lee, A.; Jiguet, F. Functional homogenization effect of urbanization on bird communities.

Conserv. Biol. 2007, 21, 741–751. [CrossRef]

11. Aronson, M.F.J.; La Sorte, F.A.; Nilon, C.H.; Katti, M.; Goddard, M.A.; Lepczyk, C.A.; Warren, P .S.; Williams, N.S.; Cilliers, S.; Clarkson, B.; et al. A Global Analysis of the Impacts of Urbanization on Bird and Plant Diversity Reveals Key Anthropogenic Drivers. Proc. R. Soc. B Biol. Sci. 2014, 281, 20133330. [CrossRef]

12. Groffman, P .M.; Cavender-Bares, J.; Bettez, N.D.; Grove, J.M.; Hall, S.J.; Heffernan, J.B.; Hobbie, S.E.; Larson, K.L.; Morse, J.L.; Neill, C.; et al. Ecological homogenization of urban USA. Front. Ecol. Environ. 2014, 12, 74–81. [CrossRef]

13. Ibáñez-Álamo, J.D.; Rubio, E.; Benedett, Y.; Morell, F. Global loss of avian evolutionary uniqueness in urban areas.Glob. Change Biol. 2017, 23, 2990–2998. [CrossRef] [PubMed]

14. Salwasser, H. Conserving biological diversity: A perspective on scope and approaches. For. Ecol. Manag. 1990, 35, 79–90. [CrossRef]

15. Chapin III, F.; Zavaleta, E.; Eviner, V .; Naylor, R.L.; Vitousek, P .M.; Reynolds, H.L.; Hooper, D.U.; Lavorel, S.; SalaI, O.E.; Hobbie, S.E.; et al. Consequences of changing biodiversity. Nature 2000, 405, 234–242. [CrossRef] [PubMed]

16. Isaksson, C.; Rodewald, A.D.; Gil, D. Editorial: Behavioural and ecological consequences of urban life in birds. Front. Ecol. Evol. 2018, 6, 50. [CrossRef]

17. Smallwood, K.S. Habitat fragmentation and corridors. In Wildlife Habitat Conservation: Concepts, Challenges, and Solutions ; Morrison, M.L., Mathewson, H.A., Eds.; John Hopkins University Press: Baltimore, MD, USA, 2015; pp. 84–101.

18. Marzluff, J.M. Avian Ecology and Conservation in an Urbanizing World; Springer Science & Business Media: New York, NY, USA, 2001.

19. Delaney, K.S.; Riley, S.P .D.; Fisher, R.N. A rapid, strong, and convergent genetic response to urban habitat fragmentation in four divergent and widespread vertebrates. PLoS ONE 2010, 5, e12767. [CrossRef]

20. Sol, D.; Lapiedra, O.; Gonz’alez-Lagos, C. Behavioural adjustments for a life in the city. Anim. Behav. 2013, 85, 1101–1112. [CrossRef]

21. Rosenberg, K.V .; Dokter, A.M.; Blancher, P .J.; Sauer, J.R.; Smith, A.C.; Smith, P .A.; Stanton, J.C.; Panjabi, A.; Helft, L.; Parr, M.; et al.

Decline of the North American avifauna. Science 2019, 366, 120–124. [CrossRef]

22. McKinney, M.L. Urbanization, Biodiversity, and Conservation. BioScience 2002, 52, 10. [CrossRef]

23. Rodríguez-Estrella, R. Land use changes affect distributional patterns of desert birds in the Baja California peninsula, Mexico.

Divers. Distrib. 2007, 13, 877–889. [CrossRef]

24. Germaine, S.S.; Rosenstock, S.S.; Schweinsburg, R.E.; Richardson, W.S. Relationships among breeding birds, habitat, and residential development in greater Tucson, Arizona. Ecol. Appl. 1998, 8, 680–691. [CrossRef]

25. Palomino, D.; Carrascal, L.M. Urban influence on birds at a regional scale: A case study with the avifauna of northern Madrid province. Landsc. Urban Plan. 2006, 77, 276–290. [CrossRef]

26. Minor, E.; Urban, D. Forest bird communities across a gradient of urban development. Urban Ecosyst. 2010, 13, 51–71. [CrossRef]

27. Kang, W.; Minor, E.S.; Park, C.R.; Lee, D. Effects of habitat structure, human disturbance, and habitat connectivity on urban forest bird communities. Urban Ecosyst. 2015, 18, 857–870. [CrossRef]

28. Blair, R.B. Creating a homogeneous avifauna. In Avian Ecology and Conservation in an Urbanizing World ; Springer Science & Business Media: New York, NY, USA, 2001; pp. 459–486.

29. Cam, E.; Nichols, J.D.; Sauer, J.R.; Hines, J.E.; Flather, C.H. Relative species richness and community completeness: Birds and urbanization in the mid-Atlantic states. Ecol. Appl. 2000, 10, 1196–1210. [CrossRef]

30. Karr, J.R. Landscapes and Management for Ecological Integrity. In Biodiversity and Landscape: A Paradox for Humanity; Kim, K.C., Weaver, R.D., Eds.; Cambridge University Press: New York, NY, USA, 1994; pp. 229–251.

Diversity 2023, 15, 1037 39 of 39

31. Scott, T.A. Initial effects of housing construction on woodland birds along the wildland urban interface. In Interface Between Ecology and Land Development in California; Keeley, J.E., Ed.; Southern California Academy of Sciences: Los Angeles, CA, USA, 1993; pp. 181–187.

32. Hostetler, M.; Duncan, S.; Paul, J. Post-construction effects of an urban development on migrating, resident, and wintering birds.

Southeast. Nat. 2005, 4, 421–434. [CrossRef]

33. Watson, D.M. The “standardized search”: An improved way to conduct bird surveys. Austral Ecol. 2003, 28, 515–525. [CrossRef]

34. Long, A.M.; Col ón, M.R.; Bosman, J.L.; Robinson, D.H.; Pruett, H.L.; McFarland, T.M.; Mathewson, H.A.; Szewczak, J.M.; Newnam, J.C.; Morrison, M.L. A before–after control–impact assessment to understand the potential impacts of highway construction noise and activity on an endangered songbird. Ecol. Evol. 2017, 7, 379–389. [CrossRef]

35. Smallwood, K.S.; Bell, D.A. Effects of wind turbine curtailment on bird and bat fatalities. J. Wildl. Manag. 2020, 84, 684–696. [CrossRef]

36. Sørenson, T. A method of establishing groups of equal amplitude in plant sociology based on similarity of species content.K.

Dan. Vidensk. Selsk. Skr. 1948, 5, 1–34.

37. Callaghan, C.T.; Major, R.E.; Wilshire, J.H.; Martin, J.M.; Kingsford, R.T.; Cornwell, W.K. Generalists are the most urban-tolerant of birds: A phylogenetically controlled analysis of ecological and life history traits using a novel continuous measure of bird responses to urbanization. Oikos 2019, 128, 845–858. [CrossRef]

38. Blair, R.B. Land use and avian species diversity along an urban gradient. Ecol. Appl. 1996, 6, 506–519. [CrossRef]

39. González-Lagos, C.; Cardador, L.; Sol, D. Invasion success and tolerance to urbanization in birds. Ecography 2021, 44, 1642–1652. [CrossRef]

40. Bolger, D.T.; Alberts, A.C.; Sauvajot, R.M.; Potenza, P .; McCalvin, C.; Tran, D.; Mazzoni, S.; Soule, M.E. Response of Rodents to Habitat Fragmentation in Coastal Southern California. Ecol. Appl. 1997, 7, 552–563. [CrossRef]

41. Sauvajot, R.M.; Buechner, M.; Kamradt, D.A.; Shonewald, C.M. Patterns of human disturbance and response by small mammals and birds in chaparral near urban development. Urban Ecosyst. 1998, 2, 279–297. [CrossRef]

42. McKinney, M.L. Effects of urbanization on species richness: A review of plants and animals. Urban Ecosyst. 2008, 11, 161–176. [CrossRef]

43. Smallwood, K.S.; Morrison, M.L. Nest-site selection in a high-density colony of burrowing owls. J. Raptor. Res. 2018, 52, 454–470. [CrossRef]

44. Smallwood, K.S.; Smallwood, N.L. Breeding density and collision mortality of loggerhead shrike ( Lanius ludovicianus) in the Altamont Pass Wind Resource Area. Diversity 2021, 13, 540. [CrossRef]

45. Andrén, H. Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: A review. Oikos 1994, 71, 355–366. [CrossRef]

46. Clergeau, P .; Jokimäki, J.; Savard, J.P .L. Are urban bird communities influenced by the bird diversity of adjacent landscapes?J.

Appl. Ecol. 2001, 38, 1122–1134. [CrossRef]

47. Hutto, R.L. Should scientists be required to use a model-based solution to adjust for possible distance-based detectability bias? Ecol. Appl. 2016, 26, 1287–1294. [CrossRef]

48. Thompson, W.L. Towards reliable bird surveys: Accounting for individuals present but not detected. Auk 2002, 119, 18–25. [CrossRef]

49. Narango, D.L.; Tallamy, D.W.; Marra, P .P . Native plants improve breeding and foraging habitat for an insectivorous bird.Biol.

Conserv. 2017, 213, 42–50. [CrossRef]

50. Adams, B.J.; Li, E.; Bahlai, C.A.; Meineke, E.K.; McGlynn, T.P .; Brown, B.V . Local and landscape-scale variables shape insect diversity in an urban biodiversity hot spot. Ecol. Appl. 2020, 30, e02089. [CrossRef]

51. Lerman, S.B.; Warren, P .S. The conservation value of residential yards: Linking birds and people.Ecol. Appl. 2011, 21, 1327–1339. [CrossRef]

52. Burghardt, K.T.; Tallamy, D.W.; Shriver, W.G. Impact of native plants on bird and butterfly biodiversity in suburban landscapes.

Conserv. Biol. 2008, 23, 219–224. [CrossRef]

53. Berthon, K.; Thomas, F.; Bekessy, S. The role of ‘nativenes’ in urban greening to support animal biodiversity.Landsc Urban Plan 2021, 205, 103959. [CrossRef]

54. Smallwood, N.L.; Wood, E.M. The ecological role of native-plant landscaping in residential yards to birds during the nonbreeding period. Ecosphere 2022, 14, e4360. [CrossRef]

55. Goddard, M.A.; Dougill, A.J.; Benton, T.G. Scaling up from gardens: Biodiversity conservation in urban environments. Trends Ecol. Evol. 2009, 25, 90–98. [CrossRef]

56. Tallamy, D.W.Nature’s Best Hope: A New Approach to Conservation That Starts in Your Yard; Timber Press: Portland, OR, USA, 2020.

57. Shuford, W.D.; Gardali, T.California Bird Species of Special Concern: A Ranked Assessment of Species, Subspecies, and Distinct Populations of Birds of Immediate Conservation Concern in California; Studies of Western Birds 1; Western Field Ornithologists and California Department of Fish and Game: Camarillo, CA, USA, 2008.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

ORIGINAL ARTICLE
Impact of Ambient Ultrafine Particles on Cause-Specific Mortality in Three German Cities
Maximilian Schwarz 1,2, Alexandra Schneider 1, Josef Cyrys 1, Susanne Bastian 3, Susanne Breitner 1,2*, and Annette Peters 1,2,4*
1Institute of Epidemiology, Helmholtz Zentrum M€unchen–German Research Center for Environmental Health, Neuherberg, Germany; 2Institute for Medical Information Processing, Biometry and Epidemiology, Medical Faculty, Ludwig-Maximilians-Universit€at M€unchen, Munich, Germany;3Saxon State Office for Environment, Agriculture and Geology, Dresden, Germany;4Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA Abstract
Rationale: Exposure to ambient air pollution has been
associated with adverse effects on morbidity and mortality.

However, the evidence for ultrafine particles (UFPs; 10 – 100 nm) based on epidemiological studies remains scarce and inconsistent.

Objectives: We examined associations between short-term exposures to UFPs and total particle number concentrations (PNCs; 10– 800 nm) and cause-specific mortality in three German cities: Dresden, Leipzig, and Augsburg.

Methods: We obtained daily counts of natural, cardiovascular, and respiratory mortality between 2010 and 2017. UFPs and PNCs were measured at six sites, and measurements of fine particulate matter (PM 2.5; <2.5 μm in aerodynamic diameter) and nitrogen dioxide were collected from routine monitoring.

We applied station-specific confounder-adjusted Poisson regression models. We investigated air pollutant effects at aggregated lags (0 – 1, 2– 4, 5– 7, and 0 – 7 d after UFP exposure) and used a novel multilevel meta-analytical method to pool the results. Additionally, we assessed interdependencies between pollutants using two-pollutant models.

Measurements and Main Results: For respiratory mortality, we found a delayed increase in relative risk of 4.46% (95% confidence interval, 1.52 to 7.48%) per 3,223-particles/cm 3 increment 5– 7 days after UFP exposure. Effects for PNCs showed smaller but comparable estimates consistent with the observation that the smallest UFP fractions showed the largest effects. No clear associations were found for cardiovascular or natural mortality. UFP effects were independent of PM

2.5 in two-
pollutant models.

Conclusions: We found delayed effects for respiratory
mortality within 1 week after exposure to UFPs and PNCs but no associations for natural or cardiovascular mortality. This finding adds to the evidence on the independent health effects of UFPs.

Keywords: ambient air pollution; ultrafine particles; particle number concentrations; particulate matter; respiratory mortality Evidence of adverse health effects of
ambient air pollution has been
consistently growing in recent decades. By
now, there are numerous studies that have
found an association between short- and
long-term exposure to particulate matter
(PM) or nitrogen dioxide (NO
2)a n d
morbidity (1, 2) and mortality (3, 4);
however, air pollution comprises a
complex mixture of many other
substances, sometimes originating from
similar sources (5).
(Received in original form October 7, 2022; accepted in final form March 3, 2023) This article is open access and distributed under the terms of the Creative Commons Attribution Non-Commercial No Derivatives License 4.0 .

For commercial usage and reprints, please e-mail Diane Gern ([email protected]). *Co–last authors.

Supported by grants from the Saxon State Office for Environment, Agriculture and Geology, Dresden, Germany.

Author Contributions: S. Bastian, A.S., J.C., S. Breitner, A.P., and M.S. involved in conceptualization of the study design. M.S. and S. Breitner analyzed the data and drafted the first version of the manuscript. S. Bastian, A.S., and J.C. provided assistance with the methodological framework. All authors read the manuscript and contributed to the final version.

Correspondence and requests for reprints should be addressed to Maximilian Schwarz, M.Sc., Institute of Epidemiology, Helmholtz Zentrum M€unchen–German Research Center for Environmental Health, Ingolst €adter Landstrasse 1, 85764 Neuherberg, Germany.

E-mail: [email protected].

This article has a related editorial.

This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.

Am J Respir Crit Care Med Vol 207, Iss 10, pp 1334–1344, May 15, 2023 Copyright © 2023 by the American Thoracic Society
Originally Published in Press as DOI: 10.1164/rccm.202209-1837OC on March 6, 2023 Internet address: www :atsjournals:org
1334 American Journal of Respiratory and Critical Care Medicine Volume 207 Number 10 | May 15 2023 Ultrafine particles (UFPs) are an
important part of particulate air pollution
but differ from PM in many ways. UFPs are
conventionally defined as particles with an
aerodynamic diameter<0.1 μm. As a result
of their small particle size, they contribute
negligibly to the total particle mass but
dominate the number concentration (6). In
addition, a large surface area per unit mass
and high surface reactivity give UFPs the
ability to transport chemical compounds;
thus, UFPs are considered more hazardous
than PM (7). UFPs are emitted directly or
formed secondarily in the air by
photochemical processes from gaseous
precursors (8). Traffic exhaust, nucleation
processes from several sources, or general
combustion have been reported to be the
main contributors to UFPs in urban air (9).

To date, UFPs are not routinely monitored
because the measurement techniques are
more elaborate and complex, and there are
no regulatory initiatives yet that would
incorporate continuous measurements (8).

We were thefirst to publish evidence of
delayed impacts of UFPs on daily mortality
in a high-pollution setting in Erfurt,
Germany, in the 1990 s (10, 11). However,
recent review articles have reported a
growing number of epidemiological studies
that suggested associations between the
number concentrations of UFPs and several
morbidity (6, 12, 13) and mortality (6, 12)
outcomes. Nevertheless, evidence was
summarized as insufficient and inconclusive
because of heterogeneity in exposure
assessment and assignment (e.g., different
measurement devices or exposure metrics)
and study methods (e.g., modeling strategies
or copollutant adjustment) (6, 8, 12, 13).

In 2021, the World Health Organization
(WHO) updated its air quality guidelines,
recommending stricter target values for some
ambient air pollutants, including PM<2.5
μm in aerodynamic diameter (PM

2.5)( W H O
2021 Global Air Quality Guidelines: annual
mean PM

2.5,5 μg/m3 [14]), based on
evidence of adverse health effects even at low
exposure concentrations (3). However, the
assessment of epidemiological literature did
not allow for the establishment of new
evidence-based reference values for UFPs
because the body of evidence is still
inadequate (14). Nevertheless, the
importance of UFPs was highlighted in a
good practice statement, which particularly
calls for more monitoring data and its use in
epidemiological studies.

Therefore, the objective of this study was
to investigate short-term associations between
the number concentrations of ambient UFPs,
total particle number concentrations (PNCs),
and daily cause-specific mortality over a study
period of 8 years in three German cities with
multiple monitoring stations. Additionally,
we investigated the impact of subfractions of
UFPs and effect modification by age, sex, and
season and assessed interdependencies
between pollutants using two-pollutant
models. Some of the results of this study have
been previously reported in the form of an
abstract (15).

Methods
Mortality Data
We obtained data on daily cause-specific
death counts for the three German cities,
Dresden, Leipzig, and Augsburg, between
2010 and 2017 through official statistics.

Cases were considered if people lived and
d i e di nt h es a m ec i t y .T h ef o l l o w i n gt h r e e cause-specificd e a t h sw e r ed efined using the
International Classification of Diseases,
10th Revision: natural (A00– R99),
cardiovascular (I00– I99), and respiratory
mortality (J00– J99). No informed consent
or approval was needed because data are
collected routinely and anonymously.

Environmental Data
Hourly air pollution data and hourly air
temperature, relative humidity, and
barometric pressure were measured at six
fixed monitoring stations that were part of
the former German Ultrafine Aerosol
Network (GUAN). A map of all GUAN
stations is shown in Figure 1. Stations
were selected based on three criteria:
1) representativeness of the exposure setting
for the population,2)a na d e q u a t en u m b e ro f
cases, and3) high standardization and good
comparability of the measurement devices.

More details are provided in the online
supplement. Four selected stations were
considered as urban background stations. In
addition, two traffic-related stations were
included to capture peak concentrations
more adequately. Particle number size
distributions were obtained in a size range of
10– 800 nm using mobility particle size
spectrometers. PM

2.5 mass concentrations
were measured by tapered element
oscillating microbalance for Augsburg and
high-volume samplers at the other stations.

Black carbon (BC) mass concentrations were
obtained by multiangle absorption
photometers for all stations except Augsburg,
where an aethalometer was used. Daily mean
concentrations were calculated if>75% of
the hourly data were available.

We considered the number
concentrations of particles in the ultrafine
range (10– 100 nm, i.e., UFPs) and total
PNCs (10– 800 nm) as exposures of primary
interest. In addition, we also assessed
nucleation-mode particles (10– 30 nm;
NuMPs), Aitken-mode particles (30– 100 nm;
AiMPs), and accumulation-mode particles
(100– 800 nm; AcMPs). Air pollutants of
secondary interest were NO
2,P M2.5,a n dB C .

At a Glance Commentary
Scientific Knowledge on the
Subject: Ambient air pollution has
been associated with adverse health
effects on morbidity and mortality,
but the epidemiological evidence for
unregulated ultrafine particles
(UFPs; 10 – 100 nm) remains scarce
and inconclusive. To date, UFPs
are not routinely monitored, and
therefore time-series analyses
assessing the link between short-
term UFP exposures and cause-
specific mortality need dedicated
monitoring campaigns.

What This Study Adds to the
Field: This multicity
epidemiological time-series study
included a highly standardized
exposure monitoring network with
data from eight consecutive years
(2010– 2017). It is one of the first
studies to incorporate multiple UFP
monitoring stations per area region
in three German cities. A novel
multilevel meta-analytical approach
showed a delayed increase in the
risk of respiratory mortality after
exposure to UFPs. These
observations were independent of
other air pollutants. Further analysis
revealed larger associations for
women and no difference by age or
season. UFPs, particularly smaller
size fractions (nucleation-mode
particles, 10 – 30 nm), may contribute
to the overall risk of mortality from
ambient air pollution.

ORIGINAL ARTICLE
Schwarz, Schneider, Cyrys, et al. : Ambient Ultrafine Particles and Mortality 1335 Statistical Analysis
We conducted a two-stage modeling
approach. In thefirst stage, we calculated
station-specific associations between air
pollutants and cause-specificm o r t a l i t yu s i n g
Poisson regression models allowing for
overdispersion. We adjusted for the
following confounders: time trend, day of the
week, vacation periods, public holidays, air
temperature, and relative humidity. Effects of
high and low temperatures were added
separately to the model according to
Stafoggia and colleagues (2013) (16). We
used cubic regression splines for time trends
(four degrees of freedom per year) and
meteorological parameters (three degrees of
freedom) to account for nonlinear
confounding. We analyzed associations
between air pollutants and mortality using
different aggregated lags. Specifically, we
assessed immediate (0– 1da f t e re x p o s u r e
[lag0– 1]), delayed (2– 4da n d5 – 7da f t e r
exposure [lag2– 4a n dl a g 5– 7, respectively])
and overall effects (lag0– 7). In the second
stage, station-specific estimates were pooled
using a novel multilevel meta-analysis that
accounts for hierarchical structures,
including random terms for cities and
stations (17). We tested for heterogeneity
between the station-specific estimates and
obtained the correspondingP value and
I
2 statistic. All results are presented as a
percent change per interquartile range
increase in the respective air pollutant
concentration to compare the relative health
effects across pollutants. A detailed
description is provided in the online
supplement.

On an exploratory basis, we conducted
several further analyses. We compared the
effects between urban background and
traffic-related stations. Two-pollutant models
were calculated if the Spearman correlation
coefficient was less than 0.7. We assessed
potential effect modifications by sex (male vs.
female) and age (,75 yr vs.>75 yr) in
stratified analyses. Seasonal differences
(October to March vs. April to September)
were analyzed using an interaction term
between the air pollutant and season. We
conducted several sensitivity analyses to test
the robustness of our results (e.g., different
model parameters, confounding variables,
and measurement stations) and also provide
the results of the main models using afixed
increment in air pollution concentration.

Detailed information on the station
characteristics (e.g., location and station
operator, environmental data, and
measurement devices) has been published
elsewhere (18, 19) and is provided in the
online supplement, together with a detailed
description of mortality data and statistical
methods.

Results
Descriptive Analysis
Total numbers and daily means of cause-
specific mortality and population data are
Figure 1. Location of German Ultrafine Aerosol Network stations across Germany and classification according to station type. Stations used for this analysis are highlighted with red boxes (dashed for sensitivity analysis). Map adapted from Birmili and colleagues, 2016 (18), and Sun and colleagues, 2019 (19).

ORIGINAL ARTICLE
1336 American Journal of Respiratory and Critical Care Medicine Volume 207 Number 10 | May 15 2023 presented in Table 1 (see Table E4 in the
online supplement).

Station-specific descriptive statistics of
air pollution and environmental data are
shown in Table 2 (a more detailed
overview is given in Table E5). In general,
median UFP concentrations at the urban
background stations were in the mid-4,000 s
in particles/cm
3, except for Augsburg, where
a median UFP concentration of 5,655
particles/cm
3 was observed. At the traffic-
related stations, significantly increased
concentrations were measured, with median
UFP concentrations of 8,637 particles/cm3
for Dresden-Nord and 10,123 particles/cm3
for Leipzig-Mitte, respectively (Table 2).

Spearman correlation coefficients
indicated mainly weak to moderate
correlations between UFPs and BC, NO
2,
and PM2.5 (Table E6); UFPs and PNCs were
highly correlated within stations (coefficients
between 0.96 and 0.98) and moderately
correlated between stations (Table E7).

Compared with UFPs, higher correlations
between PNCs and BC, NO
2,a n dP M2.5
were observed.

Main Models/Analysis
Results of the pooled analysis are presented
in Figure 2 (see Table E8). No clear
associations were observed between UFPs or
PNCs and natural or cardiovascular
mortality; however, both exposures were
associated with respiratory mortality. The
strongest effects were seen for the delayed
aggregated lags, especially lag5– 7. For
example, an interquartile range increase of
3,223 particles/cm
3 in UFP concentration
was associated with a 4.46% (95% confidence
interval [CI], 1.52% to 7.48%) increase in the
relative risk of respiratory mortality. No
heterogeneity was observed between station-
specifice s t i m a t e s(I
2 = 4.90%,P = 0.385). The
results for PNCs were comparable but with
smaller effect sizes. Looking at different
particle size modes, we observed more
pronounced effects on respiratory mortality,
predominantly for the smallest subfraction,
NuMPs (lag5– 7, 4.49% [95% CI, 1.91% to

7.14%]) (Figure 3 and Table E9). In contrast,
for natural or cardiovascular mortality, there
were no changes in risk depending on
particle size fractions, although NuMPs
i n d i c a t e dah i g h e rd e l a y e dr i s kf o rn a t u r a l mortality.

The results offixed-effects models,
station-specific estimates, single-lag models,
and a comparison between urban
background and traffic-related stations
indicated mainly higher risks for respiratory
mortality at the Leipzig stations, the single
lags of 3 and 6 days, and the urban
background (see online supplement for
detailed information; Figures E1– E3 and
Tables E10 and E11).

Table 1. Description of Population and Mortality Data (2,922 Days with Valid Data) Variable Leipzig Dresden Augsburg
Mean population, 2010 –2017 542,918 534,382 279,159
Total counts of natural mortality 43,250 36,106 20,712
Total counts of cardiovascular mortality 19,880 15,756 8,854 Total counts of respiratory mortality 2,559 2,143 1,426 Daily natural mortality 14.8 6 4.1 12.4 6 3.7 7.1 6 2.7 Daily cardiovascular mortality 6.8 6 2.7 5.4 6 2.4 3.0 6 1.7 Daily respiratory mortality 0.9 6 1.0 0.7 6 0.9 0.5 6 0.7 Values are presented as mean 6 SD where applicable. Population data based on official statistical yearbook of the cities: own calculations; natural mortality: International Classification of Diseases, 10th Revision: A00–R99; cardiovascular mortality, I00–I99; respiratory mortality, J00–J99. Source:

Research Data Centre (RDC) of the Federal Statistical Office and Statistical Offices of the Federal States (Mortality Statistics [uniform directory number: 23211], survey years, 2010–2017; DOI:

10.21242/23211.2010.00.00.1.1.0 to 10.21242/23211.2017.00.00.1.1.0, own calculations).

Table 2. Concentrations of Air Pollution and Environmental Data per Measurement Station Variable LMI LWE LTR DDN DDW AFH
Station characteristic Traffic-related Urban
background
Urban
background
Traffic-related Urban
background
Urban
background
Air pollutant
UFP (10 –100 nm),
particles/cm3
10,123 (5,156) 4,520 (3,003) 4,838 (3,154) 8,637 (4,366) 4,791 (3,156) 5,655 (3,514) PNC (10 –800 nm),
particles/cm3
11,922 (5,866) 5,748 (3,482) 6,054 (3,686) 10,292 (4,975) 6,186 (3,902) 6,909 (4,017) BC, μg/m3 2.0 (1.3) 0.8 (0.8) 0.7 (0.8) 1.5 (1.1) 0.7 (0.8) 1.4 (1.0) NO2, μg/m3 43.0 (17.0) 16.0 (11.0) NA 33.0 (14.0) 18.0 (12.0) 17.7 (12.3) PM2.5, μg/m3 13.6 (12.2) 9.6 (10.5) NA 12.3 (11.6) 10.9 (12.3) 10.2 (10.3) Meteorological
parameter
Temperature, /C14C 11.4 (12.1) 9.7 (11.7) NA 11.3 (12.5) 11.6 (12.4) 9.9 (12.2) Relative humidity, % 71.8 (19.6) 75.3 (18.6) NA 70.9 (16.8) 71.8 (17.3) 79.2 (20.3) Barometric
pressure, hPa
1,016.0 (10.0) 1,016.0 (10.0) NA 1,016.0 (10.0) 1,016.0 (10.0) 961.4 (9.0) Values are presented as median (interquartile range).

Definition of abbreviations : AFH = Augsburg-Hochschule; BC = black carbon; DDN = Dresden-Nord; DDW = Dresden-Winckelmannstrasse; LMI = Leipzig-Mitte; LTR = Leipzig-Leibniz Institute for Tropospheric Research; LWE = Leipzig-West; NA = no data available; NO 2 = nitrogen dioxide; PM2.5 = particulate matter <2.5 mm in aerodynamic diameter; PNC = total particle number concentrations (10–800 nm); UFP = ultrafine particles (10–100 nm).

ORIGINAL ARTICLE
Schwarz, Schneider, Cyrys, et al. : Ambient Ultrafine Particles and Mortality 1337 BC was not associated with any
mortality outcome (see Table E8). The
results rather indicated null effects with no
distinct pattern. For PM

2.5 and NO2,t h e
largest effects on natural or cardiovascular
mortality were observed for the aggregated
average lag2– 4, although substantial
heterogeneity was observed for natural
mortality. An increase in NO
2 of 11.00μg/m3
was associated with a 1.73% (95% CI, 0.60%
to 2.88%) higher risk of cardiovascular death
(I
2 =0 . 0 % ,P = 0.669). There were no
significant associations with respiratory
mortality for PM2.5 or NO2, although the
effect estimates for NO2 were all positive
(see Table E8).

Two-pollutant models and effect
modification analysis are reported based on
the combination of pollutant, lag structure,
and mortality endpoint, for which the
strongest effects were found in the main
analysis. The results are presented in Figure 4
and Table E12. The UFP effects on
respiratory mortality 5– 7d a y sa f t e re x p o s u r e remained rather unchanged after additional
adjustment for BC or PM

2.5, indicating an
independent effect (e.g., UFP1 PM2.5, 4.07%
[95% CI, 0.93% to 7.30%]), whereas further
adjustment for NO2 l e dt ow i d e rC I s .F o r
P N C s ,as i m i l a rp a t t e r nw a sf o u n d .H o w e v e r , it should be noted that, for NO2,t h er e s u l t s
for the Leipzig-Mitte station were excluded
from the pooled estimates because of high
Spearman correlation coefficients, leading to
more imprecise results.

Associations between respiratory
mortality and UFPs 5– 7 days after exposure
were significantly stronger in women, with a

9.57% (95% CI, 5.35% to 13.97%) increase in
risk, compared with 0.45% (95% CI,23.10%
to 4.13%) in men (Figure 4 and Table E12).

No substantial effect modifications were seen
Figure 2. Percent change in relative risk and 95% confidence interval per interquartile range (IQR) increase in air pollution concentration for natural (top), cardiovascular (middle), and respiratory mortality (bottom). The x-axis shows the 24-hour moving average lag concentrations of air pollutants. The y-axis represents the percent change of risk per IQR increase in air pollution concentration (difference between the 75th and 25th percentiles; corresponds to the spread of the middle 50% of the data). Standardization by IQR facilitates comparison between different pollutants. The shape of the estimates displays the type of pollutant. All estimates represent the pooled analysis of the measurement stations using multilevel random-effects models and were adjusted for main model covariates. BC = black carbon; NO 2 = nitrogen dioxide; PM2.5 = particulate matter <2.5 mm in aerodynamic diameter; PNC = total particle number concentrations (10–800 nm); UFP = ultrafine particles (10–100 nm).

ORIGINAL ARTICLE
1338 American Journal of Respiratory and Critical Care Medicine Volume 207 Number 10 | May 15 2023 for age and season. Generally, PNCs showed
similar results.

Sensitivity Analysis
Sensitivity analyses were again done for
selected combinations of pollutant, lag
structure, and mortality endpoints. Overall,
changing several model parameters,
adjusting for additional variables, or using
alternative definitions of UFPs and PNCs or
city-specific exposures resulted in only
minor changes in the UFP effect estimates
(Figures 5 and E4 and Table E9). For the
main analyses, an alternative standardization
method withfixed increments resulted in
larger effect estimates but also in wider CIs.

However, the direction and significance of
the estimates were not affected (Figures
E5– E7 and Tables E13 and E14). Additional
inclusion of a different urban background
station in the pooled analysis generated
comparable, albeit lower, effects, still
indicating a higher risk for respiratory
mortality (see the online supplement and
T a b l eE 9 ) .F i n a l l y ,t h ee x p o s u r e– response functions for UFPs (lag5– 7) and respiratory
mortality indicated no significant deviations
from linearity (Figure E8).

Discussion
This time-series analysis found delayed
associations between UFPs and PNCs and
respiratory mortality. The strongest effects
were seen for UFPs with a delay of 5– 7d a y s .

Consistently, we found the strongest effect
with particle number concentrations in the
size range of 10– 30 nm. In contrast, we found
no clear effects on natural or cardiovascular
mortality. For respiratory mortality, adjust-
m e n tf o ro t h e ra i rp o l l u t a n t ss u c ha sP M

2.5 or
BC indicated independent results; adjustment
for NO2 led to wider CIs and insignificant
results. Thefindings were comparable
between age groups and seasons, but more
pronounced risks were observed for women.

A multicity study conducted in eight
European cities (20) reported weak delayed
pooled effects of PNCs that were strongest
for a single lag of 6 days and respiratory
Figure 3. Percent change in relative risk and 95% confidence interval per interquartile range (IQR) increase in air pollution concentration for natural (top), cardiovascular (middle), and respiratory mortality (bottom). The x-axis shows the 24-hour moving average lag concentrations of air pollutants. The y-axis represents the percent change of risk per IQR increase in air pollution concentration (difference between the 75th and 25th percentiles; corresponds to the spread of the middle 50% of the data). Standardization by IQR facilitates comparison between different pollutants. The shape of the estimates displays the type of pollutant by particle size mode. All estimates represent the pooled analysis of the measurement stations using multilevel random-effects models and were adjusted for main model covariates.

ORIGINAL ARTICLE
Schwarz, Schneider, Cyrys, et al. : Ambient Ultrafine Particles and Mortality 1339 mortality. Thesefindings are consistent with
our results, as we found increased risks for
respiratory mortality at single lags of 3 and
6 days. In contrast to our results, Stafoggia
and colleagues observed higher effect
estimates for natural and cardiovascular
mortality. However, the authors pointed out
null effects after removing the most
influential station from the pooled analysis
(20). Another study conducted infive central
and western European cities, including
Augsburg and Dresden, Germany (21),
found positive, albeit insignificant, pooled
delayed effects for respiratory mortality after
exposure to UFPs or PNCs (e.g., UFP lag2– 5
and respiratory mortality: 8.5% [95% CI,

24.8% to 23.7%] per 2,750 particles/cm
3).

The effects were independent of PM2.5;
natural and cardiovascular mortality were
not associated with UFPs or PNCs (21).

Although differences exist between this
previous study and our study (e.g., lag
structures and lower cutoff values in the UFP
definition), we observed comparable results.

Furthermore, the CIs indicated a higher
degree of precision in our study, probably
due to the substantially longer time series. In
a single-station analysis in the German Ruhr
area, Hennig and colleagues reported higher
risks for respiratory mortality following lag2
(3.50% [95% CI,20.77% to 7.95%]) and lag6
(4.51% [95% CI, 0.37% to 8.81%]) exposures
to UFPs (22). However, no clear pattern was
found for average lag effects (22). As a
sensitivity analysis, we included the
M€ulheim-Styrum monitoring station used by
Hennig and colleagues in our main model.

Despite some methodological differences
(see online supplement), the UFP effects on
Figure 4. Percent change in relative risk and 95% confidence interval per interquartile range (IQR) increase in concentration of particles in the ultrafine range (10–100 nm; UFPs; top panel) and total particle number concentrations (10–800 nm; PNCs; bottom panel) for respiratory mortality (5–7 d after UFP exposure). The x-axis shows the results for the main (dots), two-pollutant (rectangles), and effect-modification analyses (diamonds). The y-axis represents the percent change of risk per IQR increase in air pollution concentration (difference between the 75th and 25th percentiles; corresponds to the spread of the middle 50% of the data). Standardization by IQR facilitates comparison between different pollutants. All estimates represent the pooled analysis of the measurement stations using multilevel random-effects models and were adjusted for main model covariates. It should be noted that, for the two-pollutant models for PM 2.5 and nitrogen dioxide (NO 2), the station Leipzig-Leibniz Institute for Tropospheric Research was not included in the model (no air pollution data). Additionally, the Leipzig-Mitte station was not included in the NO 2 model because Spearman correlation coefficients were greater than 0.7. BC = black carbon; PM 2.5 = particulate matter <2.5 mmi n aerodynamic diameter.

ORIGINAL ARTICLE
1340 American Journal of Respiratory and Critical Care Medicine Volume 207 Number 10 | May 15 2023 respiratory mortality slightly decreased but
remained robust. Finally, our results are
consistent with those from the extensively
studied, highly polluted area of Erfurt,
Germany, where the initial epidemiological
short-term studies in the 1990 s found
evidence of an association between UFPs and
cardio-(respiratory) mortality (10, 11, 23).

Three combined main pathways are
thought to promote the adverse health effects
of particulate air pollution, and especially
UFPs, on health (24). First, smaller particles,
and especially UFPs, can translocate from the
alveolar space by entering the endothelial
cells and the lung interstitium. It has been
demonstrated that they translocate to
epithelial cells and eventually into the
circulation, potentially causing direct adverse
effects along the way (24, 25). When in the
blood, they can reach other lung areas and
distant nonpulmonary regions and organs.

As a result of their large surface area relative
to the unit mass and their surface reactivity,
chemical compounds can be more easily
absorbed and transported, leading to further
damage (7). Second, a series of subclinical
systemic reactions can be induced from the
lung, e.g., the release of proinflammatory and
prooxidative mediators (24, 26). These can
lead to local and systemic inflammatory
processes and trigger prothrombotic effects, a
procoagulation state, and epithelial and
endothelial dysfunction (24, 26). Third,
particles that deposit in the pulmonary tree
can directly stimulate neuronal reflexes,
leading to changes in pulmonary and cardiac
autonomic regulation (24). These alterations
in autonomic tone involve multiple reflex
arcs and are often the most immediate
response to exposure to air pollution (27).

Although epidemiological studies to date can
only provide suggestive evidence on mortality,
clinical relevance is given because UFP effects
may induce endpoints such as impaired lung
function (28) and systemic inflammation (12)
or affect morbidity, particularly respiratory
health in younger people (13).

Few studies reported on potential effect
modification (e.g., age, sex, or season),
showing mixed results. Findings from one
systematic review (12) and two
aforementioned short-term analyses (20, 22)
observed increased UFP effects in the
warmer season. In contrast, the results
reported by Lanzinger and colleagues (21)
and the present study indicated a slightly
increased risk in the cold season. A possible
explanation could be a different exposure
m i x t u r eo ras m a l l e ri nfluence of Germany’s
more temperate climatic conditions (22).

Higher risk estimates for elderly people have
been reported previously (20, 21). We
observed no significant effect modification
by age, although higher risks were observed
i nt h ey o u n g e ra g eg r o u p .O u rs t u d ys h o w e d that women had a significantly increased risk
of respiratory mortality. Similar, although
insignificant, findings were reported by
Lanzinger and colleagues for respiratory
mortality (21) and by Stafoggia and
colleagues for natural mortality (20).

Differences in air pollution effects between
men and women have been extensively
studied, although thefindings remain
uncertain, and some studies reported larger
effects in women (29). Several factors have
been hypothesized that could affect and
explain these differences. Biological (i.e., sex)
factors could include, for example, different
levels of hormones and cytokines (e.g., high-
sensitivity C-reactive protein as a marker for
systemic inflammation) or a higher total
deposition fraction of UFPs in the lungs of
women (30– 32). On the contrary,
socioeconomic (i.e., gender) factors could
explain different underlying exposure
patterns, societal roles, and health behavior
in general (e.g., differences in smoking
prevalence or physical activity) (33). For
respiratory deaths, further examination of
our data showed that women were
substantially older (more were aged>85 yr)
than men, that there were no major
differences in underlying causes of death or
by station or city, and that the effects did not
change when the analysis was further
stratified by particle size fraction (data not
shown). However, the results did not change
when the analysis was additionally stratified
Figure 5. Percent change in relative risk and 95% confidence interval per interquartile range (IQR) increase in concentration of particles in the ultrafine range (10–100 nm; UFPs) for respiratory mortality (5–7 d after UFP exposure). The x-axis shows the results of the main model (dots) and different sensitivity analysis (rectangles). The y-axis represents the percent change of risk per IQR increase in air pollution concentration (difference between the 75th and 25th percentiles; corresponds to the spread of the middle 50% of the data). Standardization by IQR facilitates comparison between different pollutants. All estimates represent the pooled analysis of the measurement stations using multilevel random- effects models and were adjusted for main model covariates. App. Temp. = Apparent Temperature; Baro. Press. = Barometric Pressure; DF = Degrees of Freedom; Meteo. = Meteorology.

ORIGINAL ARTICLE
Schwarz, Schneider, Cyrys, et al. : Ambient Ultrafine Particles and Mortality 1341 by age. A more detailed investigation of other
potentially influential factors was not
possible with the available data set, and any
causal conclusions would be highly
speculative, beyond the scope of the present
paper, and unable to be adequately
supported by the evidence from our analysis
b e c a u s eo ft h ea b s e n c eo fi m p o r t a n tv a r i a b l e s and the study design itself. In summary, even
though the observed associations may be
partially explained by sex and gender
differences, larger data sets and prospective
longitudinal analyses that explicitly address
sex and gender differences in UFPs are
needed to clarify ourfindings further.

An ongoing debate concerns whether
the effects of UFPs occur independently from
PM

2.5. Different sources, temporal-spatial
patterns, and atmospheric urban
environments result in almost no
relationship between UFPs and PM

2.5 and
limited representativeness between the two
quantities (34). UFPs are assumed to be
more associated with traffic-related air
pollutants such as nitrogen oxides, carbon
monoxide, and BC (8). We found evidence
for independent effects of UFPs after
adjustment for PM

2.5 or BC. The inclusion of
NO2 l e dt om o r ei m p r e c i s ea n di n s i g n ificant results. Similar to our results, a recent review
concluded that NO2 adjustment had greater
effects on the point estimates than
adjustments for other pollutants (12). High
correlations between UFPs and NO
2 could
lead to multicollinearity or methodological
issues, resulting in unstable models and
biased effect estimates (12). As a result, there
remains uncertainty about independent
e f f e c t sw h e nU F P sa r ea d j u s t e df o ra d d i t i o n a l NO
2 coexposure (which may originate from
similar sources, e.g., traffic emissions), but
also vice versa (5, 35). Source-specifica n d
chemical-composition analyses included in
the context of large epidemiological studies
could help to further clarify this issue. In
addition, spatiotemporal modeling of short-
term UFP exposures or the inclusion of
multiple monitoring stations per area unit
could contribute to a more comprehensive
estimate of population-representative UFP
exposures. Accounting for the high spatial
variability remains challenging and will
require a greater focus in the future,
especially for long-term studies. Although
quantification of UFP risk based on number
concentration was recommended by the
WHO in 2021, there is still no national or
international consensus on what constitutes
the most important dimension of UFPs, and
standard methods still do not exist (14).

Furthermore, without adequate
characterization of the UFP source or
chemical composition, it remains unclear
whether the effects are the result of UFP
number concentrationper se or represent a
marker of combustion PM. Comparing the
health effects and related biological
plausibility of different UFP exposure
metrics (e.g., particle number or surface area)
and a more detailed characterization of UFPs
( w i t hr e g a r dt ot h e i rs o u r c e sa n dc h e m i c a l composition) would contribute to a better
and more holistic picture of UFP risk. For
example, Schmid and Stoeger have identified
surface area as a highly relevant
biological/toxicological dose metric because
i tm a yb e t t e rr e p r e s e n tt h ea r e aw h e r e molecules on the particle surface interact
with body tissues orfluids (36). However,
depending on the mode of action, other
metrics could be more biologically effective
f o rh e a l t h ,s oa e r o s o le x p o s u r em o n i t o r i n g should optimally include multiple dose
metrics simultaneously (36). In our analysis,
we had the opportunity to assess the link
between UFPs and mortality using
background and traffic-related stations.

UFPs exhibited higher risks for respiratory
mortality at urban background stations; NO
2 exhibited higher risks at traffic-related
stations. Contrarily, traffic-related stations
showed higher risks for natural mortality
after UFP exposure. Generally, urban
background stations are considered to better
reflect the exposure concentrations of the
average urban population. Nevertheless, we
included traffic-related stations to better
capture daily peak concentrations and to
evaluate potential differences on an
exploratory basis. This differentiation may
also be valuable for future research to
understand more about the differences in
risk between station types (and also within
cities) in the context of prevailing exposures
(e.g., NO
2). In addition, to date, regulatory
air-quality monitoring has focused on mass
concentrations offine PM (e.g., PM2.5 and
PM10)a n dg a s e o u sp o l l u t a n t s( e . g . ,N O2 and O3), inherently assuming that the health effects
of UFPs are well represented by monitoring
these pollutants. Given the growing body of
literature and ourfindings, it would not be
sufficient to use current monitoring standards
to assess PM risk adequately.

Strengths and Limitations
Our study represents a carefully designed
multicity study over eight consecutive years
with a harmonized exposure design for all
included stations and standardized outcome
data collection. The monitoring stations and
equipment were incorporated into the
German multiinstitutional GUAN, which
ensured routine maintenance and
standardized calibration processes to
measure particle number size distribution
through its operators. In addition, for two
cities, we included traffic-related stations in
the pooled analysis to also capture the effects
of peak concentrations to better represent the
exposure situation in these urban areas.

Therefore, we are among thefirst to compare
different risk estimates between these
two exposure settings in a multicity
epidemiological context. We thoroughly
adjusted for meteorological variables and
time trends to rule out the possibility that the
detected associations resulted from
meteorological influences or seasonal
differences. Additional sensitivity analyses
indicated that ourfinal effect estimates
seemed to be conservative and robust to
variations in the models.

This study has several limitations to
acknowledge. First, we did not have source-
specific information on particles. As a result,
we could only assume potential sources using
different size modes. Second, unlike PM

2.5,
UFPs have been reported to exhibit high
variability in space (6), which might lead to
exposure misclassification (or measurement
error), especially when a single station is used
to represent the exposure risk for an entire
city (6). In this study, we did not statistically
correct for possible measurement error, so
the effect estimates may be affected toward
or away from the null (37). However, we
included additional stations for Dresden and
Leipzig to better capture the spatial variation
in UFP concentrations. Moreover, Cyrys and
colleagues (38) have shown, for Augsburg,
Germany, that a carefully chosen urban
background station can adequately capture
the temporal variation of UFPs across the
city. However, we included only six stations
from three cities, so our analysis may lack
statistical power (e.g., when comparing risks
between station types). Third, we performed
several analyses and cannot rule out the
possibility that some results were observed
by chance. In addition, despite careful model
selection, residual confounding could be
present, especially for additional NO
2 adjustment, because the real-world exposure
environment is a complex mixture of
particles and gases that may originate from
the same sources. Fourth, the number of

ORIGINAL ARTICLE
1342 American Journal of Respiratory and Critical Care Medicine Volume 207 Number 10 | May 15 2023 deaths, especially respiratory mortalities, was
rather low, and the resulting effect estimates
might be affected, especially when examining
effect modification. Nevertheless, the CIs
generally did not show large uncertainty, and
the inclusion of additional respiratory deaths
from the Ruhr area did not substantially
change the results. Last, our study was
conducted only in Germany, so the results
may not be easily transferable to other
regions. Different meteorological or climatic
conditions can affect the concentration of
pollutants in the environment. For example,
wind speed and rain can lead to dilution or
leaching of particles in the air. In addition,
new particle formation from precursor
substances of UFPs can occur in areas with
high solar irradiation. This highlights the
need for multicity studies with different
meteorological or climatic conditions.

Conclusions
In summary, the pooled results of our time-
series study indicated an increased risk for
respiratory mortality after exposure to
UFPs. In particular, delayed effects were
seen for multiday averages and
corroboratedfindings from high-pollution
settings. No consistent associations were
found for cardiovascular or natural
mortality. The study highlights that longer
time series with more monitors per city of
high-quality UFP measurements are needed
to overcome the inconsistency in the
available evidence. It also highlights that
multiple measurements with a classification
of particle size fractions, chemical
composition, and emission source are
needed to further substantiate the impact of
UFPs as called for by the good practice
statements for UFPs published by the
WHO in 2021 (14). In general, focusing our
policies on eliminating combustion might
be the most health-protective air pollution
mitigation approach./H17039
Author disclosures are available with the
text of this article at www.atsjournals.org.

Acknowledgment: The authors thank Karen
Meyer (Research Data Centre [RDC] of the
Federal Statistical Office and Statistical Offices
of the Federal States, F€urth, Germany) and
Dana Leenders (IT service provider of North
Rhine-Westphalia, D€usseldorf, Germany) for
support with the mortality data; Christoph
Asbach and Heinz Kaminski (Institute of Energy
and Environmental Technology [IUTA],
Duisburg, Germany) and North Rhine-
Westphalia State Agency for Nature,
Environment and Consumer Protection
(LANUV) (Essen, Germany) for providing
exposure data for M€ulheim-Styrum; Maik
Merkel (Leibniz Institute for Tropospheric
Research [TROPOS], Leipzig, Germany) for
providing data of Leipzig-TROPOS; Andrea
Hausmann and Gunter L€oschau (Saxon
State Office for Environment, Agriculture and
Geology) for advice with the study concept;
Simonas Kecorius (Helmholtz Munich) for
helpful discussions on the topic; and the
editors and unknown reviewers for their
thoughtful comments and suggestions on
the manuscript. Source for mortality data:

RDC of the Federal Statistical Office and
Statistical Offices of the Federal States
(Mortality Statistics [uniform directory
number: 23211]; survey years, 2010–2017;
DOI: 10.21242/23211.2010.00.00.1.1.0 to

10.21242/23211.2017.00.00.1.1.0, own
calculations).

References

1. Atkinson RW, Kang S, Anderson HR, Mills IC, Walton HA. Epidemiological time series studies of PM2.5 and daily mortality and hospital admissions:
a systematic review and meta-analysis. Thorax 2014;69:660–665.

2. Brunekreef B, Strak M, Chen J, Andersen ZJ, Atkinson R, Bauwelinck M, et al. Mortality and morbidity effects of long-term exposure to low-level PM

2.5, BC, NO2, and O3: an analysis of European cohorts in the ELAPSE Project. Res Rep Health Eff Inst 2021;208:1–127.

3. Chen J, Hoek G. Long-term exposure to PM and all-cause and cause- specific mortality: a systematic review and meta-analysis.Environ Int 2020;143:105974.

4. Orellano P, Reynoso J, Quaranta N, Bardach A, Ciapponi A. Short-term exposure to particulate matter (PM10 and PM2.5), nitrogen dioxide (NO2), and ozone (O3) and all-cause and cause-specific mortality: systematic review and meta-analysis. Environ Int 2020;142:105876.

5. Forastiere F, Peters A. Invited perspective: the NO2 and mortality dilemma solved? Almost there! Environ Health Perspect 2021;129:121304.

6. HEI Review Panel on Ultrafine Particles. Understanding the health effects of ambient ultrafine particles. HEI perspectives 3 [serial online]. Boston, MA: Health Effects Institute; 2013 [accessed 2022 Dec 8]. Available from: https://www.healtheffects.org/publication/understanding-health- effects-ambient-ultrafine-particles.

7. Kwon HS, Ryu MH, Carlsten C. Ultrafine particles: unique physicochemical properties relevant to health and disease. Exp Mol Med 2020;52:
318–328.

8. Cassee F, Morawska L, Peters A, Wierzbicka A, Buonanno G, Cyrys J, et al. White Paper: ambient ultrafine particles: evidence for policy makers. Pfintal, Germany: European Federation of Clean Air and Environmental Protection Associations; 2019 [accessed 8 Dec 2022].

Available from: https://efca.net/files/WHITE%20PAPER-
UFP%20evidence%20for%20policy%20makers%20(25%20OCT).pdf.

9. Vu TV, Delgado-Saborit JM, Harrison RM. Review: particle number size distributions from seven major sources and implications for source apportionment studies.Atmos Environ 2015;122:114–132.

10. Breitner S, St €olzel M, Cyrys J, Pitz M, W €olke G, Kreyling W, et al. Short- term mortality rates during a decade of improved air quality in Erfurt, Germany. Environ Health Perspect 2009;117:448–454.

11. Wichmann HE, Spix C, Tuch T, W €olke G, Peters A, Heinrich J, et al.

Daily mortality and fine and ultrafine particles in Erfurt, Germany part I:
role of particle number and particle mass. Res Rep Health Eff Inst 2000;98:5–86.

12. Ohlwein S, Kappeler R, Kutlar Joss M, K €unzli N, Hoffmann B. Health effects of ultrafine particles: a systematic literature review update of epidemiological evidence.Int J Public Health 2019;64:547–559.

13. Samoli E, Rodopoulou S, Schneider A, Morawska L, Stafoggia M, Renzi M, et al. Meta-analysis on short-term exposure to ambient ultrafine particles and respiratory morbidity.Eur Respir Rev 2020;29:200116.

14. World Health Organization. WHO Global Air Quality Guidelines:
particulate matter (PM2.5 and PM10), ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide. Bonn, Germany: WHO European Centre for Environment and Health; 2021 [accessed 8 Dec 2022]. Available from: https://apps.who.int/iris/handle/10665/345329.

15. Schwarz M, Schneider A, Cyrys J, Bastian S, Peters A, Breitner S.

Impact of ultrafine particles and total particle number concentration on natural-, cardiovascular, and respiratory mortality in three German cities. ISEE Conference Abstract. Environ Health Perspect 2022; 2022.

Available from: https://ehp.niehs.nih.gov/doi/abs/10.1289/isee.2022.

O-OP-282.

16. Stafoggia M, Samoli E, Alessandrini E, Cadum E, Ostro B, Berti G, et al.; MED-PARTICLES Study Group. Short-term associations between fine and coarse particulate matter and hospitalizations in Southern Europe:
results from the MED-PARTICLES project. Environ Health Perspect 2013;121:1026–1033.

17. Sera F, Gasparrini A. Extended two-stage designs for environmental research. Environ Health 2022;21:41.

18. Birmili W, Weinhold K, Rasch F, Sonntag A, Sun J, Merkel M, et al. Long- term observations of tropospheric particle number size distributions and equivalent black carbon mass concentrations in the German Ultrafine Aerosol Network (GUAN). Earth Syst Sci Data 2016;8:355–382.

ORIGINAL ARTICLE
Schwarz, Schneider, Cyrys, et al. : Ambient Ultrafine Particles and Mortality 1343

19. Sun J, Birmili W, Hermann M, Tuch T, Weinhold K, Spindler G, et al.

Variability of black carbon mass concentrations, sub-micrometer particle number concentrations and size distributions: results of the German Ultrafine Aerosol Network ranging from city street to High Alpine locations. Atmos Environ 2019;202:256–268.

20. Stafoggia M, Schneider A, Cyrys J, Samoli E, Andersen ZJ, Bedada GB, et al.; UF&HEALTH Study Group. Association between short-term exposure to ultrafine particles and mortality in eight European urban areas. Epidemiology 2017;28:172–180.

21. Lanzinger S, Schneider A, Breitner S, Stafoggia M, Erzen I, Dostal M, et al.; UFIREG study group. Associations between ultrafine and fine particles and mortality in five central European cities— results from the UFIREG study. Environ Int 2016;88:44–52.

22. Hennig F, Quass U, Hellack B, K €upper M, Kuhlbusch TAJ, Stafoggia M, et al. Ultrafine and fine particle number and surface area concentrations and daily cause-specific mortality in the Ruhr Area, Germany, 2009-2014. Environ Health Perspect 2018;126:
027008.

23. St€olzel M, Breitner S, Cyrys J, Pitz M, W €olke G, Kreyling W, et al. Daily mortality and particulate matter in different size classes in Erfurt, Germany. J Expo Sci Environ Epidemiol 2007;17:458–467.

24. R €uckerl R, Schneider A, Breitner S, Cyrys J, Peters A. Health effects of particulate air pollution: a review of epidemiological evidence.Inhal Toxicol 2011;23:555–592.

25. Oberd€orster G, Oberd€orster E, Oberd€orster J. Nanotoxicology: an emerging discipline evolving from studies of ultrafine particles.Environ Health Perspect 2005;113:823–839.

26. Brook RD, Rajagopalan S, Pope CA III, Brook JR, Bhatnagar A, Diez- Roux AV, et al.; American Heart Association Council on Epidemiology and Prevention, Council on the Kidney in Cardiovascular Disease, and Council on Nutrition, Physical Activity and Metabolism. Particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American Heart Association.Circulation 2010;121:2331–2378.

27. Perez CM, Hazari MS, Farraj AK. Role of autonomic reflex arcs in cardiovascular responses to air pollution exposure.Cardiovasc Toxicol 2015;15:69–78.

28. Peters A, Wichmann HE, Tuch T, Heinrich J, Heyder J. Respiratory effects are associated with the number of ultrafine particles. Am J Respir Crit Care Med 1997;155:1376–1383.

29. Clougherty JE. A growing role for gender analysis in air pollution epidemiology.

Environ Health Perspect 2010;118:167–176.

30. Pilz V, Wolf K, Breitner S, R €uckerl R, Koenig W, Rathmann W, et al.; KORA-Study group. C-reactive protein (CRP) and long-term air pollution with a focus on ultrafine particles. Int J Hyg Environ Health 2018;221:510–518.

31. Prisby RD, Muller-Delp J, Delp MD, Nurkiewicz TR. Age, gender, and hormonal status modulate the vascular toxicity of the diesel exhaust extract phenanthraquinone.J Toxicol Environ Health A2008;71:
464–470.

32. Sturm R. Total deposition of ultrafine particles in the lungs of healthy men and women: experimental and theoretical results. Ann Transl Med 2016;4:234.

33. Shin HH, Parajuli RP, Gogna P, Maquiling A, Dehghani P. Pollutant-sex specific differences in respiratory hospitalization and mortality risk attributable to short-term exposure to ambient air pollution. Sci Total Environ 2021;755:143135.

34. de Jesus AL, Rahman MM, Mazaheri M, Thompson H, Knibbs LD, Jeong C, et al. Ultrafine particles and PM2.5 in the air of cities around the world: are they representative of each other? Environ Int 2019;129:
118–135.

35. U.S. Environmental Protection Agency. Integrated Science Assessment (ISA) for oxides of nitrogen— health criteria. 2016 [updated 12 Dec 2022; accessed 15 Jan 2023]. Available from: https://www.epa.gov/isa/ integrated-science-assessment-isa-oxides-nitrogen-health-criteria.

36. Schmid O, Stoeger T. Surface area is the biologically most effective dose metric for acute nanoparticle toxicity in the lung. J Aerosol Sci 2016;99:
133–143.

37. van Smeden M, Lash TL, Groenwold RHH. Reflection on modern methods: five myths about measurement error in epidemiological research. Int J Epidemiol 2020;49:338–347.

38. Cyrys J, Pitz M, Heinrich J, Wichmann HE, Peters A. Spatial and temporal variation of particle number concentration in Augsburg, Germany. Sci Total Environ 2008;401:168–175.

ORIGINAL ARTICLE
1344 American Journal of Respiratory and Critical Care Medicine Volume 207 Number 10 | May 15 2023 EXHIBIT 8
Aerosol Science and Technology, 45:1123-1134, 2011
Copyright © American Association for Aerosol Research
ISSN: 0278-6826 print/ 1521-7388 online
DOI: 10.1080/02786826.2011.582194
Very Fine and Ultrafine Metals and lschemic Heart Disease in the California Central Valley 1: 2003-2007
Thomas A. Cahill, 1,2 David E. Barnes,1 Nicholas J. Spada,1 Jonathan A. Lawton,1 and Thomas M. Cahi113
1DELTA Group, University of California, Davis, California, USA 2The Health Effects Task Force, Breathe California of Sacramento-Emigrant Trails, Sacramento, California, USA
3Division of Mathematical and Natural Sciences, Arizona State University West Campus, Phoenix, Arizona, USA
The enhancement of mortality associated with cardiovascular and specifically ischemic heart disease (IHD) has been observed in the southern California Central Valley since at least 1990, and it continues to be a major source of mortality. While there is a strong statistical association of IHD with wintertime PM2.5 mass, the causal agents are uncertain. Medical studies identify some po­ tential causal agents, such as very fine and ultrafine metals, but they have not been fully characterized in most Central Valley re­ gions. To provide improved information on specific and potentially causal agents, a five site aerosol sampling transect was conducted from Redding to Bakersfield during a 17-day period of strong stag­ nation, January 5-22, 2009. Mass and elemental components were measured every 3 h in eight particle size modes, ranging from 10 to 0.09 µ,m, while the ultrafine particles ( <0.09 µ,m) were col­ lected on Teflon filters. Ancillary studies were performed including direct upwind-downwind profiles across a heavily traveled sec­ ondary street near a stoplight.

Very fine and ultrafine iron, nickel,
copper, and zinc were identified as vehicular, with the most prob­ able sources being brake drums and pads and the lubrication oil additive zinc thiophosphate. High correlations, many with r2 > Received 7 August 2010; accepted 27 November 2010.

The authors first wish to gratefully acknowledge the support of the Resources Legacy Fund of Sacramento, California, for encouragement and financial support for this study.

The staff of all of the Air Resources Board (ARB) sampling sites were key to successful execution of the project.

The authors also wish to gratefully acknowledge the financial sup­ port of the Sacramento Metropolitan Air Quality Management District under its CEO, Larry Greene.

The authors wish to acknowledge the input and oversight of the Health Effects Task Force (HETF) of Breathe California of Sacramento­ Emigrant under J ananne Sharpless, Chair; Betty Turner, consultant to the HETF, and especially members Ralph Propper and Earl Withy­ combe. Helpful suggestions and review were provided by William Wilson, US Environmental Protection Agency (EPA).

Address correspondence to Thomas A. Cahill, DELTA Group, Uni­ versity of California-Davis, One Shields Ave., Davis, CA 95616, USA.

E-mail: [email protected]

0.9, were found between these vehicular metals and IHD mortality, enhanced by the meteorology, terrain, and traffic patterns of the southern Central Valley. The braking systems of cars and trucks must now be considered along with direct exhaust emissions in estimating the health impacts from traffic.
[Supplementary materials are available for this article. Go to the publisher's online edition of Aerosol Science and Technology to view the free supplementary files.]

INTRODUCTION
The task of determining the causal factors that lie behind the statistically sound association of mortality with aerosols con­ tinues to labor under severe difficulties. Reliance on Federal Reference Method (FRM) methodologies, even if enhanced by periodic compositional analysis, does not provide enough in­ formation to allow health research scientists to identify causal factors (Lippmann 2009). Intensive research studies, which can provide such data, are so expensive as to limit their scope in space and time and thus pose problems with statistical signifi­ cance. An alternative is to identify regions with persistent and specific health impacts of suspected aerosol origins and then augment monitoring data and special studies to provide a suite of data from which causal factors may be identified. However, other factors such as meteorology, land use, and socioeconomic factors can often weaken the significance (Pope et al. 1995).

The California Central Valley, however, is a region of such ho­ mogeneity that such efforts there might be fruitful.

The southern part of the California Central Valley, the San Joaquin Valley, is the largest contiguous area of the United Sates in serious violation of both ozone (summer) and PM2.s (winter) ambient air quality standards. While the summer ozone levels are relatively uniform over the Central Valley, the winter PM 10 and PM2.5 concentrations vary strongly, from low in the northern Sacramento Valley to high in the southern San Joaquin Valley.

During individual winter stagnation episodes, the south to north 1123
Cahill -- 00000286
1124 T. A. CAHILL ET AL.
(Bakersfield to Redding) mass ratio is up to a factor of 8 for PM10, and a factor of 5 for PM2.s.

In addition to improving monitoring data, a series of ex­ tensive research studies by and through the California Air Re­ sources Board (ARB), such as California Regional Particulate Air Quality Study (CRPAQS 2001), Central Coast Ozone Study (CCOS), and Fresno Asthmatic Children's Environment Study (FACES), were initiated in 1999 through 2001 that greatly added to our knowledge of San Joaquin Valley aerosols. However, most field data were taken before the current understanding of aerosols and cardiovascular impacts was achieved (Devlin 2003, Lippmann 2009) and thus lack some critical measurements, es­ pecially ultrafine metals (Chow et al. 2006, 2007, 2008; Chen et al. 2007; Kleeman et al. 2009). One of the major results of these studies is an understanding that most of the PM2.s mass in the winter in the San Joaquin Valley is ammonium nitrate and wood smoke, neither of which are a known cardiovascular threat.

The mortality data in the entire Central Valley are relatively uniform for three of the four major causes of mortality, pul­ monary disease, stroke, and cancer, but the 4th and most impor­ tant, heart attacks from all causes, is higher by about 20% in the central and extreme southern California Central Valley [Cali­ fornia Department of Health Services (CA DHS) 2010] and es­ pecially in Bakersfield. For one specific and major component of heart disease, ischemic heart disease (IHD), the incidence rises to about 35% (CA DHS 2010). In this research, we have measured some of the components of aerosols found by recent health studies most likely to be causal factors in this statistically strong association to cardiovascular disease, namely very fine and ultrafine metals, and size and compositionally-resolved or­ ganics (Cahill 2010), over the entire Central Valley, with the goal to enhance data from the earlier studies and provide additional material for future epidemiological analyses.

EXPERIMENTAL METHODS
Study Region Geography and Climatology
The Central Valley of California (Figure 1) has remarkable uniformity in terrain and meteorology. It is one of the richest agricultural areas in the world, with two major metropolitan areas, the Sacramento-Stockton and Fresno, with smaller cities of a few 100,000 to small towns spread throughout the Valley.

The Valley is a major transportation corridor for road and rail traffic but generally lacks strong industrial sources. There are active areas of oil extraction near and north of Bakersfield. The Central Valley offers an ideal situation to examine the effects of pollutants on health since many potential confounding factors, such as meteorology, land use patterns, socio-economic factors, and cigarette smoking, are very similar throughout the Valley (Cahill et al. 1998; note that the values in that report are slightly different in the present report since we have now available CORE Report #2 (1996), while the original report used CORE Report #1 (1994)).

Saa-am.mt.a
Valley
region
• N
1---1
100 lar.

FIG. 1. Map of the Central Valley of California. The northern part is referred to as the Sacramento Valley while the southern part is called the San Joaquin Valley. The areas within the dotted circumference vary in altitude from almost sea level in the Delta to a few hundred feet in elevation. The small arrows denote the typical winter wind patterns (Hayes et al. 1984 ). The cities marked were used in this study.

In summer, high temperatures in the Valley (mean daily high, July, 36°C ± 1 °C) lower the surface pressure and draw massive inputs of marine air into the Valley from the San Francisco Bay, bringing with it oceanic aerosols and typical urban pollutants enhanced by the major petrochemical industrial sites at the Car­ quinez Strait, between Sacramento and San Francisco. These winds often funnel south down the entire San Joaquin Valley, south of the strait, but less often funnel north up the entire Sacramento Valley, north of the strait. Rainfall in the Valley is rare during the period from May through October (Hayes et al. 1984).

In winter, the Central Valley air basin tends to become stag­ nant due to its flat valley floor and circumferential mountains, except at the Straits (Figure 1 ). There is slow drainage of cold air from the Sierra Nevada Mountains each night to the Valley floor (mean daily low, January, 3°C ± 1 °C). The air drains slowly, at the rate of about 50 km/day, from both the northern and southern ends of the valley towards the lowest point, the Sacramento-San Joaquin Delta, southwest of Sacramento and directly east of the Carquinez Strait. This pattern results in persistent stagnation episodes that are periodically interrupted by synoptic storms.

Typical Pollution Patterns
The annual pattern of PM2.s mass reflects the effects of the winter stagnation, with elevated levels from about November through February (Figure 2). All valley sites have similar annual Cahill -- 00000287
METALS AND ISCHEMIC HEART DISEASE 1 1125
90 ... 80
"' ~ 70 E
:S 60
:::i
~ 50
"' Q,
"' 40 E
~ 30
:: <.>
:i
20 10
0 ~il
2 3 4 5 6 7 8 9 10 11 11 12
2 3 4 5 6 6 7 8 9 10 11 12
Monlh of lhe year 2002
FIG. 2. Annual pattern of PM2.s mass as seen at Fresno. patterns, high in winter and low in summer. The PM2.s peak seen in August was from a forest fire in Oregon. While the southern reaches of the Central Valley have serious violations of federal and state mass standards (ADAM 2010), other parts of the valley that share agriculturally based land uses and meteorology do not exhibit serious violations of these standards. For PM 10 mass, the ratio from Bakersfield in the south to Redding in the north can be as high as a factor of 9 in winter stagnation events, while for PM2.s, the ratio can be as high as a factor of 5 (ADAM 2010).

The pattern of winter air movement makes the high aerosol mass levels at Bakersfield difficult to explain, since the nighttime airflow into Bakersfield is from the mountains to the south, southeast, and east of the city. There is essentially no influence from the Los Angeles basin due to the blocking mountains. The air at Bakersfield then flows slowly north towards Fresno, which has generally lower PM2.s mass levels despite greater population and industry.

Epidemiological Data
Greatly improved data on mortality and morbidity became available circa 2001 from the California Department of Health Services (CA DHS 2010), which allow detailed analyses of death by hundreds of cause on an annual basis. These data show a persistent enhancement, 2003-2007, of IHD mortality of roughly 50% in the central region and at Bakersfield, at the extreme southern edge of the San Joaquin Valley, as compared with the northern valley (Figure 3).

Aerosol Sampling Methods
Aerosol samples were collected along a transect in the Cen­ tral Valley to investigate the causal factors behind the increased mortality due to IHD in the Central Valley of California. Since routine air quality measurements were available in this region, we collected and analyzed aerosols not routinely measured by existing monitoring programs that were known or suspected to be causal factors in cardiovascular disease. Further, we selected a period of stagnation that maximizes winter particulate mass 2!10
y= 00732x + 129.48
r2 = 0 2614
l
□------------------------(l 100 200 300 400 ~00 600 700 Distance from north end of Central Valley (km)
FIG. 3. IHD in the Central Valley of California, 2003-2007. Distances span from Redding in the north (0 km) through Chico (170 km), Sacramento (220 km), Fresno (400 km) to Bakersfield (680 km).
concentrations, avoiding the synoptic fronts and rain that pe­ riodically clean out the valley in winter. Using meteorological predictions, we simultaneously sampled continuously by size, time, and composition for 17 days starting on Jan 5, 2009, at five sites from the extreme north, Redding, to the extreme south, Bakersfield.

Rotating drum impactors (DELTA Group 8 DRUMs; Cahill
et al. 1985; Raabe et al. 1988) were used at all sites to provide continuous samples capable of 3-h data in eight impaction size modes from 10 to 0.09 µ,m diameter plus an integrating Teflon after filter. The impaction samples were collected onto Apiezon­ L greased Mylar stages (Wesolowski et al. 1977). Analyses were made for mass and elements from aluminum to molybdenum
plus lead by synchrotron-induced x-ray fluorescence (Bench et al. 2002). The quality assurance validations, including a year­ long inter-method comparison to ARB's FRM particular sam­ plers, are summarized in the supplemental information. For the ARB side-by-side over the entire year, agreement for PM2.s mass was better than ±10%, DRUM versus FRM, with the win­ ter agreement much better than that, a few percent, probably because of higher winter mass values. Note that 49 individual mass values from a DRUM with 3-h time resolution and 6 sub-

2.5-µ,m size cuts plus a <0.09 µ,m after-filter were required to match a single 24-h PM2.s mass value.

Aerosols destined for organic analyses were likewise col­ lected with DRUM samplers at the same time and locations as the elemental samplers. However, the organic samples were an­ alyzed averaging over the entire 17-day study, and the aerosols were collected onto fired aluminum substrates (Cahill 2010).

The study included three components, all conducted in win­ ter conditions and using the same equipment, including inte­ grating ultrafine Teflon filters: (1) an initial year-long study of the DRUM sampler side-by-side with the ARB's FRM at the Cahill -- 00000288
1126 T. A. CAHILL ET AL.

Sacramento test site at 13th and T Street to establish equiva­ lency of the sampling and analysis (Cahill and Barnes 2009), (2) a simultaneous transect across a heavily traveled secondary street to identify very fine and ultrafine aerosols from roadways, and (3) the main transect study in winter, 2009. Summaries of these studies are included in supplemental information.

For the transect study, samples were collected at five existing ARB and district monitoring sites covering the entire Central Valley (Figure 1 ):

Redding (roof of Health Department),
Chico (ARB Manzanita Avenue site),
Sacramento (ARB 13th and T Street site),
Fresno (ARB First Street site), and
Bakersfield (ARB California Street site).

Paired UC Davis DELTA Group 8 DRUM samplers were
used at all sites except Sacramento, which had to use a PM 2.s 3 DRUM for lack of equipment. One 8 DRUM was used at
each site for mass and elements (3-h time resolution), sampling onto lightly greased Mylar, and the other for organic matter (17- day average), sampling onto fired aluminum foils. All included identical Teflon ultrafine filters that integrated the entire 17-day period. The parameters measured at all sites include but are not limited to

1. time-averaged ultrafine particulate elemental composition, Al to Mo, plus lead, and mass, 0.09 > Dp > 0.0 µ,m aerody­ namic diameter,

2. time-dependent (3 h) mass and elements (see above) for the eight particle sizes 10.0-5.0, 2.5, 1.15, 0.75, 0.56, 0.34,

0.26-0.09 µ,m aerodynamic diameter,

3. time-averaged organic matter by size, 10.0-5.0, 2.5, 1.15,

0.75, 0.56, 0.34, 0.26---0.09 and <0.09 µ,m diameter, includ­ ing polycyclic aromatic hydrocarbons (PAHs), sugars (in­ cluding levoglucosan), fatty acids, and n-alkanes have been published (Cahill 2010).

RESULTS
Meteorology
After the sites were selected, deployment was delayed un­ til weather predictions indicated at least a 10-day period of low winds and stagnation, conditions that produce the highest PM2.s mass. The prediction was accurate and the meteorology was favorable and allowed us to extend the study to a 17-day period characterized by high mass values and persistent hazes (Table 1).

Aerosol Mass Values
PM10 mass values were available at all sites on January 19, while all sites except Redding had daily PM2.s mass values collected by the ARB and local agencies (ADAM 2010). The PM25 values were averaged over the entire 17-day study period ·--- P'M 10 mass. 1119 --+- Pll 211 man, 1119 -• · PM2_5 mass. study avg 100
90 1 80 E
.l:! 70 ..c ::,
u 60
iii so a.
.. 40 E
e 30 C)
e ,,
u 20 .,
i 10 II"
0 Redding
., .,
,.
r , , , , ,,
Chico Sai:ramento Fresno Bakersfield
Transect north to south
FIG. 4. PM2.s mass values were available every day, but only January 19th had PM10 mass at all valley sites. The high PM10 to PM2.s ratio on January 19 is unusual and unexplained.
and were typical of prior winter stagnation events (Table 2 and Figure 4 ). Looking in more detail, we can see that the north to south increase in average PM2.s mass is driven not as much by the increase in mass on a given day but the number of days that saw the high mass values.

Aerosol monitoring by state and local agencies with com­ positional data for PM2.s aerosols was only available at the Fresno site via the IMPROVE sampler (Malm et al. 1994 ).

The evolution of the aerosol event is shown in Figure 6, show­ ing that the study period would represent a winter day in vi­ olation of the PM2.s mass standard. The important role am­ monium nitrate plays during these aerosol episodes is clearly indicated.

Size-Resolved Mass and Inorganic Aerosols
For the period from January 5 through 22, over 6400 mea­ surements were made of mass and inorganic elements in nine size modes, measured every 3 h except for the integrating ul­ trafine filters. In addition, a simultaneous study (Cahill 2010) measured organic aerosols in four classes (alkanes, PAHs, .! It
l 10
1iH
il so
i 40
t ,e
~ 20
0 u 1tl
---- Bakersfield, - +- Fresno, _.,_ Sacramento, · • • Chico ' \
' , r
i o-~~~~~~~~~~~~~~~~~~~~...,....,
31 2 4 t 8: 11 12 14 16 18: 20 22 24 H
1 3 S 7 I 11 13 1 S 17 19 21 2S 25
January 2009
FIG. 5. PM2.s mass values before, during, and after the study period of January 5-22, 2009.

Cahill -- 00000289
METALS AND ISCHEMIC HEART DISEASE 1 1127
TABLE 1
Summary of meteorological parameters during sample collection period of January 5-22, 2009. The values reported are the average over the entire aerosol sampling period except for rainfall, which includes the rain on the days of setup and take down, January 5 and 22
Parameter Redding
Average temperature (°C) 11.3
Average high (°C) 19.3
Average low (°C) 3.0
Relative humidity(%) 56.4
Average wind speed (m/s) 1.92
Rain (cm) 1.2
Number of rain events 4
Number of fog events 0
Average visibility (km) 14.6
sugars, and organic acids) in nine size modes at all sites. In this article, we focus on those particles that have the best support from the health literature as having the potential for causing or exacerbating IHD. Thus, we focus on only those insoluble par­ ticles in the very fine (0.26-0.09 µ,m) and ultrafine ( <0.09 µ,m) modes that are able to penetrate deeply into the lung and pass into the cardiovascular system. However, before we focus on these particles, we need to examine the overall aerosol behavior important to eventually isolate the sources of these particles.

Using the well-studied Fresno site, we can see that the size distribution roughly matches expectations with a few differences (Figure 7). In Figure 7, three points stand out. First, the DRUM sampler with its coated substrates operated well, with negligible soil mass penetrating into the accumulation mode as shown by the calcium results. Second, there is a distinct deviation in the size profile of iron, which shows an enhancement in the 1.15-2.5 µ,m size mode not seen in other soil elements. Third, potassium becomes enhanced in the finest modes, clearly from non-soil sources. It is also much finer in size than typical wood smoke, which almost always peaks in the 0.34-0.75 µ,m mode.

Examining the non-soil iron anomaly (Figure 8), the size and 3-h time data show enhancement in the 2.5-1.15 µ,m mode each night not seen in other soil elements and not seen at any other valley site. The very fine iron often tracks the fine iron, but sharp deviations occur, indicating complex source behavior.

Chico Sacramento Fresno Bakersfield

9.6 9.0 9.4 10.6

15.5 16.1 15.4 16.5

3.7 1.7 3.2 4.4

71.6 73.4 71.8 66.4

1.27 0.47 0.62 1.17

2.2 2.5 1.2 0.1
3 3 2 2
2 11 2 2

12.3 7.1 6.9 6.6
Bakersfield lacks time information due to a system failure, and thus its behavior in time is unknown.

Further insight on zinc in Fresno can be obtained by examin­ ing zinc over most of a year (Cahill et al. 2003). The persistent presence of very fine zinc occurs throughout the entire year and becomes dominant in wintertime, closely associated with other vehicular aerosols. The zinc to very fine mass ratio was close to that seen in laboratory diesel tests (Zielinska et al. 2003).

Data on Very Fine and Ultrafine Aerosols
Very fine aerosols were collected on DRUM stage 8, 0.26 to

0.09 µ,m, and integrated over the 17-day study period to match the integrating ultrafine filters. Ultrafine aerosols were collected on the same 47-mm Teflon filters as the organic samples and av­ eraged over the same 17-day period. The filters were analyzed by scans of S-XRF every 0.5 mm across the filter diameter. All filters were uniform except for Chico, which had contamination on one edge. This was easily avoided by modifying the aver­ aging program. Two different averages were generated: one by summing the spectra, then analyzing and the other by averaging the typical eighteen individual elemental results.

In Table 3a, we show a summary time averaged DRUM very
fine data and, in Table 3b, the integrated ultrafine filter data for the study period.

TABLE 2
Mass values for PM10 on January 19 and PM2.s averaged from January 5 though January 22, 2009. The value of ~9 for PM2.s is estimated assuming the same PM10/PM2.s ratio as the other cities Site Redding Chico Sacramento Fresno Bakersfield
Size (µ,m) PM10 PM10 PM10 PM10 PM10
Mass (1/19) (µ,g/m3) 12 37 29 46 95
Size (µ,m) PM2.s PM2.s PM2.s PM2.s PM2.s
Mass average (µ,g/m3) ~9 27.3 30.2 40.9 51.0
Cahill -- 00000290
1128 T. A. CAHILL ET AL.
.......- PM,, mus (IMPROVE!, ·.., · Organic carbon, . ..,.. • PM,_, mass (ARBI -• Ammonium nitrate -- Elemental c,nbon
60 ,---------------------------,
.! ,,,,✓----,
• 50 E I,-,-=-,
~~41 ,.,,~ ✓ \. .? 15 40 :,
(J
g_ 30
..
~ 20 ;;,
§ 10
i
/ '\ /
··········••r-••·······················\·~,...~··· . -: ,----•, ~ ., ' ✓~ ' ~
✓ ' ......
.... __ - ·-::-~ .... ·- ' ... .,,, - -- ..... ,._ __ ..... ~ ... --·-~---lt•-·-·-·-irr" .... --·~ ~ ..... - ~ 4 7 10 13 16 19 22
Jenuery 2009
FIG. 6. IMPROVE and ARB PM2.s data at the Fresno site during the study.

Note the dominance of ammonium nitrate in the mass maxima.

The values in bold represent vehicular sources identified in the next section of this article. The copper ultrafine values are suspect from contamination by copper containing filter holders.

Knorr is non-soil potassium, corrected by the Calcium content, and a standard IMPROVE tracer of wood smoke (Malm et al. 1994). MDL is minimum detectable limit. Uncertainties are nominally ±5%, and detailed quality assurance and uncertain­ ties are in the supplemental information.

The correlations of sulfur (ammonium sulfate) and selenium are driven by the oil extraction and refining near Bakersfield (Figure 10). But there are strong associations between many metals and IHD in both the very fine and ultrafine modes. Al­ though mass levels in the very fine and ultrafine modes are low, the concentration of metals was significant. Thus metals are a major contributor to particle numbers, many presumably in insoluble compounds. Examining spatial trends for very fine metals with known vehicular origins, we see at Bakersfield a clear north to south enhancement of many very fine metals that far exceeds the x 1.6 increase in PM2.s mass for the same period (Figure 9).
·•- Potauh.1m, • • • Calcium, -- Iron
,eo
.!l 5110 .,
E
14 400
::,
u
~ JGO
"' E
f 280
Cl!
'
0 C
"' 180 :z '-.
•·-· 0
<0.09 l.2i - 0.3 ◄ l.5i - 0.75 1.15 - 2:.5 5.0 to - 10
!Ul9 - l.2i 0.3 ◄ - 0.56 0.75 - 1.15 2.5 - 5.0
Particle aerodynamic diameter micrometers
FIG. 7. Size distribution for Fresno aerosols.
.. .. -;
E
u
:i:i ::,
" ... .

Q.
• E . ..

G
0 C
• :z
1200
1000
800
800
-400
200
._ 2.5 - 1.1~ iron, • • • 0.26 - 0.0& iron x 100
5 6 7 8 9 10 11 12 13 1-4 15 16 17 18 19 20 21
January
FIG. 8. Fine and very fine nonsoil iron at Fresno. Recall that the study began directly after a rainfall event and had almost no winds to stir up dust. Thus we propose that most of the iron is roadway and vehicle derived.

The upward trend of sulfur must be interpreted in accord with the meteorology, which moves air from Bakersfield towards the Delta. Typical elements associated with vehicles include non­ soil iron, phosphorus, zinc, nickel, and copper. From the ratio present in the zinc thiophosphate in lubricating oil, there are clearly additional zinc sources present at all sites except perhaps Bakersfield.

In summary, there are significant contributions of metals in the very fine and ultrafine mode, many of which rise to their highest levels at Bakersfield (Figures 11 and 12) and correlate strongly with the IHD mortality data.

DISCUSSION
Sources of the Metals
The generation of metals in the ultrafine mode requires high temperatures and/or pressures. Thus, many sources of metals from vehicles, such as zinc from tire wear, are not candidates for sources ofultrafine aerosols. Two potential sources are engines, and especially diesel engines because of the high temperatures and pressures involved, and the braking systems in vehicles. .l! .

E 15
.!:!
..0 ::, u .. 10 .

0..
.
ti
15, 5
g . :z
Raddirtg Cllllico Sacreune..-0 Fre••o Bak.era1ield
Transect north to south
Nonsoil Iron
-,
Copper
•➔-1
Zmc
Phosphorus
-o~
FIG. 9. Very fine transition metals with known vehicular sources, zinc and phosphorus from zinc thiophosphate in lubricating oil, iron, copper, and zinc potentially from braking systems.

Cahill -- 00000291
METALS AND ISCHEMIC HEART DISEASE 1 1129
TABLE 3a
Aerosol mass data from the study and their correlation to IHD mortality. Uncertainties are nominally ±5%, and detailed quality assurance and uncertainties are in the supporting materials. Concentration (ng/m3) of very fine (0.09 to 0.26 µ,m) aerosols Site Redding Chico Sacramento Fresno Bakersfield Correlation r2 to IHD Ammonium sulfate 45 82 75
Chlorine <0.4 2.5 1.8
Phosphorus <0.1 0.4 0.8
Potassium (non-soil) 18.0 54.1 10.9
Vanadium <0.1 0.03 0.04
Chromium <0.1 O.Ql 0.01
Iron (non-soil) 2.2 1.8 1.0
Nickel <0.03 0.03 0.07
Copper 0.7 0.6 0.16
Zinc 1.9 3.9 1.1
Arsenic 0.15 0.2 0.07
Selenium <0.1 <0.1 0.09
Bromine 0.5 0.7 1.3
Lead 0.1 0.4 0.4
While there is considerable information on the total par­ ticulate emissions from vehicles, the information on ultrafine metals from vehicular exhaust is limited (Kleeman et al. 2000; Zielinska et al. 2003, 2004). The Zielinska et al.'s (2003) study involved detailed analysis of diesel exhaust. As part of the study, we analyzed for 32 elements collected in 11 tests with a 14-stage nano-MOUDI (lowest stage 0.01 µ,m). Sulfur, phosphorus and the metals zinc, calcium, and, in one of the 11 tests, lead, all peaked generally in the 0.10-0.056 µ,m modes (Zielinska et al. 2003). In addition to elemental and organic carbon, the tracer elements were assigned to sources: sulfur was from the fuel 120 580 0.98

0.7 2.6 0.22

0.7 4.0 0.98
45 32.7 0.00

0.2 <0.1 0.04

0.6 0.2 0.04

2.3 15.5 0.95

0.05 0.16 0.95

0.27 1.1 0.47

4.9 12.5 0.88

0.54 1.07 0.87

0.10 0.24 0.95

1.1 4.2 0.98

0.55 2.0 0.97
(CA low sulfur), the zinc and phosphorus from the zinc thio­ phosphate stabilizer in the lubricating oil, and calcium from an antacid additive. No other transition metals were seen in statis­ tically significant amounts.

A second potential source of very fine and ultrafine metals is the braking systems of cars and trucks, since high tempera­ tures and/or pressures are involved in this process. The brake drums have evolved from the massive, heat-conductive struc­ tures and generally well cooled in the past and now are often made of "grey iron," roughly 90% iron with the admixture of a few percent carbon, plus copper, silicon, and other metals in TABLE 3b
Concentration (ng/m3) of ultrafine ( <0.09 µ,m) aerosols Site Redding Chico Sacramento Fresno Bakersfield Correlation r2 to IHD Ammonium sulfate 122 215 270 470 999 0.95
Chlorine 2.5 5.1 0.6 3.0 0.5 -0.38
Phosphorus 1.3 0.6 4.0 3.3 10.5 0.96
Potassium 40.7 138.1 76.9 176.1 70.9 -0.05
(non-soil)
Vanadium 0.05 0.15 0.10 0.21 0.02 0.00
Chromium 0.13 0.04 0.75 0.28 0.32 0.03
Iron (non-soil) 3.9 3.9 14.3 7.7 27.7 0.88
Nickel 0.9 0.1 0.6 0.3 1.6 0.70
Copper na na na na na na
Zinc 8.6 11.2 14.3 22.4 12.3 0.00
Arsenic 0.41 0.64 1.1 2.6 1.5 0.12
Selenium 0.47 0.64 0.62 1.1 0.9 0.05
Bromine 11.2 12.4 13.7 25.3 18.9 0.16
Lead 1.7 2.2 2.5 3.6 2.6 0.06
na: not available.

Cahill -- 00000292
1130 T. A. CAHILL ET AL.
..... Potnslul'II, -- Sulfur (t'or amm.sulfat11. x 4.12t!} -- Chromium, ·-+ - Nickel, -- Lead, · ..__ Arsenic 300 4 --------------------~
2110
'.€ 200 ,,,
E
~ 1!10
g'
C
~ 100
CIO
0 L.__...J._ ___ -'---------'-----'-------'---'
R11ddlng Chico Sa~ra111•nto Fresno Bt1kersfi11ld
CA C•ntral Vall•y north to south transect
FIG. 10. Ultrafine potassium and sulfur.
smaller amounts, making them like brake pads an erodible sur­ face. Brake pads are far more complex in composition and are traditionally designed to erode.

Limited information is available on a complete inventory of vehicular expendables. An early one was developed as a part of the extensive California freeway lead and particulate studies in the 1970s. In Cahill and Feeney (1973), the UC Davis vehicle fleet was studied for all forms of expendables, including fuel, oil, brake wear, exhaust train erosion, tires, etc., by directly mea­ suring mass loss at replacement intervals. While many of the results are uncertain due to technological changes in the fuels and engines, measurements of brake drums and brake pads gave ~5 and 108 gm, respectively, at a replacement cycle of 58,400 km, or < 1 and 15 mg/km. However, these data were based on asbestos-containing brake pads and may not be reliable, which also impacts the US Environmental Protection Agency (EPA) PARTS model estimate of 20.6 mg/km, based on 1985 data.

In terms of brake drums, changes in brake drums from iron to "grey iron" containing graphite could materially increase that source of particles from prior values. Even with the uncertain­ ties, it is clear that brakes contribute a significant component of typical vehicle total emission rates. However, averages are mis­ leading because braking is only used for limited times. Thus the emission rates at stoplights and especially extensive mountain 30 2(1
i
20 E
~ 1!!
"' 0
C:
"' 10 z
5 0
---- Phosphorus, ·-+-- Iron (nonsoll~ ......,_ Zinc
'
• f
~·~- ,' / --~ --.///
-----······----R•dding Chieo Sa~rn,tnto Fresno Bakersfield CA Central Valley north to south transect
FIG. 11. Typical vehicular ultrafine metals.
j
C>
E 3 u
:.0
::I
u
ii 2 a. .,
E
! OI
0 C .. z
0 Redding Chico Sacramento Fresno Bakersfield
CA Central Valley north to south transect
FIG. 12. Additional ultrafine elements.
down grades will be much higher than these values. It is worth noting that traditional dynamometer studies and roadside side studies near freeways are likely to miss brake wear as an emis­ sion source, thus these emissions may be largely unrecognized by the majority of vehicle emission and assessment studies.

Information on brake pad composition was reviewed by
Kennedy et al. (2002) and Chan and Stachowiak (2004). The lat­ ter report, "A multitude of different brake pads have sprung onto the market in the post-asbestos brake pad revolution, each with their own unique composition," often proprietary. Brake pads comprise four subcomponents: (1) frictional additives, abra­ sives, and lubricants; (2) fillers; (3) a binder; and (4) reinforcing fibers for strength. The brake pads are described as metallic, semimetallic, with organic components, and organic. Asbestos to be used mainly as a reinforcing fiber in brake pads, and after its removal, a variety of materials have been used, including iron fibers. Some of them include potentially hazardous materi­ als potassium titanate and sepiolite, a magnesium-silicon com­ pound, both of which have potential health hazards. Organic aramaid fibers (e.g., Kelvar) and ceramics such as alumina and carbides are becoming more popular as prices decline. Fillers include a wide variety of organic (cashew hulls, old tires) and inorganic materials (barium sulfate is popular). The frictional components include graphite and metal sulfides including cop­ per, antimony, tin, zirconium, and lead. Copper in particular has been identified as a problem in aquatic systems, and legislation to remove it from brake pads has been passed in the states of Washington and California.

While there is enormous variability in the data, as shown by the percentile distributions, the four main elements are iron, copper, zinc, and nickel (Table 4). In summary, there is a wide variety of very fine and ultrafine metals that may arise from brake drums and pads, but iron and copper are clearly two major components.

Information on ambient and near source very fine and ultra­ fine metals was derived from two additional studies in Sacra­ mento. The first was a complete analysis ofultrafine ( <0.09 µ,m) Cahill -- 00000293
METALS AND ISCHEMIC HEART DISEASE 1 1131
TABLE 4
Summary of concentrations of elements in New Zealand
brake pads, from largest to smallest concentrations of
transition metals plus heavier elements. The distributions are shown to emphasize with wide variability between
manufacturers
Element (ppm) 10th percentile Median 90th percentile
Iron 11,700 18,300 190,000
Copper 29 5000* 116,000
Zinc 127 1630 37,400
Nickel 44 342 652
Manganese 143 315 1088
Lead 6 50 949
Barium 558 3195 6144
*Preferred interim median for copper due to high variability.

Barium (median = 3100 ppm) was not resolved in our data due to interferences.
aerosols over a 2-week period, November-December, 2007, as part of an 8-month DRUM to FRM intercomparison with the California ARB in central Sacramento at the 13th and T Street ARB site (Cahill and Barnes 2009, supplementary materials), and the second was a 2007 study directly measuring aerosols across a heavily traveled (65,000 v/day, 1.5% diesels) secondary street in Sacramento, Watt Avenue (Cahill et al. 2007, supple­ mentary materials).

One period in the Sacramento ARB intercomparison in
November and December was chosen for mass closure through addition of surrogate organic measurements (Cahill et al. 1989; Malm et al. 1994 ). As anticipated, most of the ultrafine mass is organic (Table 5). Note also that the only nondetermined major species, nitrate, could not have been a major factor in Sacra­ mento or the agreement would not have been as robust. Farther south in the San Joaquin Valley, nitrates are always a major fac­ tor. The presence of the fine transition metals in these quantities represents a very large number of particles. Note that the zinc value seen in Sacramento in 2007, 11.5 ng/m3, is similar to the transect zinc value in 2009, 14 ng/m3•
The second study was performed at Watt Avenue, the site of several studies on the impact of Watt Avenue on Arden Middle School (Cahill et al. 2007, supplementary materials). The Watt Avenue study utilized the same sampler as the ARB study, with two identical eight stage DRUM samplers with greased Mylar substrates and a47-mm stretched Teflon after filter for the <0.09 µ,m ultrafine mode. The school sampler was 15 m downwind from the nearest traffic lane on the roof of the one story building and 50 m south of a stoplight on Arden Way that backed traffic up to and south of the Arden Middle School site. Thus consid­ erable braking occurred directly upwind of the Arden Middle School sampler. The far upwind sampler was 500 m away in a residential neighborhood. Samples were collected continuously Phosphorus
--. nonsoil Iron
--· Copper
Zin<:
-♦•
Redding Chico Sacramento Fresno Bakersfield
FIG. 13. Very fine plus ultrafine aerosols with known vehicular sources. over 8 weeks in February and March, 2007. The results of these studies are shown in Table 5.

Through the direct upwind downwind measurements across
Watt Avenue, the source of the metals seen in downtown Sacra­ mento can also be shown to be largely derived from traffic sources despite the fact that the nearest freeway to the 13th and T Street site is over 1 km away.

Almost identical values of the same four elements are seen in Sacramento and especially in Bakersfield (Table 6 and Figure 13).

Since the ultrafine four elements, non-soil iron, nickel, cop­ per, and zinc, and a few others, are traffic related, we can ex­ amine the mass levels in the Central Valley to see how the levels correspond to traffic. Since the wind flows downslope across Bakersfield from the south, where 1-5 descends from the Grapevine and Hwy 58 descends from Tehachapi Pass, we used the sum of the cars plus 10 times the trucks, to roughly match to emission rates, and compared the ratio of traffic to PM2.s at the ends of the Valley and across the Chico and Fresno lateral transects for 1-5 plus Hwy. 99. For the valley ends, the traffic ra­ tio was 4.6 and the aerosol ratio 4.4, while for the central lateral transect, traffic was 2.1 and the aerosol was 1 .4. Therefore the PM2.s mass values in the study were similar to the local traffic volumes, and the ultrafine metals at Bakersfield are from the two major highways south of Bakersfield climbing and descending the mountain passes.

Support for this hypothesis comes from historical data taken in Bakersfield in winter, 1974-1976 (Cahill and Flocchini 1974; Flocchini et al. 1976). The historical data show that particulate pollution from cars was high in Bakersfield in the 1970s data.

Lead levels at Bakersfield, population circa 75,000 in 1975, in the average winter period were higher on average than in parts of the Los Angeles urban area. This is a clear indication that the local meteorology and geographical setting of Bakersfield results in it being highly impacted by vehicular sources, most likely vehicles on 1-5 and Hwy 58 as they climb out of or descend into the valley.

Cahill -- 00000294
1132 T. A. CAHILL ET AL.

TABLE 5
Measured and reconstructed ultrafine mass and elemental components in two studies in Sacramento, 2007. Watt 0.5 km is 500 m upwind; Watt 15 mis downwind. Sac Center is the ARB 13th and T Street site. There are no elemental carbon data or nitrate data, so the reconstructed mass is incomplete
Watt ave. upwind ( ~ 0.5 km)
Major components {-lglm3
Mass (gravimetric) Na
Mass (recon.) Na
Organics (H) Na
Sulfur (ammonium sulfate) 0.09
Chlorine (NaCl) 0.00
Soil 0.04
K non-soil (wood smoke) 0.04
Metals 0.01
Minor components ng/m3
Phosphorus 1.1
Vanadium 0.06
Chromium 2.2
Iron (non-soil) 5.2
Nickel 0.3
Copper 0.1
Zinc 1.2
Arsenic 0.16
Selenium 0.10
Bromine 0.8
Lead 0.5
Examining the current IHD mortality data from Figure 3, we observed that if we remove a totally arbitrary average back­ ground IHD mortality rate of 120 from the total rate ( about 2/3 of the valley average rate), and matching the Bakersfield result, the excess IHD mortality roughly scales with local truck traffic on the major freeways (CalTrans 2010) and the very fine-ultrafine iron (Figure 14).

The pattern reflects the pattern that the two major north-south freeways, 1-5 and Hwy 99, come close to each other only Watt ave. downwind (15 m) Sacramento center
{-lglm3 {-lglm3
Na 2.04
Na 2.15
na 1.72

0.18 0.34

0.00 0.04

0.08 0.048

0.14 0.053

0.09 0.035
ng/m3 ng/m3

1.0 2.4

0.18 0.15

1.3 0.45

34.9 17.0

16.3 3.5

9.8 8.3

17.3 11.5

0.46 0.6

0.13 0.3

1.1 3.7

2.3 4
at Sacramento-Stockton and Bakersfield. Note that car traf­ fic has a very different pattern because of the large auto­ mobile traffic on local Sacramento highways. When ultrafine aerosols alone are correlated with IHD mortality for the en­ tire Central Valley, the highest correlations are for phosphorus (r2 = 0.96) and iron (r2 = 0.88), both with known vehicular sources.

The association of IHD with toxic organic contaminants, specifically benzo[a]pyrene (BaP) and coronene (Cahill 2010), TABLE 6
Very fine and ultrafine metals with known vehicular sources Site Redding Chico Sacramento Fresno Bakersfield
Class Very fine + ultrafine Very fine+ ultrafine Very fine + ultrafine Very fine + ultrafine Very fine+ ultrafine Size 0.26-0 f-lm 0 .26---0 f-lm 0.26-0 f-lm 0.26-0 f-lm 0.26-0 f-lm Species
Phosphorus 1.3 1.0 1.2 4.0 14.5
Iron 6.1 5.7 15.3 10.0 43.2
Nickel 0.9 0.13 0.67 0.35 1.8
Copper 4.2 3.1 2.8 2.9 3.6
Zinc 10.5 15.1 15.4 27.3 24.8
Cahill -- 00000295
METALS AND ISCHEMIC HEART DISEASE 1 1133
-- AADTTruck trmnc m~or highways, ·• Iron (vf + uf) (nl)lm' '1000) CONCLUSIONS • .. Sc•l•d lechemle heart dlUHI mortality
!50,--------------------~
RtddinQ Chlto Sacramento Fruno B1ktrtfleld
Central Valley tan ■ ect north to aouth
FIG. 14. Association between local average annual daily truck (AADT) traffic on major freeways, scaled ischemic heart disease, and very fine plus ultrafine iron.
was examined but the pattern did not produce a fit to the IHD increase (Figure 15). The relatively high values of BaP at Chico in winter, observed for years in prior CARB toxics data reported in the ADAM database (ARB ADAM 2010), was traced to the burning of primarily pine wood.

Coronene is normally associated with the exhaust of auto­ mobiles, while BaP has known vehicular sources (diesels and cars) as well as wood smoke (Fujita et al. 2007; Riddle et al. 2007; Kleeman et al. 2009; Cahill 2010). The non-soil, nonwood smoke potassium was also observed by Gertler et al. (2003) in the Tuscarora Tunnel studies from light duty vehicles, but not from diesels. Thus, there appears to be a spark emission source of potassium and coronene in the very fine/ultrafine. A south valley enhancement was also seen in petroleum derived alka­ nes (Cahill 2010), which were present in the largest amounts in Fresno, lesser amounts in Bakersfield, and negligible elsewhere.

In summary, it does not appear that organic aerosols from wood smoke, diesels, and automobiles are a component in the IHD excess.
- • - B&nzo1alp)'l'1lno/3, -- Potassium (nonsoil} (ngtm• I. -+- Coronon& (x 2) ?01 ,----------------------,
' --~-----~---~-~---·
Redding Chico Sacram&nto Frnsno Bal(orsfi&ld
North to south transect
FIG. 15. Comparison ofultrafine BaP, potassium, and coronene.

Persistent enhancement of mortality associated with cardio- vascular and specifically IHD has been observed in the southern San Joaquin Valley since 1990, yet while there is a strong sta­ tistical association with mass, most of the mass is known to be relatively innocuous to the cardiovascular system. Most of the mass is ammonium nitrate, which is soluble in lung fluid. This makes ammonium nitrate unlikely as a source of cardiovascular disease. In order to examine potential causal agents, a profile was made of mass, inorganic, and organic components of mass in nine size modes, including ultrafines <0.09 µ,m, at five sites from Redding to Bakersfield during a 3-week period of strong stagnation, January 5-22, 2009. The strongest correaltions to IHD mortality were found in very fine (0.26-0.09 µ,m) to ul­ trafine metals, with most tied to vehicular sources. This result is supported by several independent lines of reasoning. First, there is an association with truck traffic and IHD throughout the Valley. Second, the four key transition elements, non-soil iron, nickel, copper, and zinc, are closely tied to vehicular sources through upwind-downwind measurements at Watt Avenue, a
secondary artery in suburban Sacramento. Third, the same four elements are found at the downtown ARB 13th and T Street site.

Fourth, there are known diesel sources of zinc and phosphorus, from zinc thiophosphate in lubricating oil, plus calcium as an antacid. No iron, nickel, or copper were seen in the diesel dy­ namometer tests, so these elements are not coming from engine wear. Conversely, brake pads include many elements, includ­ ing iron, nickel, copper, and zinc. The Watt Avenue data were taken just south of a stoplight, so braking was occurring. Heavy braking also occurs on the 1-5 "Grapevine" and Highway 58 downgrades, each more than a 1000 m descent, and both carry heavy truck traffic into Bakersfield, where the same elements are seen again correlating with increased rates of IHD.

Thus, in summary, we present here evidence that, while not conclusive, strongly supports the hypothesis that very fine and ultrafine transition metals are a causal factor in IHD in the Cen­ tral Valley of California. Removal of zinc thiophosphate from lubricating oil could greatly reduce the zinc concentration, while changes in brake drums and pads could reduce the non-soil iron, copper, zinc, and nickel concentrations in ambient air. After a 15-year fight that surfaced with concerns about the San Fran­ cisco Bay, on October 5, 2010, Governor Arnold Schwarzeneg­ ger signed SB346, which requires brake pad manufacturers to reduce the use of copper to not more than 5% by 2021 and no more than 0.5% by 2025. The reasons were based on the toxicity of copper in water run-off from roads and follows similar action by Washington State.

Finally, it is clear that the lack of correlation between PM2.s (or PM1.o) composition and very fine and ultrafine aerosol composition makes measurements of the composition of very fine and ultrafine particles critical for obtaining causal relations to health impacts, both for organic and inorganic species.

Cahill -- 00000296
1134 T. A. CAHILL ET AL.

REFERENCES
ADAM. (2010). California Environmental Protection Agency-Air Resources Board. Retrieved from http:/ /www.arb.ca.gov/ ADAM
Bench, G., Grant, P., Ueda, D., Cliff, S., Perry, K., and Cahill, T. A. (2002). The Use of STIM and PESA to Measure Profiles of Aerosol Mass and Hydrogen Content, Respectively, Across Mylar Rotating Drums Impactor Samples.

Aerosol Sci. Technol., 36:642-651.

California Department of Health Services. (2010). Web Based Inquiry for Mortality Data. Retrieved from http://www.applications.dhs.ca.gov/ vsq/default.asp accessed 3 May, 2011.

Ca!Trans. (2010). CA Transportation Authority, Average Daily Truck Traffic on the California Highway System. Retrieved from http://www.dot.ca.gov/ hq/traffops/saferesr/trafdata/truck2008final.pdf
Cahill, T., Cliff, S. S., Jimenez-Cruz, M., and Perry, K. (2003). Aerosols by Size, Time, and Composition in Support of the Fresno Asthmatic Children's Environment Study, 2001. Final Report to the California Air Resources Board, Sacramento, CA.

Cahill, T., Dutcher, D., Clark, C., Martin, J., McCarthy, T., and Lipnick, D. (1998). Comparison of Cardiac and Stroke Mortality to Carbon Monoxide, Ozone, and Particulate Air Pollutant Concentrations in the Sacramento Valley Region. Report to the Health Effects Task Force, American Lung Association Sacramento Emigrant Trails, Sacramento, CA.

Cahill, T. A., and Barnes, D. E. (2009). Comparison of Fine Mass, UC Davis DRUM versus FRM, at the Air Resources Board 13th and T Street Site. Final Report to the California ARB, Sacramento, CA.

Cahill, T. A., Cahill, T. M., Spada, N. J., and Barnes, D. E. (2007). Organic and Elemental Aerosols near Watt Avenue, Late Winter-Spring, 2007. Final Report to the Sacramento Metropolitan Air Quality Management District, Sacramento, CA.

Cahill, T. A., Eldred, R. A., Motallebi, N., and Malm, W. C. (1989). Indirect Measurement of Hydrocarbon Aerosols Across the United States by Non­ sulfate Hydrogen-Remaining Gravimetric Mass Correlations. Aerosol Sci.

Technol., 10:421-429.

Cahill, T. A., and Feeney, P. J. (1973). Contribution of Freeway Traffic to Airborne Particulate Matter. Final Report of the California Air Resources Board, Contract ARB-502, Sacramento, CA.

Cahill, T. A., and Flocchini, R. G. (1974). Regional Monitoring of Smog Aerosols. Annual Reports to the California Air Resources Board, Contract 2---006, 1974, 1975, 1976, 1977, Sacramento, CA.

Cahill, T. A., Goodart, C., Nelson, J. W., Eldred, R. A., Nasstrom, J. S., and Feeney, P. J. (1985). Design and Evaluation of the Drum Impactor, in Pro­ ceedings of International Symposium on Particulate and Multi-phase Pro­ cesses, T. Ariman and T. Nejat Veziroglu, eds., Hemisphere Publishing Corporation, Washington, DC, Vol. 2, pp. 319-325.

Cahill, T. M. (2010). Size-Resolved Organic Speciation of Wintertime Aerosols in the California Central Valley. Environ. Sci. Technol., 44:2315-2321.

Chan, D., and Stachowiak, G. W. (2004). Review of Automotive Brake Friction Materials. Proc. Instn. Mech. Engrs., 218(9):953-966.

Chen, L. W. A., Watson, J. G., Chow, J. C., and Magliano, K. L. (2007).

Quantifying PM2.5 Source Contributions for the San Joaquin Valley with Multivariate Receptor Models. Environ. Sci. Technol., 41(8):2818- 2826.

Chow, J. C., Chen, L. W. A., Watson, J. G., Lowenthal, D. H., Magliano, K.

A., Turkiewicz, K., and Lehrman, D. E. (2006). PM2.5 Chemical Com­ position and Spatiotemporal Variability During the California Regional PM! 0/PM2.5 Air Quality Study (CRPAQS). J. Geophys. Res.-Atmos., 111:

Dl0.

Chow, J. C., Watson, J. G., Lowenthal, D. H., Chen, L. W. A., Zielinska, B., Mazzoleni, L. R., and Magliano, K. L. (2007). Evaluation of Organic Markers for Chemical Mass Balance Source Apportionment at the Fresno Supersite. Atmos. Chem. Phys., 7(7):1741-1754.

Chow, J. C., Watson, J. G., Lowenthal, D. H., and Magliano, K. L. (2008).

Size-Resolved Aerosol Chemical Concentrations at Rural and Urban Sites in Central California, USA. Atmos. Res., 90(2-4):243-252.

Devlin, R. (2003). What We Know and What We Think We Know About Health Effects of Fine Particles; Is It Enough? American Association of Aerosol Research (AAAR), Annual Meeting, Charlotte, NC.

Flocchini, R. G., Cahill, T. A., Shadoan, D. J., Lange, S. J., Eldred, R. A., Feeney, P. J., Wolfe, G. W., Simeroth, D. C., and Suder, J. K. (1976). Monitoring California's Aerosols by Size and Elemental Composition. Environ. Sci.

Technol., 10:76--82.

Fujita, E. M., Zielinska, B., Campbell, D. E., Arnott, W. P., Sagebiel, J. C., Mazzoleni, L., Chow, J. C., Gabele, P. A., Crews, W., Snow, R., Clark, N. N., Wayne, W. S., and Lawson, D. R. (2007). Variations in Speciated Emissions from Spark-Ignition and Compression-Ignition Motor Vehicles in California's South Coast Air Basin. J. Air Waste Manage., 57(6):705-720.

Gertler, A., Abu-Allaban, M., Coulombe, W., Gillies, J. A., Pierson, W. R., Rogers, C. F., Sagebiel, J.C., Tarnay, L., and Cahill, T. A. (2003). Measure­ ments of Mobile Source Particulate Emissions in a Highway Tunnel. Int. J.

Vehicle Des., 27:86-93.

Hayes, T. P., Kinney, J. J. R., and Wheeler, N. J.M. (1984). California Surface Wind Climetology State of California Air Resources Board, Aerometric Data Division, Sacramento, California.

Kennedy, P., Gadd, J., and Moncrieff, I. (2002). Emission Factors for Contam­ inants Released by Motor Vehicles in New Zealand. Ministry of Transport, Auckland, New Zealand.

Kleeman, M. J., Riddle, S. G., Robert, M.A., Jakober, C. A., Fine, P. M., Hays, M. D., Schauer, J. J., and Hannigan, M. P. (2009). Source Apportionment of Fine (PMl.8) and Ultrafine (PM0.l) Airborne Particulate Matter During a Severe Winter Pollution Episode. Environ. Sci. Technol., 43(2):272-279.

Kleeman, M. J., Schauer, J. J., and Cass, G. R. (2000). Size and Composi­ tion Distribution of Fine Particulate Matter Emitted from Motor Vehicles.

Environ. Sci. Technol., 34(7):1132-1142.

Lippmann, M. (2009). Semi-Continuous Speciation Analyses for Ambient Air Particulate Matter: An Urgent Need for Health Effects Studies. J. Expo. Sci.

Epidemiol., 19:235-247.

Malm, W. C., Sisler, J. F., Huffman, D., Eldred, R. A., and Cahill, T. A. (1994).

Spatial and Seasonal Trends in Particle Concentration and Optical Extinction in the United States. J. Geophys. Res., 99(Dl):1347-1370.

Pope, C. A., III, Thun, M., Namboodiri, M. M., Dockery, D. W., Evans, J. S., Speizer, F. E., and Heath, C. W., Jr. (1995). Particulate Air Pollution as a Predictor of Mortality in a Prospective Study of U.S. Adults. Am. J. Respir.

Crit. Care Med., 151(3) (Part 1):669---674.

Raabe, 0. G., Braaten, D. A., Axelbaum, R. L., Teague, S. V., and Cahill, T. A. (1988). Calibration Studies of the Drum Impactor. J. Aerosol Sci., 19(2):183-195.

Riddle, S. G., Jakober, C. A., Robert, M. A., Cahill, T. M., Charles, M.

J., and Kleeman, M. J. (2007). Large PAHs Detected in Fine Particu­ late Matter Emitted from Light-Duty Gasoline Vehicles. Atmos. Environ., 41(38):8658-8668.

Wesolowski, J. J., John, W., Devor, W., Cahill, T. A., Feeney, P. J., Wolfe, G., and Flocchini, R. (1977). Collection Surfaces of Cascade Impactors, in X­ Ray Fluorescence Analysis of Environmental Samples. T. Dzubay, ed., Ann Arbor Science Publishers, Ann Arbor, Michigan, pp. 121-130.

Zielinska, B., Goliff, W., McDaniel, M., Cahill, T., Kittelson, D., and Watts, W. (2003). Chemical Analyses of Collected Diesel Particulate Matter Sam­ ples in the E-43 Project. National Renewable Energy Laboratory, Golden, COLORADO.

Zielinska, B., Sagebiel, J., McDonald, J. D., Whitney, K., and Lawson, D.

R. (2004). Emission Rates and Comparative Chemical Composition from Selected in-use Diesel and Gasoline Fueled Vehicles. J. Air Waste Manage.

Assoc., 54:1138-1150.

Cahill -- 00000297

EXHIBIT 9
AR002810
Aerosol Science and Technology, 45:1135-1142, 2011
Copyright © American Association for Aerosol Research
ISSN: 0278-6826 print/ 1521-7388 online
DOI: 10.1080/02786826.2011.582196
Very Fine and Ultrafine Metals and lschemic Heart Disease in the California Central Valley 2: 197 4-1991
Thomas A. Cahill, 1,2 David E. Barnes,1 Earl Withycombe, 2 and Mitchell Watnik3 1DELTA Group, University of California, Davis, California, USA 2The Health Effects Task Force, Breathe California of Sacramento-Emigrant Trails, Sacramento, California, USA
3The Statistical Laboratory, University of California, Davis, California, USA The southern part of Central Valley of California in winter has long had high PM10 mass, which until about 1990 included sul­ fate, vanadium, and nickel from the burning of crude oil used to generate steam to enhance heavy petroleum recovery. In roughly 1990, natural gas became the major energy source used for steam generation. In 1989-1991, data were collected throughout Cali­ fornia on the mortality from strokes and ischemic heart disease (IHD). Although no spatial variability was seen for strokes, the southern San Joaquin Valley was found to have IHD mortality rates roughly 60% greater than the rest of the valley. PM10 was statistically identified as the major factor associated with the IHD mortality. However, when the rate of IHD was reexamined in the 2003-2007 period, a sharp reduction, about 30%, was seen in the relative rates for southern San Joaquin Valley as compared with the northern Sacramento Valley. We have measured very fine and ultrafine vanadium and nickel aerosols in a winter experiment in 2009, which shows an order of magnitude reduction in vanadium and nickel aerosols as compared with the pre-1990 data, which is a consequence of the switch from burning crude oil to natural gas to generate the steam. The inference of a causal relationship between the reduced vanadium and nickel and the improved IHD rate is supported by a growing body of laboratory and epidemio­ logical work on the toxicity of vanadium and nickel, including from oceangoing ships burning crude or residual oil.

Received 14 October 2010; accepted 27 November 2010.

The authors wish to gratefully acknowledge the input and over­ sight of the project by the ALASET Health Effects Task Force under Jananne Sharpless, Chair, and Betty Turner, ALASET liaison, HETF members Tom Cahill, Mel Knight, Michael Lipsett, Ralph Propper, Marc Schenker, Steve Van den Eden, and Earl Withycombe, as well as the suggestions of William Wilson, US EPA. The authors also wish to gratefully acknowledge the financial support of the Sacramento Metropolitan Air Quality Management District under its CEO, Norm Covell, and the Yolo-Solano AQMD, under its CEO, Larry Greene.

Finally, the authors wish to thank Ms. Jeanette Martin for her untiring assistance in designing and assembling the original report.

Address correspondence to Thomas A. Cahill, DELTA Group, Uni­ versity of California-Davis, One Sheids Ave., Davis, CA 95616, USA.

E-mail: [email protected]

INTRODUCTION
The Central Valley of California is well known for exhibiting in much of its area serious violations of federal PM10 , PM2.s, in winter, and ozone standards, in summer, matching or exceeding those seen for any commensurate area of the United States. Yet other parts of the valley share agriculturally based land uses and meteorology and do not exhibit serious violations of these stan­ dards. Although efforts continue to control particulate matter (PM) and ozone violations, this pollutant gradient offers an ideal situation to examine the effects of pollutants on health. Many of the confounding variables seen in comparison studies, such as meteorology, land use patterns, and cigarette smoking, are very similar throughout the Valley, and a wealth of data exists on the pollutants and their sources from district air monitoring and Air Resources Board (ARB) studies. The presence of detailed anal­ ysis of mortality from ischemic heart disease (IHD) and stroke during 1989-1991 reported by the Cardiovascular Disease Out­ reach and Epidemiology (CORE) Program (Kao et al. 1994) and the California Department of Health Services (CADHS 2010) provides a natural opportunity to compare the pollutant data and mortality data for the identical time ranges. While this ar­ ticle focuses on PM effects, the original report (Cahill et al. 1998) also examined the effects of ozone and carbon monoxide (CO) on IHD and stroke mortality; some of which are presented in this article for completeness. However, complexities in the spatial and temporal gradients of ozone require further inter­ pretation, and thus, they are not examined in any detail in this article.

BACKGROUND
The northwest-southeast trending 550 km by 100 km Cen­
tral Valley of California is one of the richest agricultural ar­ eas in the world (Figure 1). In addition, there are two major metropolitan areas in the Valley, the Sacramento-Fresno regions each with about 1 million residents, and otherwise towns are generally medium (circa a few 100,000) to small and spread 1135
Cahill -- 00000277
AR002811
1136 T. A. CAHILL ET AL.
1----1
1001:.m
FIG. 1. Map of the Central Valley of California. The northern part is referred to as the Sacramento Valley while the southern part is called the San Joaquin Valley. The areas within the dotted circumference vary in altitude from almost sea level in the Delta to a few I 00 m in elevation. The small arrows denote the typical winter wind patterns (Hayes et al. 1984).
throughout the Valley. The mean population density is about 60 people/km 2•
High summer temperatures in the Valley (mean daily high, July, 36°C ± 1 °C) lower the surface pressure and draw massive inputs of cool air into the Valley from the San Francisco Bay area, bringing with it oceanic aerosols and typical urban pollu­ tants enhanced by the major petrochemical industrial sites at the Carquinez Strait. These latter sources lie directly in the path for the cold, dense winds from the San Francisco Bay. These winds funnel down the entire San Joaquin Valley, south of the strait, but rarely penetrate north very far in the Sacramento Valley, north of the strait. Rainfall in the Valley is rare during the period from May through October.

In winter, there is slow drainage of cold air off the Sierra Nevada to the Valley floor (mean daily low, January, 3°C ± 1 °C) and then, at the rate of about 50 km/day, the air drains toward the lowest point, the Sacramento-San Joaquin Delta, directly east of the Carquinez Strait. Thus, little impact from the San Francisco Bay area is seen on most winter days in the Central Valley beyond the immediate Delta region. Winter rainstorms and strong northwest winds periodically clean out the Valley.

Thus, the Valley tends to fall into two distinct and very different conditions: hot, dry summers with strong diurnal winds, and cool, foggy winters with long periods of stagnation. This results in two very different aerosol types: summer and winter (Figure 2).

In Figure 2, 24 h, one day in six PM10 mass at the Fresno 1st Street "Super-site" is shown (ADAM 2010). The site is in a largely residential neighborhood roughly 1 km east of the 2 3 4 s 7 8 9 10 11 12
Month of th• Year 2002
FIG. 2. PM10 mass at the Fresno "Super-site", 2002.
nearest major freeway, well away from agricultural operations. It is thus a good representation of what the average Fresno resident inhales. The annual pattern of PM10 mass has been consistent in the valley for many years, but this data only really became available around 1997, hence why Figure 2 uses more recent data to demonstrate the traditional annual cycle of particulate mass.

The health effects data are derived from CORE (Kao et al. 1996) and CADHS (2010). CORE collected city (>20,000 in­ habitant) and countywide data on mortality from stroke and IHD; the latter is the largest source of mortality in the Califor­ nia Central Valley accounting for about one fourth of all deaths.

The CORE data were corrected for age, > 35 years, and reduced to incidence per 100,000 residents. The results were then pre­ sented as a ratio to the California average values the both males and females. Individual city data were weighted and averaged to obtain countywide averages in order to match the air pollution data.

The effect of race was examined over the entire California dataset. The data were available only in three classifications:
"White," "Black," and "Other," with "Other" in the Central Val­ ley being largely Latinos. No statistically significant difference was seen in the IHD rates for the dominant > 75 years age group, but a slight enhancement of the rates for Blacks was seen in the 64- to 75-year cohort. For all ages, the mortality/100,000, 95% confidence limit, was 196 ± 3 for white males, 212 ± 9 for black males, and 199 ± 4 for other males; the corresponding values for females were 127 ± 3, 166 ± 5, and 135 ± 3. The stroke deaths, on the contrary, while only roughly one third of the rate of IHD mortality, showed stroke rates roughly twice as high for the "Black" and "Other" cohorts than the "White" cohort. Consideration was given to cigarette-smoking patterns, but no significant difference was seen along the length of the Valley (Table 1).

Four metrics for air pollution were studied, each with very different cardiovascular impacts: ozone (top 30 h and hours above 0.09 ppm), annual average CO, and annual average PM10 .

The approximate cigarette-smoking patterns were calculated Cahill -- 00000278
AR002812
METALS AND ISCHEMIC HEART DISEASE 2 1137
TABLE 1
Ischemic heart disease (IHD) and stroke mortality versus air pollutants in the Central Valley of California, 1989-1991 IHD rate vs. Stroke rate vs.

County (north CA average CA average Ozone top 30
to south) (%) (%) h (ppm)
Shasta -14 +9 0.095
Butte -27 -3 0.078
Sutter -4 +26 0.090
Placer -17 -4
Sacramento -6 +2 0.107
Yolo +2 +13
San Joaquin +15 +17 0.096
Stanislaus -7 -3 0.102
Madera +11 -23
Fresno +3 -12 0.129
Kings +17 +20 0.106
Tulare +22 +8 0.093
Kern +33 -11 0.111
from cigarette sales data. Correlations were calculated for all parameters and are presented in Table 2 and Figures 3 and

4. At some sites, measurements were lacking. No data were deleted.

To further investigate the statistical robustness of these data, we used "proc glm" on SAS version 9.1 for the analyses. Be­ cause there are only 13 observations in total, we did not feel that imputation would prove fruitful. Therefore, we ran mul­ tiple regression models with "percentage change of IHD" and "percentage change of stroke" as the outcome variables. The possible explanatory variables were "number of hours of ozone above 0.09 ppm" ("ozone"), "annual average PM less than 10 /lm, measured in micrograms per cubic meter" ("PM10"), and "percentage rate of smoking" ("smoking"). We checked the assumption of normality of the errors by formally using the Shapiro-Wilk test and the assumption of homoskedasticity in­ formally using residual plots.

The overall stroke model was not significant (p = 0.1108). We did not proceed further for that outcome. However, the overall IHD model was significant (p = 0.0330). We used backward elimination as the model selection strategy, for example, that ofNeter et al. (1996). Ozone and smoking were eliminated-in that order-leaving PM10 as the sole significant explanatory variable (p = 0.0026). The resulting linear regression is given PM10 annual
Ozone hours CO annual average Smoking rate
>0.09 (h) average (ppm) (/lg/m 3) average(%)
20 0.54 30 22.4
6 0.93 38 22.4
4 44 22.8
125 30 22.8
153 1.29 44 23.4
16 30 22.4
36 1.13 51 22.8
119 0.75 48 22.0
30 53 22.0
369 0.87 77 22.0
21 61 22.5
192 0.88 67 22.5
605 0.80 68 22.5
by
estimated percentage change of IHD = -41.5 + 0.87 x PM10 , and the corresponding r-squared value is 0.576. It is noteworthy that Fresno appears to be an outlier. If this point is removed, the slope and r-squared values jump dramatically. We have no cause to remove that point and we therefore use the model, as indicated earlier, as our finding. However, the Fresno value includes semirural sampling sites at the urban fringe that are not present in other San Joaquin Valley cities, which had at that time only a single city center site.

However, the strong association between aerosol mass and IHD posed a problem. As additional data were developed in the Central Valley from the IMPROVE program (Malm et al. 1994) and from extensive California studies (CRPAQS 2001, Cahill et al. 2003), it became clear that the major aerosol species that dominated fine mass was ammonium nitrate, with much smaller contributions from organic matter, including a wood smoke component, and ammonium sulfate. These are
however water soluble and have not been closely linked to IHD in animal and laboratory studies. (Devlin 2003; Lippmann 2009).

TABLE 2
Correlation of air pollution to mortality by IHD and stroke in the Central valley of California, 1989-1991 Mortality
Correlation to IHD
Correlation to stroke
Ozone average top 30 h (ppm) Ozone number hours> 0.09 (ppm) r2 = 0.18
r2 = 0.21
r2 = 0.27
r2 =0.l9
CO (ppm)
r2 < 0.05
r2 < 0.05
r2 = 0.56
r2 < 0.05
Cahill -- 00000279
AR002813
1138 T. A. CAHILL ET AL.
■ IHD II Annual average PM10 mass
FIG. 3. Comparison of death rate from IHD, compared to the California av­ erage, and annual PM10 inhalable aerosol mass. Shasta County is at the very northern end of the California Central Valley, and Kern County is at the southern end.
200 ~--------------------~
:€ 150
Q.
"' 1:11
f : 100
..
i
~ 50
.. " . " •
o~---------------------~
60 70 80 !10 100 110 120 130 140 150 160
Relative rate of IHD mortality to CA average 1%)
FIG. 4. IHD mortality rate relative to the California average versus PM10. The parameters are associated by a regression r2 = 0.56.
90 ,------------------------,
80 70
60 1 50
::i.40 ~ 30
20 10
0'--------='-------------------'
2 3 4 5 6 7 8 9 10 11 11 12
2 3 4 5 6 6 7 8 9 10 11 12
Month of the Year 2002
FIG. 5. PM2.s mass at the Fresno "Super-site," 2002, the same period used in Figure 2.
--- 1989-1991 -- 2003-2007
FIG. 6. IHD in the California Central valley, 1989-1991 versus 2003-2007.

The 1989-1991 data are scaled to match the Shasta-Butte data, 2003-2007, as diagnostic protocols differed.

Reduction of I HD Rates: 2003-2007 Versus 1989-1991
Beginning around 1999, data on aerosol in the Central Valley were enhanced by the availability of PM2.s mass and chemical speciation at some San Joaquin Valley sites. In Figure 5, the same period for the Figure 2, one day in six PM 10 mass is shown for 24 h daily fine PM25 mass (ADAM 2010). The old federal 24-h PM2.s standard was 65 {lglm 3 and the current 24-h standard is 35 {lglm 3, which is routinely violated in winter. The annual average standard is 15 {lglm 3•
The question of IHD and aerosols was revisited for the period 2003-2007 (Cahill et al. 2010). From the recent data, it became 200
100
!iO
0 _.... <0.6 micron• -- 0.6-3.S micron•
~ Vanadium
I\
) N\_ f"-.
'-. V
1 3 15 7 9 11 13 16 17 19 21 23 26 27 29 31
2 4 11 8 10 12 14 16 18 20 22 24 211 28 30
_,,_.. -<0.6 micron• -+- 0.6-3.6 micron•
250 r-----------------------~
200
e: 1so
0 .::: 100
60 Nickel
1 3 5 7 II 11 13 16 17 111 21 23 26 27 29 31
2 4 6 8 10 12 14 11!1 18 .0 22 2A 2a 28 30
January 1976
FIG. 7. Vanadium and nickel aerosols in Bakersfield, January 1976.

Cahill -- 00000280
AR002814
METALS AND ISCHEMIC HEART DISEASE 2 1139
ra Elemental carbon ■ Organic aerosols ■ Ammonium nitrate ~ Ammonium sulfate ■ Soil
30 25
20 \ 15
10 5
0 11 11 11 12 12 12 12 12 1 1 1 1 1 2 2 2
(a)
ail! PM 2_5 mass (gravimetric) ■ PM 2_5 mass (sum of speciu) 30 25
20 .., 15
i
:::t 10
5 0
11 11 11 12 12
November
12 12 12
December
(b)
1 1 1
January
1 1 2 2
February
2 FIG. 8. (a) and (b) showing PM2.s mass closure and major species. clear that a comparison of the IHD mortality, 1989-1991 versus 2003-2007, shows a major reduction, roughly 30%, that occurs in 2003-2007 for the counties north of Bakersfield (Figure 6).

The trend for the northern half of the Central Valley is similar, while Bakersfield in Kern County is only slightly reduced. Thus, the question arises as to the causes for the improvement in the IHD rate.

The compositional analysis of Central Valley aerosols dates back to the early 1970s, including an innovative program of UC Davis for the California Air Resources Board (CARB), 1973-1977 (Cahill et al. 1974; Flocchini et al. 1976; Barone et al. 1978; Motallebi et al. 1990a, 1990b). A total of 14,100 daily aerosol measurements were taken in three size modes (15-3.5 µ,m, 3.5-0.50 µ,m, and 0.50--0.0 µ,m) and analyzed for elemental content by particle-induced x-ray emission (PIXE) (Cahill 1995) at about fourteen separate sites, including the valley sites of Chico, Sacramento, Visalia, (100 km NNE of Bakersfield), and Bakersfield. Bakersfield had one of the highest levels of vanadium and nickel aerosols in California, but port areas (Oakland, Richmond, and Los Alamitos) were also high (Figure 7).

Neither vanadium nor nickel was seen in sizes above 3.5 µ,m diameter. The radically different vanadium/nickel ratios oc­ curred when winds came from the western (Taft, 40 km SW) or northern (Oil dale, 5 kmN) oil fields periods of stagnation, giving the highest pollution levels, favored sources in the much closer northern oil fields, some of which were within Bakersfield itself (Motallebi et al. 1990b ). In addition, there was an oil refinery in Oildale, with potentially different emissions than the crude oil burned for enhanced oil recovery (EOR). Also it is notewor­ thy that while the vanadium and nickel concentrations can vary by over an order of magnitude, the lead levels were relatively Cahill -- 00000281
AR002815
1140 T. A. CAHILL ET AL.
-- Sulfur 1!174 - Lead 1!174 Sulfur 200!1
Redding Chico Sacramento Vlsali:llfresno B11ker11fleld
Vanadium 1974 - • Nickel 1974 - Zinc 1974
711
!ill
◄ II
311
211
10 II
ReddiR<J Chico 811.aamenlo Vis11.lil!Jf'msno Bllkerslield North to south transect
FIG. 9. Lead and sulfur, 197 4-1976, versus sulfur, 2009, and vanadium, nickel, and zinc 1974-1976. Visalia lies between Bakersfield and Fresno. constant, indicating a continuous impact of traffic-derived aerosols at Bakersfield.

Recent compositional analysis of the PM2.s aerosols in the southern San Joaquin Valley also became available through the IMPROVE site at Sequoia National Park (NP) starting in 1992 (Malm et al. 1994). The Sequoia NP Ash Mountain site was at low elevation, near 600 m, and thus received aerosols from the valley floor whenever the inversion lifted to that level.

IMPROVE had anticipated the value of PM2.s mass and full speciation, using quality assurance protocols of "integral redun­ dancy" to measure major parameters by two or more indepen­ dent methods. Sulfur from PIXE on Teflon in Channel A (Cahill 1995) is compared with sulfates after a denuder on Nylon filters in Channel B. Organic matter measured by combustion from tandem quartz filters in Channel C is compared to organic mat­ ter estimates by nonvolatile hydrogen of Teflon in Channel A (Cahill et al. 1989). Iron was measured by both PIXE and x-ray fluorescence (XRF) on the Teflon filters of Channel A. Thus, the gravimetric mass can be directly compared to the sum of all species (Figure 8b) and all major components (Figure 8a).

Figure 8 shows periodic incursions of high aerosol levels throughout the winter whenever the valley inversion rises to the altitude of Ash Mountain. Other than during these incursions, aerosol mass levels are very low, a few {lglm 3• By choosing an arbitrary concentration of 10 {lglm 3, we can be assured that we are seeing San Joaquin Valley aerosols diluted by the high inversion levels necessary for the Ash Mountain incursions.

Finally, the excellent agreement between the mass of aerosol measured gravimetrically and the sum of species for all three IMPROVE channels (Figure 8a) shows that we are not missing any major component of the PM2.s mass.

Figure 5 showed that the violations of the annual and 24-h PM25 standards in the San Joaquin Valley are driven by the winter aerosols. The high PM2.s values generally correspond to stable conditions with low inversions, and the clean periods to synoptic rainstorms passing through the Valley. Figure 8b shows that winter composition is dominated by fine Dp < 2.5 /lm nitrates with smaller contributions from organic matter, wood smoke and vehicular exhaust, and some sulfates (IMPROVE 2010). The IMPROVE data also include fine metals, thus aiding our intercomparisons.

Winter data from the earlier efforts (Cahill et al. 1974, 2003; Flocchini et al. 1976) can be roughly compared with the current work (Cahill et al. 2010) by summing the four finest DRUM stages to obtain <0.56 /lm particles (Table 3 and Figure 6).

Addressing the potential health impacts of these data, vana­ dium and nickel have the highest mortality risk coefficients of seventeen fine particulate matter (FPM) components (Lippmann et al. 2006). The elevated levels seen in 1974-1976 were sharply reduced as early as 1992-1993, using the vanadium and nickel levels in the southern San Joaquin Valley from the Sequoia IM­ PROVE site, although the Sequoia NP site's PM2.s data are not directly comparable to the valley floor's very fine ( <0.5 /lm) particle data. These data are shown in Figure 9. For compari­ son's sake, vanadium and nickel have current annual averages in FPM of 1.9 ng/m3 across sixty metropolitan areas (MSAs) (Lippmann et al. 2006).

Examining the potential sources for the very fine (in this case <0.5 /lm) particles (Table 3 and Figure 6), we see sulfur plus metals associated with crude oil combustion (vanadium and nickel), as well as lead and other vehicular metals. Kern County has very large reserves of crude oil, most of which lie just north and west of Bakersfield, although there was and is some oil extraction within the city itself. The Kern County oil fields have a heavy crude oil that requires EOR techniques such as steam injection to extract. Initially, the steam was generated by burning the crude oil itself, resulting in high pollution lev­ els. In 1979, violations of federal air quality standards required Kern County to apply strict controls on sulfur dioxide, nitro­ gen dioxide, and particulate emissions, which included metals vanadium and nickel. EOR by crude oil combustion dropped slowly, 1980-1990, and then dramatically with the increasing availability of natural gas, with an immediate improvement in air quality by the early 1990s (CA Almanac). Thus, the popula­ tion subject to IHD mortality in Bakersfield 1989-1991 had at least two decades of exposure to elevated high levels of very fine to ultrafine transition metals, specifically vanadium and nickel at 10-20 times the current US averages.

Laboratory data on the impact of vanadium and nickel
aerosols has become more extensive (Zhang et al. 2009).

Thus, we can hypothesize that the effective control of the Cahill -- 00000282
AR002816
METALS AND ISCHEMIC HEART DISEASE 2 1141
TABLE 3
Comparison between January 1974-1976, averaged, and January 2009. The four finest DRUM stages of the 2009 data have been summed to be PM < 0.56 µ,m to better match the PM < 0.5 µ,m data of the earlier ARB/UCD aerosol network. The Sequoia NP and Fresno winter 2002-2003 data are of PM2.s
Valley transect 2009 [January 2009 (ng/m 3) <0.56 µ,m]
Redding
Chico
Sacramento
Fresno
Bakersfield
Fresno PM2.s (ng/m 3)
Winter 2002
Sequoia NP ( ~600 m), 10 episodes PM2.s > 10 µ,g/m 3
Sulfur

88.8 136.9
246.0
261.4
505.0
266.2
Vanadium

0.2 0.5

0.4 0.3

0.2 0.47
Nickel

1.0 0.4

0.1 0.5

2.3 Zinc

13.9 5.2

6.9 40.3

32.0 12.7
Lead

2.5 2.1

1.6 4.9

9.4 1.2
Winter
2002-2003 (ng/m 3) 194 0.2

0.22 0.1

0.1 4.7 1.0
Winter 1992-1993 (ng/m 3) 562 0.7 6.0 2.0
Valley transect 1974-1976 [January 1974-1976 (ng/m 3) average, <0.5 µ,m diameter] Redding na na na na na
Chico 170 7 10 483
Sacramento 375 0 1
8 38
18 720
Visalia 225 17 19 593
Bakersfield 1685 19 61 1714
Port of Los Angeles
August-September, 2008
crude-oil-derived metals in the late 1980s may be responsible for the decreased IHD death rate at sites north of Bakersfield in the 1990s and beyond. This hypothesis is supported by other studies, such as the sharp drop in mortality after a sharp drop in sulfur, vanadium, and nickel in Hong Kong when the sul­ fur levels in fuels were lowered by edict (Hedley et al. 2002; Lippmann et al. 2006). The presence of high levels of nickel in New York City from residual oil combustion in buildings is hypothesized to be a causal factor in the enhanced mortality Nlcktl 0.25-0.09
25 ,------------------------.

FIG. 10. Very fine (0.26 > Dp > 0.09 µ,m) vanadium and nickelin Wilmington, CA, downwind during daylight hours at the Port of Los Angeles. 2565 16 4 13 6
associated with PM2.s (Peltier et al. 2009; Peltier and Lippmann 2010).

The current data also have relevance to the emission of sulfur, vanadium, and nickel by oceangoing ships. Recent data in New York and New Jersey (Peltier et al. 2009), and the Port of Los Angeles (Cahill et al. 2009), show the impacts of oceangoing ships burning residual or bunker oil. In the Los Angeles study, the average concentrations of fine sulfur, vanadium, and nickel in the coastal town of Wilmington in August 2008 were 2565, 16, and 4.4 ng/m 3, respectively (Table 3). Southern winds bring aerosols from the Port of Los Angeles into Wilmington routinely during part of each day (Figure 10). Ultrafine particles were not collected in this study, so the actual values could have been significantly higher than these values. Thus, on the basis of the Bakersfield data, the enhanced rates of IHD are predicted in the area.

CONCLUSION
IHD mortality in the Central Valley of California, corrected for age, race, sex, and smoking, was almost 60% greater at the more polluted southern end of the Central Valley than the less polluted northern end in 1989-1991 despite similar meteorol­ ogy and land use patterns throughout the Valley. A significant association, r2 = 0.56, is seen between IHD mortality and PM10, but negligible correlation, r2 < 0.05, is seen with stroke mortal­ ity. When PM10 data are converted to an estimated PM2.s mass Cahill -- 00000283
AR002817
1142 T. A. CAHILL ET AL.
by using more recent measurements, these results are similar to other studies such as the "6 Cities" study (Dockery et al. 1993) in mortality increase per particulate mass; 1.9% mortality in­ crease per {lglm 3 of PM2.s for "6 Cities," while 1.8% mortality increase per {lglm 3 of estimated PM25 for this study, despite major differences in meteorology and pollutant mix. No associ­ ation was seen between CO and any health impact. Ozone was only weakly correlated with HID and stroke. However, strong east-west ozone gradients were not corrected for in this work, and thus, these results for ozone must be viewed as merely indicative of the possible impacts.

The sharp reduction seen, when the IHD data were reexam­ ined in the period 2003-2007, was coincident with a reduction in the high levels of vanadium ( 19 ng/m 3) and nickel (38 ng/m3) due to oil field operations from before 1974 to after 1990, when new protocols were adopted. The vanadium and nickel levels in 2009 were close to the national average values, both roughly

1.9 ng/m3• The relatively smaller reduction in IHD rate in Bak­ ersfield itself is associated with the continuing impacts of ve­ hicular very fine and ultrafine metals (Cahill et al. 2010, this issue).

REFERENCES
ADAM. (2010). ADAM California Environmental Protection Agency-Air Resources Board, ADAM Data Summaries. Retrieved from
www.arb.ca.gov/adam
Barone, J. B., Cahill, T. A., Eldred, R. A., Flocchini, R. G., Shadoan, D. J., and Dietz, T. M. (1978). A Multivariate Statistical Analysis of Visibility Degradation at Four California Cities. Atmos. Environ., 12:2213-2221.

Cahill, T., Cliff, S. S., Jimenez-Cruz, M., and Perry, K. (2003). Aerosols by Size, Time, and Composition in Support of the Fresno Asthmatic Children's Environment Study, 2001. Final Report to the California Air Resources Board, Sacramento, CA.

Cahill, T., Dutcher, D., Clark, C., Martin, J., McCarthy, T., and Lipnick, D. (1998). Comparison of Cardiac and Stroke Mortality to Carbon Monoxide, Ozone, and Particulate Air Pollutant Concentrations in the Sacramento Valley Region. Report to the Health Effects Task Force, American Lung Association Sacramento Emigrant Trails, Sacramento, CA.

Cahill, T. A. (1995). Compositional Analysis of Atmospheric Aerosols, in Particle-Induced X-Ray Emission Spectrometry (Chemical Analysis Series, Vol. 133), S. A. E. Johansson, J. L. Campbell, and K. G. Malmqvist, ed., John Wiley & Sons, Hoboken, NJ, pp. 237-311.

Cahill, T. A., Barnes, D., and Smeltzer, K. (2009). Deposition of Coarse Toxic Particles in Wilmington, CA. Final Report for the Department of Toxic Substances Control (DTSC), Sacramento, CA.

Cahill, T. A., Barnes, D. E., Spada, N. J., Lawton, J. A., and Cahill, T. M. (2010).

Very Fine and Ultra-Fine Metals and Ischemic Heart Disease in the Califor­ nia Central Valley 1: 2003-2007. Aerosol Sci. Technol., 45:1124-1134.

Cahill, T. A., Eldred, R. A., Motallebi, N., and Malm, W.C. (1989). Indirect Measurement of Hydrocarbon Aerosols Across the United States by Non­ Sulfate Hydrogen-Remaining Gravimetric Mass Correlations. Aerosol Sci.

Tech., 10:421-429.

Cahill, T. A., and Flocchini, R. G. (1974). Regional Monitoring of Smog Aerosols, Annual Reports to the California Air Resources Board, Contract 2---006, 1974, 1975, 1976, 1977, Sacramento, CA.

California Department of Health Services. (2010). California Department of Health Services: Center for Health Statistics Office of Health Informa- tion and Research (OHIR). Retrieved from http://www.dhs.ca.gov/hisp/chs/ OHIR/vssdata/Tables.htm
CORE. (September 1994). Cardiovascular Disease Outreach and Epidemiology Program. The Local Burden of Cardiovascular Disease; Death from Heart Disease and Stroke in California Cities. Report No. 1. Kao, C., Gazzinaga, J. M., Cowling, D. W., Cohen, R. D., Fox, P., Kaplan, G. A. California De­ partment of Health Services and the University of California, San Francisco, CA.

California Regional Particulate Air Quality Study (CRPAQS). (2001).

California Regional Particulate Air Quality Study. Retrieved from http:/ /www.arb.ca. gov /airways
Devlin, R. (2003). What We Know and What We Think We Know About Health Effects of Fine Particles: Is It Enough? Annual Meeting. American Associ­ ation of Aerosol Research (AAAR), Charlotte, NC.

Dockery, D. W., Pope, C. A. III, Hu, X. P., Spengler, J. D., Ware, J. H., Fay, M. E., Ferris, B. G., and Speizer, F. E. (1993). An Association Between Air Pollution and Mortality in Six US Cities. N. Engl. J. Med., 329:1725-1759.

Flocchini, R. G., Cahill, T. A., Shadoan, D. J., Lange, S. J., Eldred, R. A., Feeney, P. J., Wolfe, G. W., Simeroth, D. C., and Suder, J. K. (1976). Monitoring California's Aerosols by Size and Elemental Composition. Environ. Sci.

Technol., 10:76--82.

Hayes, T. P., Kinney, J. J. R., and Wheeler, N. J.M. (1984). California Surface Wind Climatology, State of California Air Resources Board, Aerometic Data Division, Sacramento, CA.

Hedley, A. J., Wong, C.-M., Thach, T. Q., MStat, S. M., Lam, T.-H., and Anderson, H. R. (2002). Cardiorespiratory and All-Cause Mortality After Restrictions on Sulfur Content of Fuel in Hong Kong: An Intervention Study.

Lancet, 360(9346):1646--1652.

Interagency Monitoring of Protected Visual Elements (IMPROVE). (2010).

Retrieved from http://vista.cira.colostate.edu/IMPROVE/ Kao, C., Gazzinaga, J. M., Cowling, D. W., Cohen, R. D., Fox, P., and Kaplan, G. A. (1994). Caralio Vascular Disease Outreach and Epidemiology Pro­ gram. The Local Burden of Cardiovascular Disease; Death from Heart Dis­ ease and Stroke in California Cities. Report No. 1. California Department of Health Services and the University of California, San Francisco, CA.

Lippmann, M. (2009). Semi-Continuous Speciation Analyses for Ambient Air Particulate Matter: An Urgent Need for Health Effects Studies. J. Expo. Sci.

Epidemiol. Stud., 19:235-247.

Lippmann, M., Ito, K., Hwang, J. S., Maciejczyk, P., and Chen, L. C. (2006).

Cardiovascular Effects of Nickel in Ambient Air. Environ. Health Perspect. 1141(11):1662-1669.

Malm, W. C., Sisler, J. F., Huffman, D., Eldred, R. A., and Cahill, T. A. (1994).

Spatial and Seasonal Trends in Particle Concentration and Optical Extinction in the United States. J. Geophys. Res., 99(Dl):1347-1370.

Motallebi, N., Cahill, T. A., and Flocchini, R. G. (1990a). Influence of Particulate Size on Statistical Studies of Visibility at California Regions. Atmosfera, 3(2):111-126.

Motallebi, N., Flocchini, R. G., Myrup, L. 0., and Cahill, T. A. (1990b). A Prin­ cipal Component Analysis of Visibility and Air Pollution in Six California Cities. Atmosfer., 3(2):127-141.

Neter, J. W., Kutner, M. H., Nachtsheim, C. J., and Wasserman, W. (1996).

Applied Linear Statistical Models (4th ed.). Irwin, Chicago, IL, p. 353.

Peltier, R. E., Hsu, S.-I., Lall, R., and Lippmann, M. (2009). Residual Oil Combustion: A Major Source of Airborne Nickel in New York City. J.

Expo. Sci. Epidemiol., 19:603---612.

Peltier, R. E., and Lippmann, M. (2010). Residual Oil Combustion: 2. Distribu­ tion of Airborne Nickel and Vanadium Within New York City. J. Expo. Sci.

Epidemiol., 20:342-350.

Zhang, Z., Chau, P. Y., Lai, H. K., and Wong, C. M. (2009). A Review of the Effects of Particulate Matter-Associated Nickel and Vanadium Species on Cardiovascular and Respiratory Systems. Int. J. Environ. Health Res., 19(3):175-185.

Cahill -- 00000284

EXHIBIT 11

UNCORRECTED PROOF

ARTICLE
INFO
Article history:

Received 17 June 2016
Received in revised form 5 September
2016
Accepted 12 September 2016
Available online xxx
Keywords:

Aerosols
Highway
Size profiles
Lateral transects
Ultra fine metals

ABSTRACT
As
one component of a study investigating the impact of vehicle emissions on near-road air quality, human exposures, and potential health effects, particles were measured from September 21 to October 30, 2010 on both sides of a major roadway (Interstate-96) in Detroit. Traffic moved freely on this 12 lane freeway with a mean velocity of 69 mi/hr. with little braking and acceleration. The UC Davis DELTA Group rotating drum (DRUM) impactors were used to collect parti- cles in 8 size ranges at sites nominally 100 m south, 10 m north, 100 m north, and 300 m north of the highway. Ultra-fine particles were continuously collected at the 10 m north and 100 m north sites. Samples were analyzed every 3 h for mass (soft beta ray transmission), 42 elements (synchrotron-induced x-ray fluorescence) and optical attenuation (350–800 nm spectroscopy). A three day period of steady southerly winds along the array allowed direct measurement of freeway emis- sion rates for coarse (10 > Dp > 1.0 μm), PM2.5, very fine (0.26 > Dp > 0.09 μm), and ultra-fine (Dp < 0.09 μm) particles.

The PM2.5 mass concentrations were modeled using literature emission rates during the south to north wind periods, and averaged 1.6 ± 0.5 μg/m3, versus the measured value of 2.0 ± 0.7 μg/m3. Using European freeway emission rates from 2010, and modeling them at the I-96 site, we would predict roughly 3.1 μg/m3 of PM2.5 particles, corrected from the 4.9 PM10 value by their measured road dust contributions. Using California car and truck emission rates of 1973, this value would have been about 16 μg/m3, corrected down from the 19 μg/m3 PM5.0 using measured roadway dust contributions.

This would have included 2.7 μg/m3 of lead, versus the 0.0033 μg/m3 measured. Very fine particles were distributed across the array with a relatively weak falloff versus distance. For the ultra-fine particles, emissions of soot and metals seen in vehicular braking studies correlated with traffic at the 10 m site, but only the soot was statistically significant at the 100 m north site. Otherwise, the 10 m north and 100 m north sites were essentially identical in mean concentration and highly correlated in time for most of the 5 week study. This result supports earlier publications showing the ability of very fine and ultra-fine particles to transport to sites well removed from the freeway sources. The concentrations of very fine and ultra-fine metals from brake wear and zinc in motor oil observed in Detroit have the potential of being a significant component in statistically established PM2.5 mortality rates. © 2016 Published by Elsevier Ltd.

Atmospheric Environment xxx (2016) xxx-xxx
Contents lists available at ScienceDirect
Atmospheric Environment
journal homepage: www.elsevier.com
Transition
metals in coarse, fine, very fine and ultra-fine particles from an interstate highway transect near Detroit
Thomas A. Cahill a, ∗, David E. Barnes a, Jonathan A. Lawton a, Roger Miller a, Nicholas Spada b, Robert D. Willis c, Sue Kimbrough d
a DELTA Group, University of California, Davis, CA 95616, USA b University of Houston, Houston, TX, USA
c US Environmental Protection Agency, National Exposure Research Laboratory, Research Triangle Park, NC 27711, USA d US Environmental Protection Agency, National Risk Management Research Laboratory, Research Triangle Park, NC 27711, USA

1. Introduction
Enormous advances have been made in the past 40 years in reduc- ing particulate pollutants from highway vehicles, including the elim- ination of lead from gasoline. Additionally, important advances have been made in reducing emissions from diesel trucks by improved en- gine design, low-sulfur fuels, and especially diesel particulate filters.

However, studies show that highways continue to be a source of both coarse and fine particulate matter (PM) including known toxics such as diesel exhaust (Zhu et al., 2002; Karner et al., 2010; Cahill and Cahill, 2013). In addition, particles associated from vehicular wear such as from brake pads and drums are currently roughly equal to tailpipe emissions and are on track to exceed tail pipe emissions in the near future (Denier Van der Gon et al., 2013). Only limited data ∗ ∗Corresponding author.

Email address: [email protected] (T.A. Cahill)
are available on wear particles by size and composition, but a major source is known to be ultra-fine metals from abrasion of brake pads and drums (Cahill et al., 2014).

In vitro laboratory studies of nanoparticles raise concerns about the health impacts of highway emissions, especially very fine and ul- tra-fine metals due to their propensity for penetrating to the deepest portions of the lung and diffusing into the circulatory system (Lewis et al., 2005; Chen and Lippmann, 2009; Lippmann, 2009; Oberdorster et al., 2007; Denier Van der Gon et al., 2013). Additionally, epidemi- ological health studies continue to show that highway emissions are damaging to both pulmonary and cardiovascular systems of people living near highways (Cahill et al., 2011), with reduced lung function in children (Peters et al., 1999a,b); Gauderman et al., 2000; HEI, 2009; Lin and Peng, 2010; Karner et al., 2010). These data have been com- bined with data on the health impacts of roadway pollutants, including potential cancer impacts largely from diesel exhaust, and used to gen- erate estimated health impacts in models such as Emfac2007 (ARB, 2007).
http://dx.doi.org/10.1016/j.atmosenv.2016.09.023
1352-2310/© 2016 Published by Elsevier Ltd.

Cahill -- 00000324
AR002331
,: -· -- --

ELSEVIER

UNCORRECTED PROOF
2 Atmospheric Environment xxx (2016) xxx-xxx
A recent European survey noted that wear particles were approach- ing and would soon surpass exhaust particles near roadways, with un- certain health impacts. Their consensus statement concludes,“In light of the continuous increase of the relative contribution of non-exhaust emission to ambient PM, where it is becoming the dominant emission process for urban transport, it is more than timely to devote greater efforts to properly quantifying non-exhaust emissions and assessing health relevance.” (Denier Van der Gon et al., 2013).

The U.S. Environmental Protection Agency (EPA) and the Uni- versity of Michigan conducted the Near-road Exposures to Urban Air Pollutants Study (NEXUS)– a research project to study the impact of vehicle emissions on near-road air quality, human exposures, and po- tential health effects. Components of NEXUS include but are not lim- ited to human exposures in near-road residences (Vette et al., 2013), studies of near roadway allergic impacts in mice (McGee et al., 2015).

The NEXUS aerosol component reported in this publication was de- signed and conducted to update our information on highway emissions and their potential human impact with a component on vehicular wear emissions.

1.1. Experimental methodologies

1.1.1. Sampling array
Particles were measured from September 21 to October 30, 2010 on both sides of the Interstate 96 freeway 9 miles west of downtown Detroit. The I-96 monitoring location was established as part of a col- laborative research study conducted by the U.S. Environmental Pro- tection Agency (EPA) and the U.S. Department of Transportation Fed- eral Highway Administration (FHWA) (US EPA, 2013). The test sec- tion was a 12-lane section of I-96 with an at-grade roadway config- uration aligned east to west. Obstructions from vegetation between the test array of four sites, which extended from 100 m south of the highway to 300 m north of the highway, were negligible, as seen in Fig. 1. As part of NEXUS, additional PM monitoring was conducted at these sites.

Particles were collected in 8 size ranges at sites 100 m south, 10 m north, 100 m north, and 300 m north of the highway, while ultra-fine particles were continuously collected at the 10 m north and 100 m north sites. (Table 1)All sampling sites were on the top of trailers or one-story buildings with minimal obstructions (approximately 4 m from ground to inlets). The exception was the 100 m south site that had heavy trees south of the site, which did not obstruct the air flow from the highway.

1.1.2. Weather
Wind direction and wind speed were measured continuously at the 100 m north site (Fig. 2). During most of the study winds were weak, and extensive periods of calms occurred, especially at night. Wind di- rections were most frequent from the northwest (315°) through north northeast (35°). Given the southeast to northwest orientation of the sampling array, an upwind/downwind analysis of I-96 traffic emis- sions is best conducted when winds are from the southeast to south- west sector.

Thus, in order to address the prime goal of the study, a small frac- tion of the entire five-week period was chosen for which there was a clean upwind-downwind profile along the array. Since the goal was to quantify aerosol transport downwind of the freeway, results were limited to periods when the upwind-downwind trajectory was within ±67° of the north-south axis across the east-west freeway with wind speeds >1 km/hr. Periods of extreme wind or rain events were ex- cluded as well as wind trajectories pointing back to strong upwind sources such as the Monroe coal-fired power plant.

In terms of duration, while any 3 h period provides a valid signa- ture in time, the ideal is several days in a row. The reasons include re- ducing the inherent 1½ hr uncertainty in the elemental data set by the width of the x-ray analysis microprobe to roughly 30 min, which al Fig. 1. Monitoring site transect array across I-96 west of Detroit established for the EPA/FHWA National Near Road Study and used as a component of the NEXUS study. I-96 runs east-west and Telegraph Road is the multi lane road running north-south to the west of the sampling sites.

Cahill -- 00000325AR002332

UNCORRECTED PROOF
Atmospheric Environment xxx (2016) xxx-xxx 3
Fig.

2. a) Wind direction and rainfall, and b) wind speed, with the dotted box showing the periods with the wind from the south, 180° ± 67°, measured at the 100 m north site. Data below the dashed line are calm periods with wind speed <1.0 km/hr. lows better accord with meteorology and traffic patterns. It also allows better sensitivity by summing the periods for averaging actions such as size profiles. Limiting the intensive to week days avoided the com- plexity of highly variable weekend conditions, especially in terms of the truck traffic.

1.2. Experimental techniques

1.2.1. Sample collection
The primary aerosol collection instrument was the UC Davis DELTA Group rotating drum (DRUM) impactor (Cahill et al., 1985; Raabe et al., 1988; ) delivering aerosol samples in 8 size ranges: 10 to 5.0, 5.0 to 2.5, 2.5 to 1.15, 1.15 to 0.75, 0.75 to 0.56, 0.56 to 0.34,

0.34 to 0.26 and 0.29 to 0.09μm aerodynamic diameter. Samples were impacted at the rate of 4 mm/day onto 168 mm long, lightly greased Mylar foils, which were then transferred to plastic frames for analy- ses. Excellent agreement (r 2 = 0.99, slope = 0.99 ± 0.011, and inter- cept = 14 ng/m3) was seen in a side by side comparison of the DRUM mass (summed over all stages including ultra-fines) versus a Califor- nia Air Resources Board PM 2.5 Federal Reference Monitor (Cahill et al., 2014; Nichols, 2009, Supplemental Materials A).

For this study, a newly developed continuous ultra-fine stage was added at the 10 m north and 100 m north sites. These allow continu- ous collection of ultra-fine (<0.09 μm) particles directly after the last stage of the DRUM and keyed to deliver a stretched Teflon filter strip exactly matching in time the rotating drum stages (Cahill and Barnes, 2009).

1.2.2. Sample analysis
Details of the accuracy and precision of the mass, optical, and elemental analyses of the DRUM are included in Supplemental Materials. The measurements were made in 3 h increments using 500 μm wide excitation sources, and included:
• Mass (soft beta ray transmission, Ni63 source, MDL = 0.7 μg/m3). • Elements sodium through molybdenum, plus lead (Synchrotron-in- duced X-Ray Fluorescence (S-XRF), MDLs typically ∼ 0.1 ng/m3). • Diesel soot (optical attenuation, 380–820 nm, 50 nm bites, Ocean Optics spectrophotometer).

Cahill -- 00000326
AR002333
- Wind direction at 100 m north • Rainfall (cm x 100)
vi ., ., c,
~ 200 - - --++-+-+-,1-H+---_.,.._---._..---+<,_.., ___ _.._. ___ --ll- ___ -lH+-H-> ------«---¥--'- - ----l-+ '----IH C:
,g
<J
~ a 100
0 29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1
October , 2010
Wind speed, 100 m North
12 10
vi
]_ 8
-c .,
Q) 6 C.
"' -c
C: 4 ~
2 0
l
I ~
I
I J ~ .I ,I 1/
~ ~ ~~
~I
~ ~ ~ I
-•-~,-- ------ .... -- - -----------
29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
October, 201 O

UNCORRECTED PROOF
4 Atmospheric Environment xxx (2016) xxx-xxx

1.3. Black carbon data
Measurements of black carbon were also collected at the I-96 sites based on optical absorption (880 nm) using a Magee Scientific rack mount aethalometer (US EPA, 2013) and were compared to the DRUM optical attenuation.

1.4. Traffic data
Traffic volume during the study period showed a highly repro- ducible weekly pattern (Fig. 3). Weekday traffic averaged 175,000 cars and light trucks/day along with 12,000 heavy-duty trucks/day, or 6.6% trucks. Weekend traffic averaged 110,000 cars and light trucks/day, and about 3% trucks. Traffic data were collected as part of the EPA/FHWA study in 5 min increments individually for each of the 12 traffic lanes and for 6 classes of vehicle by length: 10 ft. (3 m), 30 ft. (9 m), 40 ft. (12 m), 50 ft. (15 m), 60 ft. (18 m), and 70 ft. (21 m) to 80 ft. (24.2 m), the maximum length allowed (US EPA, 2013). Traffic moved freely on the test section during the study, with a mean speed of 69 ± 3 miles/hr (110 ± 5 km/hr), with little braking activity. Below we plot the data reduced to one hour averages over all 12 lanes, using the 10 ft. (3 m) and 30 ft. (9 m) categories for cars and pickup trucks (light-duty vehicles or LDVs, mostly gasoline powered) and 40 ft. (12 m) to 70 ft. (24.2 m) for heavy trucks (mostly diesel).

The very strong morning and afternoon peaks seen in the light duty vehicle (LDV) data is much less pronounced in the truck data, which tends to be more uniform all day. Traffic data were not available be- fore Sep. 29.

1.5. Results of the 5-week study
The meteorology encountered during most of the study resulted in aerosol concentrations that, even at the close-in 10 m north site, displayed little or no correlation with the local traffic volume (Fig. 3). In Figs. 4–10 below we show four examples of typically regional particles and three examples of particles likely to have local road- way influence. Regional particles include PM 10 and PM2.5 mass, sil- icon (soil), sulfur (sulfates) and fine potassium (wood smoke). Par- ticles with known roadway sources include zinc in coarse mode Fig.

3. Traffic measured on I-96 west of Detroit on the EPA/FHWA National Near Road Study site. Heavy diesel truck numbers are scaled by a factor of 15 to facilitate comparison.

Weekends occurred on Sept. 25–26, Oct. 2–3, Oct. 9–10, Oct. 16–17, Oct. 23–24, and Oct. 30–31.

Fig.

4. PM10 and PM2.5 mass (without ultra-fines < 0.09 μm), at the 10 m north site. Each data point represents a 3 h measurement. The notation “tm” denotes a 6 mm long integral blank for timing and background subtraction validation.

Cahill -- 00000327
AR002334
20 ~----------------------------~
~
::,
0 .c
~
"' C..:;
Cl> C
~ Heavy diesels> 40 ft (x 15) - Cars and light trucks < 40 15 t------,-,----------++....--:----r--.-r+-11------ --!T-t-,------, § ~ 10 f--llH .... II-- -H----.im-lls-tl- - --...llHIIHll ~ - D-IIHl-1-111 - ~ -IHll-a..._ -----, ~~
<.>
If=
~ 5
30 25
l~ 2D
16 10
6 0
29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1
October 2010
- FMIOnuss · - • - F'llll2.6 nuss
~ ~ ~ 25 ~ ~ 2 4 6 8 ~ fl ~ ~ ff ~ ~ ~ 25 ~ ~
~ Z 25 ~ ~ 1 3 6 7 9 ~ ~ ~ ~ ~ 2D ~ ~ 25 ~ ~
S::fAe,, let O:tdJer

UNCORRECTED PROOF
Atmospheric Environment xxx (2016) xxx-xxx 5
Fig.

5. Silicon from 10 to 0.09 μm derived from soil. The notation “tm” denotes a 6 mm long integral blank for timing and background.

Fig.

6. Sulfur (for sulfate, x 3.0) at 10 m north.
(5.0–1.15 μm) from tire wear, zinc in the very fine mode (0.26 > Dp > 0.09 μm) from motor oil combustion, and black carbon (diesel soot) (Malm et al., 1994; Supplemental Materials 1).

Zinc is seen routinely both in transects from the roadway and in the ambient atmosphere. Zinc at the 10 m north site was highly variable in concentration, with highest concentrations typically occurring when the site was upwind of I-96. Note that unlike the potassium in Fig. 7, there are strong sources of coarse zinc not associated with the very fine zinc. Since there were no known sources close to the NEXUS site, the calculated rate of settling favors regional industrial processes with stack emissions.

Optical absorption was measured by two very different techniques, namely light transmission through the DRUM Mylar foils using a multi-wavelength optical spectrometer (Fig. 9), and direct measure- ment of optical absorption (b abs) from an aethalometer (Fig. 10; data provided by the EPA). The average b abs in units of ng/m 3 of Black Carbon (BC ≅ soot) for the entire period was 648 ng/m 3 at 100 m south, 1014 ng/m3 at 10 m north, 638 ng/m3 at 100 m north, 438 ng/m3 at 300 m north, showing the impact of soot from the freeway.

1.6. Results of the 3-day intensive
The conclusion from the aethalometer data is that even close to the freeway, regional sources are a major factor compared to roadway par- ticles for much of the five week study. Thus, for the remainder of this report, we will focus on those periods during which the criteria for a clean upwind-downwind profile were achieved. The results of the transects can then be compared to the regional particles seen during the entire study, isolating roadway impacts.

In order to identify those periods that met these criteria, nonpara- metric trajectory analysis (NTA) was performed using the meteoro- logical data collected at the 100 m north site (Henry et al., 2011).

Cahill -- 00000328
AR002335
1000
900
C"l 800
-E 700
II)
E 600
lU 500 c,
0 400
C:
300 ttl
z 200
100
0 800
700
C"l 600 -E
II)
500 E
ttl 400 C)
0 300 C
ttl 200 z
100
0 Detroit transect 1 0 m north
Silicon (for soil, - x 4)
- - 10 to 5 .0 - 2.5 to 1.15 - 0.75 to 0.56 - 0.34 to 0.26 - 5 .0 to 2.5 - 1.15 to 0.75 - 0.56 to 0.34 - 0.26 to 0.09 22 24 26 28 30 2 4 6 8 10 tm 12 14 16 18 20 22 24 26 28 30 23 25 27 29 1 3 5 7 9 11 tm 13 15 17 19 21 23 25 27 29 31 September October
Detroit transect 10 m north
Sulfur(for ammonium sulfate x 4.125)
- -10 to 5.0 - 2.5 to 1.15 - 0.75 to 0 .56 - 0.34 to 0.26 - 5.0to2 .5 - 1.15to0 .75 - 0.56to0 .34 - 0.26to0 .09
23 25 27 29 1 3 5 7 9 11 11 13 15 17 19 21 23 25 27 29
22 24 26 28 30 2 4 6 8 10 tm 12 14 16 18 20 22 24 26 28 30 September October

UNCORRECTED PROOF
6 Atmospheric Environment xxx (2016) xxx-xxx
Fig.

7. Potassium at 10 m north, with coarse modes derived from soil and fine modes largely from wood smoke and industrial combustion processes.

Fig.

8. Zinc from 10 μm to 0.09 μm at 10 m north.

Hourly averaged met data were aggregated into 3 h averages to bet- ter match the 3 h aerosol data. Fig. 11 shows the fraction of time in each 3 h period when the wind was blowing across I-96 from the south to the north. Upon early analysis of the data, an additional criterion was added, namely avoidance of sulfur (MDL = 5 ng/m 3) and sele- nium-rich (Se and As, MDL = 0.1 ng/m3) particles transported by SSE winds that passed over the Ford Rouge River complex, other indus- trial sources, and the large Monroe coal fired power plant 34 miles (54 km) south (185°) of the array (Fig. 6). This eliminated the po- tential October 25 day, while October 23 and 24 were both weekend days and impacted by upwind aerosol sources. October 30 was also eliminated as a weekend day. Although October 8 and 20 met the trajectory criteria, both had extended periods of calm winds making clean upwind-downwind analysis uncertain. A final criterion for the upwind-downwind study was that the 10 m north site should be mini- mally impacted by Telegraph Road, a multi lane road west of the I-96 sites (Fig. 1). Fig. 12 shows the subset of trajectories for which the 10 m north site was downwind of I-96 and without significant influ- ence from Telegraph Road.

Although there was some influence of Telegraph Road traffic on Oct. 27 and 28 at the 10 m north site, the distance to Telegraph road was much farther than at the 100 m north site and a belt of trees in a residential neighborhood interfered with wind and aerosol transport. (Fig. 1). Also, Telegraph Road traffic volume was a tiny fraction of I-96 traffic volume. For these reasons, the impact should be small, confirmed by the low aethalometer soot readings (Fig. 13 below). The 300 m north site, on the other hand, is much closer to Telegraph Road and was probably significantly impacted by Telegraph Road traffic for much of the study period. The longest period during the NEXUS cam- paign that met all criteria for a north-south transect occurred from Oct 26 through Oct. 28, 24–3 h periods, shown in the dashed box in Fig.

11. Cahill -- 00000329
AR002336
150
C')
_g 100
"' E
!!!
0)
0 C: 50 RS
z
40 (')
-E 30
(I)
E
~ 20
C)
0 C:
~ 10
0 0
Detroit transect 10 m north
Potassium
- - 10 to 5.0 - 2.5 to 1.15 - 0 .75 to 0 .56 - 0.34 to 0 .26 - 5 .0 to 2 .5 - 1.15 to 0 .75 - 0 .56 to 0 .34 - 0 .26 to 0 .09 22 24 26 28 30 2 4 6 8 10 tm 12 14 16 18 20 22 24 26 28 30 23 25 27 29 1 3 5 7 9 11 tm 13 15 17 19 21 23 25 27 29 31 September October
Detroit transect 10 m north
Zinc
- -10 to 5.0 - 2.5 to 1.15 - 0.75 to 0.56 - 0.34 to 0.26 - 5.0 to 2.5 - 1.15 to 0.75 - 0.56 to 0.34 - 0.26 to 0.09 21 23 25 27 29 1 3 5 7 9 11 12 14 16 18 20 22 24 26 28 30 22 24 26 28 30 2 4 6 8 10 tm 13 15 17 19 21 23 25 27 29 31 September October

UNCORRECTED PROOF
Atmospheric Environment xxx (2016) xxx-xxx 7
Fig.

9. Optical extinction at two wavelengths for very fine particles at the 10 m north site.

Fig.

10. Optical absorption (babs) from an aethalometer at both 10 m north and 100 m south.

The wind speed for the 3-day intensive averaged 10.7 ± 3.0 km/ hr, and the average direction was SSW, 208± 43°. However, the 10 m north site was in the turbulent mixed zone of the northernmost lane of the freeway, as evidenced by the winds from passing trucks. Thus, this site was impacted by one lane of the freeway under most wind condi- tions, representing only a small fraction of all traffic volume on the 12 lane freeway.

1. Validation of the meteorological analysis
Validation of the Oct. 26–28 choice for the upwind-downwind study was obtained by correlation of soot and typical roadway parti- cles with the daily daytime traffic maxima as shown in Figs. 13 and

14. The days chosen for the intensive are in the dotted box area. Data are from the 10 m north site.

During the intensive period, the calcium (roadbed abrasion) and iron (vehicle “wear”) particles (Fig. 14a and b) are correlated with traffic volume (Fig. 14d), with daytime peaks and nighttime low val- ues, unlike during most of the prior sampling period. Both elements are strongly enhanced over Earth crustal average, calcium by a factor of 5, and iron by a factor of 10, suggesting resuspension of roadbed concrete erosion and vehicle “wear” particles by wind and vehicular turbulence. There was almost no braking of vehicles during the study, so concentrations of the very fine Fe, Ni, Cu and Zn particles charac- teristic of braking (Cahill et al., 2014) are limited.

Highway zinc coarse particles (Fig. 14c) are enhanced by the zinc in tires and brake wear, while fine and ultra-fine zinc are observed in both brake wear and the burning of the zinc thiophosphate stabilizer used in many motor and lubricating oils, including diesels (Zielinska et al., 2003).

2. Size profiles of the elements
Fig. 15a, b and c plot particle size distributions during the intensive for crustal species (a), transition metals and lead (b) and sulfur (c). The mean size distributions for crustal elements peak in the supra-micron particles, indicating a mechanical source, such as mechanical wear and resuspension of contaminated soils. Iron is very interesting, as it peaks in the 2.5–5.0 μm mode but in concentrations far in excess of what would be expected in soil.

Cahill -- 00000330
AR002337
2 8 1.5
u
C:
~
o'i c.,
a
0 0.5
0 Very fine aerosols 0.26 > Dp > 0.09 micrometers
m Far red , 780 nm - Far blue, 380 nm
21 23 25 27 29 1 3 5 7 9 11 tm 14 16 18 20 22 24 26 28 30 1 22 24 26 28 30 2 4 6 8 10 Im 13 15 17 19 21 23 25 27 29 31 2 September October
Aethalometer Data
- 10 m north D 100 m south

6.-------------------------------,
5 1--------- -+------------------ -,----,
~4---------~~--------------------, </I C:
~~ 3----------~---------~----------, J:l 8
~2---•~----~--~---,--~---~~-----~---,
28 30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1 3
27 29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2
September October November

UNCORRECTED PROOF
8 Atmospheric Environment xxx (2016) xxx-xxx
Fig.

11. Trajectory analysis for the 100 m north site isolating those periods when the northern sites were approximately downwind of I-96 and the 100 m south site was a clean upwind background site. The period chosen for the intensive is shown in the dashed box, October 26 through October 28.

Fig.

12. Trajectory analysis isolating those periods when the 100 m south site was a clean upwind background for 100 m north. The 10 m north site had less Telegraph Road interfer- ence.

Size distributions for transition metals (Fig. 15b) also peak in the supra-micron range. Iron (brake drums and rust), copper (brake pads), zinc (stabilizing agent in lube oil, tire wear), and lead (“legacy lead ” from past emissions) are known to be associated with traffic (Cahill et al., 2011). Note that in addition to a coarse mode (re-suspended road dust), zinc, lead and copper also exhibit an ultra-fine mode. Very fine and ultra-fine zinc is most likely associated with lubricating oil.

A very different profile is shown for sulfur (Fig. 15c). Typically the accumulation mode around 0.5 μm represents regional ammonium sulfate, while very fine and ultra-fine sulfate has local combustion ori- gins, including from diesel combustion in trucks and cars.

3. Transects in the three day intensive: Coarse (5.0 > D p > 2.5) parti- cles – experimental
For the time-resolved coarse particles, we used only part of the distribution since the time resolution of the coarsest DRUM stage, 10 to 5.0 μm, was large enough to blur the day/night differences. (Supplemental Materials B).

Results of the highway transect (Fig. 16) show highest concentra- tions at the near freeway site at 10 m north, supporting traffic-derived sources. Concentrations fall off from the roadway to 100 m north, but concentrations of the coarsest particles increase again at 300 m north, especially for crustal species. The presence of the heavily traveled Telegraph Road approximately 380 m west of the 300 m site (Fig. 1) as well as local exposed soils near the site might be contributory fac- tors. In addition, the sampler inlet cut point may have been raised by an estimated 10% due to a non-standard, slightly wider inlet slot. This effect disappeared for the PM2.5 particles which are not influenced by coarse local soils.

4. Transects in the three day intensive: Coarse
(10 μm > Dp > 1.15 μm) particles – theoretical
I-96 traffic during the intensive was typical of the entire 5 week period (Fig. 2). For modeling purposes, traffic data for the intensive was broken into two periods: daytime (6 a.m. through 9 p.m.), and nighttime (9 p.m. through 6 a.m.). Daytime traffic averaged 10,479 Cahill -- 00000331
AR002338
1 "C
C
"j

0.8 ~ 0
"C

0.6 a,
:§
0 0.4
C
0 ~ 0.2
~
IL
0 1 ,::,
C
"j 0.8 C
~ 0 ,::,

0.6 a,
~
0 0.4
C
.Q
ti
~
0 .2
LL
0 D Trnjectories downwind of 1-96 and Telegraph Road
I
- f-- f--
rvr fl )_ f ~-
I
- f-- - I 1" - -I
I
~- -- ~ - I 11- -~ I
I
I
~
-- - ~ I r
n-
I
j I
I
29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2
30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1 3
-
-
-
-
September October November
D Trajectories downwind of 1-96 but not downwind of Telegraph Road I
I
I
I
I
I
I
I
I
I
t
I
~
I _,_
A
I
~ I I\ J
I
I 1 /,,,.
29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2
30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1 3
-
September October November

UNCORRECTED PROOF
Atmospheric Environment xxx (2016) xxx-xxx 9
Fig.

13. Aethalometer data on optical absorption (∼soot) from the EPA/FHWA National Near Road Study. The days chosen for the intensive are in the shaded area. October 25 and earlier days were rejected because of substantial upwind aerosol pollution seen in Figs. 5–7. cars and light trucks/hr, and 824 heavy trucks/hr, while the numbers at night were 3874 and 187 vehicles/hr respectively.

For the theoretical analysis of the data, all coarse particles from 10 to 1.15 μm were used so as to allow comparisons to other work.

We used a Sliding Box Model Estimate of the I-96 10 to 1.15 μm mass emission rate for the 3-day intensive. The “sliding box model ” (Cahill and Feeney, 1973, Feeney et al., 1975; Courtney et al., 1978) is a well-tested way to measure roadway emissions when one has ac- cess to both upwind and downwind information on pollutants, a flat terrain, and a lateral wind across the roadway. The dimensions of the box are set by the roadway width, including lateral turbulence, and the top of the mixed zone, set by vehicle height and velocity. The lat- ter is the greatest uncertainty, but extensive data on the height of the mixed zone were taken in Los Angles in 1973 and used in several stud- ies. The length of the box is arbitrary, and we use 1.6 km, which was needed since the project accepted winds at ± 45° from normal and still had to meet the line source assumption even 300 m downwind.

The box so defined represents a volume into which pollutants are uni- formly mixed by vehicular turbulence. The box then slides laterally across the freeway while vehicles are emitting into the box. A new box takes its place upwind in a smooth transition and the process repeats.

The model was validated with data taken in 80 two-hr up- wind-downwind periods of lead transport from five freeway sites in 1972 (Cahill and Feeney, 1973; Feeney et al., 1975). Since accurate traffic data and meteorology were available and emissions of auto- motive lead were precisely known from the literature (Habibi, 1973), the sliding box model gave a roadway edge prediction at the at-grade freeway section of 4.0 ± 0.4 μg/m3 for 5000 vehicles/hr. The three day average measured at the at-grade site was in excellent agreement,

4.0 ± 0.15 μg/m3. The sliding box model was also successfully ap- plied in General Motors test facility studies of sulfur from catalytic converters (Courtney et al., 1978)
Sliding box model calculations were run separately for daytime, 6 a.m. to 9 p.m., and nighttime, 9 p.m. to 6 a.m., keyed to the traffic flow. Traffic flow, wind velocities were also separated in the analy- sis. It was assumed that trucks had 10 times the emission rates as cars (Gertler et al., 2003). Table 2 below shows the analysis for cars and trucks for calcium, and then the net result for calcium, iron, and zinc.

Coarse calcium and iron are enhanced by enrichment factors 5 and 10 times Earth crustal averages respectively (Fig. 14a and b), likely a consequence of the erosion of the concrete highway and iron-rich debris from vehicles. The transition and heavy metals, vanadium, copper, zinc, and lead, are enormously enhanced compared to Earth crustal averages and represent vehicular debris, including brake pads, brake drums and tire wear.

In this analysis, the unknown emission rates were calculated by matching the predicted concentrations to the measured concentrations at the 10 m north site after correcting for concentrations at the 100 m south site.

In Table 2, the emission rates in mg/km were fit to the NEXUS background corrected concentrations (μg/m 3) at 10 m north (Fig. 14a–c). Matching the upwind-corrected data at the 10 m north site for freeway impact to the predicted concentrations, emission rates of 1.5 mg/km for calcium, 2.8 mg/km for iron, and 0.09 mg/km for zinc are calculated, with truck emission rates 10 times these values.

Direct comparisons are possible to similar analyses in Los An- geles in 1972 from an at-grade concrete freeway. The NEXUS free- way-sourced calcium level in the mixed zone, 0.24 μg/m3, is about 1/ 5 of that seen in 1973, 1.13 μg/m3. Note that the Los Angeles freeway had not seen rain in over 2 months, while rain fell about once/week in the NEXUS study. For iron, which has both a concrete component and one associated with vehicle wear, the NEXUS values are 0.52 μg/m3, versus 1.43 μg/m3 in Los Angeles. For zinc, the NEXUS level of 0.017 μg/m3 is far less than the 0.22 μg/m3 seen in Los Angeles in 1973.

The California data (Cahill et al., 2003) for 15 > D p > 5.0 μm parti- cles from tire wear were run through the sliding box model of Table 2 and yielded 120 ng/m 3 of zinc, about 10 times what was observed in NEXUS for < 5.0 μm particles (Fig. 14c). However, the upper cut point for the Los Angeles particles was ∼15 μm, which increased the numbers, while better high mileage tires and aerosol size differences reduce the NEXUS value.

A very similar protocol (upwind-downwind samplers, rotating drum impactors, x-ray analysis) was utilized in Switzerland (Bukowiecki et al., 2010) for an extensive series of measurements to separate vehicular wear from exhaust emissions. The Swiss results showed most vehicular wear particles in the 10 to 1 μm range. The vehicular wear results were 50 ± 13 mg/km per vehicle for light duty Cahill -- 00000332
AR002339
4 3
"' ~
~ lJl 2
"' :J .0 0 .c
I-
24 Aethalometer Data
10 m north - - - 100 m north ~ 100 m south
25 26 27
October
28 29 30

UNCORRECTED PROOF
10 Atmospheric Environment xxx (2016) xxx-xxx
Fig.

14. a, b, and c. Typical roadway particles during the Oct. 26 through Oct 28 intensive (in shaded box). d. Traffic from Fig. 3 on the same time scales (trucks x 15). vehicles, 288 ± 72 mg/km for heavy duty vehicles, with 3% brake wear, 56% suspended/abraded road dust, and no evident tire wear.

This gives roughly 3.2 mg/km brake wear, assuming 6% trucks in their mix. If the NEXUS iron and zinc were all from brake wear, it would amount to roughly 5.2 mg/km, similar to the Swiss results.

This is an overestimate, since the NEXUS iron value should be re Cahill -- 00000333
AR002340
..,
900
800
700
E soo 'in
5 500
~ 400
0 fi 300
z
"" E "ii;
E
I!
Cl
0 fe :z
"" ~ E
I!
Cl
0 C:
"' :z
200
100
0 500
400
300
200
100
0 14
12 10
e
6 4
2 0
15 "'10 -0
<=
"' <J)
::,
0 I= 5
Calcium
-- 5.0 to 2 .5 - - 2.5 to 1 .15 -- 1.15 to 0.75 micrometers ~------------------~--------------------~------,
23 23
...
24 24
26 Iron
-- 5.0 to 2 .5 · - - 2.5 to 1 .15
27 28
1 .15 to 0. 75 micrometers
I ✓
.l...l.lww 1..lal..l...l.L~ ' ' .I..J..,.I,,,~---- -~~~~~~ 25 27 29
26 28
Zinc
-- 5.0 to 2.5 - - 2.5 to 1.15 -- 1.15 to 0.75 micrometers ~------------------~--------------------~------~
~
23 24 25 26 27 28 29
October
c::J Truck traffic (scaled x 15) -- Car Traffic
24 25 26 27 28 29 30
Octo ber

UNCORRECTED PROOF
Atmospheric Environment xxx (2016) xxx-xxx 11
Fig.

15. Size distributions of a) crustal species; b) transition metals and lead; and c) sulfur at the 10 m north site during the NEXUS intensive. duced due to iron from vehicle debris and roadbed wear.

5. Transects in the three day intensive: Fine PM 2.5 particles – experi- mental
Fig. 17a and b shows transects for fine crustal species and fine transition metals and lead, respectively, during the Oct. 26–28 inten- sive. Fine crustal species, like coarse, show a near-freeway enhance- ment, especially for iron, enhanced by a factor of 20 over typical soils. The fall-off versus distance is slower than with coarse species, with aluminum and silicon being essentially the same at the 10 m north and 100 m north sites The 300 m north site is almost identical to the 100 m south site, consistent with a regional background.

The data clearly show the impact of highway derived particles.

First, the aluminum and silicon have about the correct ratio for Earth's crustal averages, and therefore represent crustal materials either in the roadway aggregate or the surrounding soils. The parking Cahill -- 00000334
AR002341
300
('") 250
,€ 200 VJ
E
~ 150 o
0 100 c;
~
z 50
0 10
..,
E 8
"iii
E 6
~ ...

Cl 4 0 c;
~
2 z
0 100
90 § 80
VJ 70
E 60
:! 50
g, 40
ij 30
z 20
10 0
10 m north, south winds Oct. 26 - Oct. 28
--- Silicon • .. Calcium · ... Iron -- Potassium
-.. - •""
< 0.09 0.25 to 0.34 0.56 to 0.75 1.15 to 2.5 5.0 to 10

0.09 to 0.26 0.34 to 0.56 0.75 to 1.15 2.5 to 5.0
Particle aerodynamic diameter (micrometers )
10 m north, south winds Oct. 26 - Oct. 28
- Copper -,...... Nickel ..... · Chromium ·- • Lead
Zinc -- Vanadium
/---
/
'
◊
/ '- .,, .
/
" / - _. - v \
.._ ..., ,.... _, v· '-~' - .,,,,,,,,,,., _ a, ,, _ ...... _ \ ............. '- -- - - - .o- - ... ..,,v- - ~ __,,~ ~
- ....... •
~
< 0.09 0.25 to 0.34 0.56 to 0.75 1.15 to 2.5 5.0 to 1 O

0.09 to 0.26 0.34 to 0.56 0.75 to 1.15 2.5 to 5.0
Particle aerodynamic diameter (micrometers )
10 m north, south winds Oct. 26 - Oct. 28
_._ Sulfur (for ammonium sulfate, x 4.125)
..
" I
I
' I•
' .,, ., ..... "'\
I T
' ,
\ -T
' ., ' '
'
.,
' _.,
----__ .,.,.
~ ~
< 0.09 0.25 to 0.34 0.56 to 0.75 1.15 to 2.5 5.0 to 1 O

0.09 to 0.26 0.34 to 0.56 0.75 to 1.15 2.5 to 5.0
Particle aerodynamic diameter (micrometers)

UNCORRECTED PROOF
12 Atmospheric Environment
xxx (2016) xxx-xxx
Fig.

16. Coarse aerosol transect during the Oct. 26–28 intensive: a) crustal species; b) transition metals and lead.

Table
1 Distance of samplers from roadway median and nearest active traffic lane of the 10 lane freeway. We also include the relative roadway impacts on cancer from the widely used model Emfac2007 (ARB, 2007).

Site
Distance from
median
Distance from closest traffic
lane
Relative cancer
impact
100 m
south
92 m south 71 m south 0.13 upwind
10 m
north
26 m north 10 m north 1.00 downwind
100 m
north
142 m north 100 m north 0.25 downwind
300 m
north
332 m north 270 m north 0.10 downwind
area
just upwind of the 100 m north site may also be a source of such materials. Calcium, however, is enhanced by a factor of about 3 over Earth's crustal average, and is consistent with roadway abrasion from the concrete highway. Iron is enhanced by a factor of about 10, which is consistent with brake wear and resuspended rust in a region where road salting is a common source of vehicle degradation. Fine cop- per (Fig. 17b) also shows enhancement, possibly from brake pads. In summary, fine concentrations of these resuspended soil and vehicular wear particles are about ½ of the observed mass, similar to European estimates that “wear” particles are about equal to tailpipe emissions (Denier Van der Gon et al., 2013).

All the non-crustal elements as well as zinc show a near freeway enhancement during the Oct. 26–28 intensive. All are strongly en- hanced over what would be expected from soil, suggesting roadway pollution. The iron is about 16×the copper, while the iron/copper ratio from braking is only about 3. Thus, most of the iron is not from brake drums and pads, but probably represents re-suspended rust particles from degraded cars and trucks. Recall that little vehicular braking oc- curred in this freeway section. The 300 m north site is almost identical to the 100 m south site, establishing a small regional background. The fall-off versus distance for zinc is much slower, and the upwind value elevated, showing regional fine zinc aerosol sources. The enhanced zinc at 10 m north may be due to the zinc thiophosphate oil additive.

6. Transects in the three day intensive: Fine PM 2.5 particles – theoret- ical
In the sliding box model, we assumed an I-96 PM 2.5 mass emis- sion rate (row labeled “Source PM2.5 mass” in Table 3) based on re- cently measured car and truck emission rates for PM 2.5 in the eastern US (Gertler et al., 2003). The sliding box model was then used to cal- culate PM2.5 values downwind of the freeway which were compared to measured PM2.5 concentrations.

The results shown in Table 3 can be compared to the PM concen- trations measured at the 10 m north site which was just on the edge of the lateral turbulence zone. The 24 h average PM 2.5 estimated for the 3 day intensive was 1.6 ± 0.5 μg/m3, versus the measured 2.0 ± 0.7 Cahill -- 00000335
AR002342

5.0 to 2.5 micron aerosol transect during NEXUS
South winds Oct. 26 through Oct. 28
100 m S 10m N 100 m N 300 mN
Transect south to north

5.0 to 2.5 micron aerosol transect during NEXUS
South winds Oct. 26 through Oct. 28
14 ~------------------~
M 12 1-------- -------------- --1
E vi 10 t------- - - -"c-------------1
E S f----- ~ ~~.----4,-----------~ (II ,_
~ 6 1----- - - --~--_,,_ ________ ___.
0 ~ 4 f--- r'Cf-- -:7' ....... .c-- ~ ~ ------=,._.-=----~ z 2
0 100 m S 10 m N 100 m N 300 m N
Transect south to north
Aluminum
-Silicon
-Calcium
-Iron
-
Zinc
-Copper
Lead
-Vanadium
-lron/30
·-. -

UNCORRECTED PROOF
Atmospheric Environment
xxx (2016) xxx-xxx 13
Table
2 Sliding box model of coarse (10 > Dp > 1.15 μm) particles for daytime periods during the NEXUS 3 day intensive. LDV = Light duty vehicles, cars and light trucks (Class 1, 2); HDV = Heavy duty vehicles, trucks (Class 3–6).

Site I-96 Detroit I-96 Detroit I-96 Detroit I-96 Detroit I-96 Detroit Date October 26–28, 2010 Calcium Calcium Calcium Iron Zinc Time 15 h blocks Days 6–21 Days 6 –21 Days 6 –21 Days 6–31 Days 6–21 Vehicle type Classes LDV HDV Total cars and trucks Total cars and trucks Total cars and trucks Fit to data mg/km 1.5 15.0 1.5, 15 2.8, 28.3 0.09, 0.90 Box dimensions height m 5 5 5 5 5
width m 54 54 54 54 54
length m 1000 1000 1000 1000 1000
Box volume x 1000 m3 270 270 270 270 270
Traffic vehicles/hr 10,479 824 11,303 11,303 11,303
Speed mi/hr 69 69 69 69 69
Speed km/hr 110.4 110.4 110.4 110.4 110.4
vehicles in box # vehicles 95 7 102 102 102
Emissions mg/min 142 112 254 480 15.3
Concentration in box per minute μg/m3 0.53 0.41 0.94 1.78 0.057 Wind velocity m/second 3.1 3.1 3.1 3.1 3.1
Sliding box translation Seconds 17.4 17.4 17.4 17.42 17.4 Sliding box translation Fraction of minute 0.29 0.29 0.29 0.29 0.29 Calculated concentration at 10 m north μ μg/m3 0.153 0.120 0.273 0.517 0.0164 Measured concentrations 10 m north μ μg/m3 0.242 ± ±0.050 0.516 ± ±0.05 0.0169 ± ±0.008 Los Angeles 1972

5.0–15 μm mg/mi 6.2, 62 8, 80 1.2, 12
Concentration in mixed zone μ μg/m3 1.13 1.43 0.22
Fig.

17. Fine aerosol transect during the Oct. 26–28 intensive: a) crustal species; b) transition metals and lead.

Cahill -- 00000336
AR002343
PM2.5 ( 2.5 to 0.09) aerosol transect during NEXUS
South winds Oct. 26 through Oct. 28
M 250 t--------.,...--------------,
_§
r,i 200 t------- - - --"<-------------,
E r: 150 t------------~---- ..,.._-------------1
Cl
0 ij 100 1---- ~ ___,,,.,,,.==-----="""'--= c-'ll l.:?' "--.::---------i z
100 m S 10 m N 100 m N 300m N
Transect south to north
PM 2.5 (2.5 to 0.09) aerosol transect during NEXUS
South winds Oct. 26 through Oct. 28
100 m S 10 m N 100 m N 300 m N
Transect south to north
Aluminum
~
Silicon
-Calcium
Iron
-
Zinc
-Copper
Lead
-Vanadium
-lron/30
- . -

UNCORRECTED PROOF
14 Atmospheric Environment xxx (2016) xxx-xxx
Table 3
Sliding box model of the three day intensive, when meteorological and traffic conditions were suitable. Also shown are the Swiss data (2010) and Los Angeles data (1973) estimated from measured values.

Emission rate estimates Gertler et al., 2003
eastern US
Gertler et al., 2003,
eastern US
Gertler et al., 2003,
eastern US
Bukowiecki et al., 2010,
Swiss freeways
Cahill and Feeney, 1973,
LA freeways
Time 15 h
increment
Daytime 6 to 21 Daytime 6 to 21 24 h average 24 h average 24 h average Vehicle type Classes LDV HDV All vehicles All vehicles All vehicles Source PM2.5 mass mg/km 14 135 14/135 50/288 1056
Box dimensions height m 5 5 5 5 5
width m 54 54 54 54 54
length m 1000 1000 1000 1000 1000
Box volume x 1000 m3 270 270 270 270 270
Traffic vehicles/hr 10,479 824 7682 7682 7682
Speed km/hr 110 110 110 110 110
vehicles in box # vehicles 95 7 70 70 70
Emissions mg/min 1329 1008 1528 4538 17,636
Concentration in box per minute μg/m3 4.9 3.7 5.7 16.9 65.3 Wind velocity m/second 3.1 3.1 3.1 3.1 3.1
Sliding box translation Seconds 17.4 17.4 17.4 17.4 17.4 Sliding box translation Fraction of
minute

0.29 0.29 0.29 0.29 0.29
Calculated PM2.5 Concentration μg/m3 1.43, 1.08 0.53, 0.25 1.6 ± ±0.5 4.9 ± ±1.2 19.0 ± ±3 Size ranges μg/m3 PM2.5 PM2.5 PM2.5 PM10 PM5.0
Measured PM2.5 Concentrations
(some estimated)
μg/m3 2.0 ± ±0.7 3.1 ± ±1(est) 16 ± ±2 (est)
μg/m3. Note that the Gertler et al., 2003 emission values were for Cal- ifornia cars and fuels.

Table 3 also includes estimates of the 10 m north concentrations based on the Swiss (Bukowiecki et al., 2010) and Los Angeles (Habibi, 1973; Cahill and Feeney, 1973) freeway emission rates. The Swiss study gave PM 10 emission rates of 50 ± 13 mg/km for light duty vehicles, and 288 ± 72 mg/km for heavy duty vehicles. Apply- ing these emission rates to the NEXUS configuration, one would pre- dict about 4.9 μg/m3 PM10 mass at the freeway edge, slightly less than the 10 μg/m3 to 19 μg/m3 PM2.5 values reported as the traffic-im- pacted Central European average (Denier Van der Gon et al., 2013).

The Swiss data included about ½ roadway crustal particles, which would likely reduce their 4.9 μg/m3 PM10 values to around 3.1 μg/m3 of PM2.5.

The right hand column of Table 3 is the predicted 24 h PM5.0 mass at the 10 m north site using the emission rates from the 1973 Los An- geles study (Habibi, 1973). The mean estimated PM2.5 mass concen- tration of 19μg/m3, would also have included 2.7μg/m3 of lead. Mak- ing the correction for presumed coarse roadway wear particles, the 1973 PM5.0 data would be reduced to roughly 16 μg/m3 equivalent PM2.5 in the NEXUS configuration.

7. Transects in the three day intensive: Very fine
(0.26 μm > Dp > 0.09μm) particles
Very fine particles (Fig. 18) were less impacted by distance from the freeway compared to fine and coarse particles. These results rein- force the behavior seen in the Los Angeles I-710 studies (Fig. 19, Zhu et al., 2002) in which very fine and ultra-fine components extended both upwind and downwind from the array with little change in con- centration with distance from the freeway once one was farther away than about 150 m (see Fig.19 ).

1.7. Ultra-fine particles
The development of a means to measure ultra-fine particles as a function of time allows for the first time the ability to match com- positionally-resolved ultra-fine particles to potential sources and me- teorology. Two such units were built and tested to run behind the DELTA 8 DRUM, which has a validated lower cut point of 0.09μm.

Thus, for this study, ultra-fine particles are particles having aerody- namic diameters <0.09 μm. These two units were placed at the 10 m north site and the main site at 100 m north, and ran for 5 weeks to match the 8 DRUMs.

The ultra-fine data did not have an upwind site, but the agreement between the 10 m north and 100 m north sites is confirmation of a re- gional distribution. The major exception to this pattern was optical at- tenuation (Fig. 13), which in the 0.26 > Dp > 0.09 μm size mode is al- most entirely soot, largely from diesel trucks. We did not measure soot in the DRUM ultra-fine mode.

An examination of the 3 day intensive period at 10 m north showed that, of the four elements known to be present in the ultra-fine size ranges from traffic, zinc and iron show some modest correlation with daytime traffic peaks (Fig. 21). Recall, however, that relatively little braking occurs on this stretch of I-96.

Ultra-fine particles at the 100 m north site showed essentially no correlation with daytime traffic, in accord with the factor of ∼10 re- duction seen in previous studies (Cahill and Feeney, 1973; Zhu et al., 2002; Emfac2007). For diesel-generated ultra-fine sulfur, (Zielinska et al., 2004), the freeway contribution is largely lost in high levels of ul- tra-fine sulfur associated with upwind sources possibly including the Monroe coal-fired power plant.

2. Discussion
The particles in the NEXUS study ranged from coarse concrete erosion and resuspended roadway particles contaminated with tire and brake drum debris, to very fine and ultra-fine particles generated by traffic and diverse regional sources.

The behavior of the particles during the three day transect inten- sive was roughly as expected, with coarser particle concentrations de- clining rapidly downwind while fine and especially very fine particle concentrations dropped more slowly, as shown in Figs. 16–18.

The agreement between the predictions of the sliding box model and the observed near-roadway PM2.5 concentrations (Table 3) gives confidence that literature emission rates used in the model must be close to reality. The results also highlight the enormous improvement in vehicle tailpipe emissions in the past 40 years.

Cahill -- 00000337AR002344

UNCORRECTED PROOF
Atmospheric Environment xxx (2016) xxx-xxx 15
Fig.

18. Very fine aerosol transect during the Oct. 26–28 intensive. a) crustal species; b) transition metals and lead.

Fig.

19. Transect of Interstate 710 in Los Angeles from Zhu et al. (2002). Superimposed are lead data from 1973 (Cahill and Feeney, 1973; Feeney et al., 1975). The arrow marks the distance of the 100 m north site from the I-96 median in the present study.

The situation with the ultra-fine particles, on the other hand, in- dicates a dominance of regional concentrations little affected by the nearby freeway traffic with the striking exception of optical absorp- tion (diesel soot) as seen in both optical spectroscopy from the DRUM finest stages and the aethalometer data (Fig. 13). The ul- tra-fine particles have elemental signatures consistent with brake wear and zinc in motor oil, but the concentrations are small, consistent with little braking on that section of I-96 during the 3 day intensive.

Cahill -- 00000338
AR002345
C")
_§
(/J
E ro ...

Cl
0 C: ro z
(")
E
-(/J
E ('ll
I..

Cl
0 C:
('ll
z
"' <I)
:,
cij
>
<I)
>
~
<ii
Cl'.'.

Very fine (0.26 to 0.09) aerosol transect during NEXUS
South winds Oct. 26 through Oct. 28
100 m S 10 m N 100 m N 300m N
Transect south to north
Very fine (0.26 to 0.09) aerosol transect during NEXUS
South winds Oct. 26 through Oct. 28
5 4
3 2
0 ... 'Ill -
... ,,. - ... .. ,.... ... .. ... - -----...
----
~
---
......_
=:::::-.
100 m S 10 m N 100 m N 300 m N
Transect south to north
Aluminum
~
Silicon
-Calcium
-Iron
-
Zinc
-Copper
-Lead
-Vanadium
-lron/30
-•-
- CO - Black Carbon - Number - Lead (1974)
1000
900
800 1----------- •1--,,..._ __ _. Location of 100 111 north 700 site from I-96 rnedian
600
500
400
300
· - .... ■ - ■ - • ■ 11111 ■ •
.. ,. .
200
100
0 -240 -160 -80 0 80 160 240 320 400
-200 -120 -40 40 120 200 280 360
Upwind Distance from Freeway 710 (m) Downwind

UNCORRECTED PROOF
16 Atmospheric Environment xxx (2016) xxx-xxx
Fig.

20. Ultra-fine sulfur, potassium, and zinc at the 10 m N and 100 m N sites. The large ultra-fine sulfur peaks on October 24 through 25 were likely from the Monroe coal fired power plant. The Sept. 30 and Oct. 12 plumes were from a source to the north. Black lines above the dates in the zinc plot are weekends.

These three days were selected when there was minimal impact from Telegraph Avenue which had extensive stop and go periods. The ul- tra-fine data are a more extreme case of the behavior seen in the very fine particles, which had large regional sources and fell off slowly downwind.

The potential exposure and human health implications of these par- ticles can be estimated by comparison with recent work tying road- way-derived ultra-fine particles to ischemic heart disease in the Cal- ifornia Central Valley (Cahill et al., 2011). An increase in IHD mor- tality of ∼30% at Bakersfield, CA, was associated with elevated lev- els of very fine and ultra-fine roadway-derived transition metal parti- cles, mostly from brake drums and pads (Table 4 below). However, the concentrations in Detroit were generally far lower than those mea- sured in Bakersfield.

As shown in Table 4, ultra-fine concentrations measured at the 10 m north and 100 m north sites are similar to those at Watt Avenue (Spring) in Sacramento, CA measured 20 m downwind of a 65,000 ve- hicles/day secondary street, at a stop light, during non-inversion con- ditions (Cahill et al., 2014). In both cases, little influence was seen from the nearby roadway, and the ultra-fine particles were regionally distributed. However, there was only very limited braking and accel- eration during the NEXUS intensive, thus limiting concentrations of brake wear particles. Nevertheless, if the observed ultra-fine metallic particles see in the NEXUS experiment extend over large numbers of people, it could potentially enhance the ischemic heart disease death rate in the same manner seen in Bakersfield, CA (Cahill et al., 2011).

3. Conclusions
Important differences in the behavior of particles of different sizes were observed during this study. For coarse species, the fall-off in concentration versus distance was rapid, roughly in accord with cur Cahill -- 00000339
AR002346
(")
E <ii
Ii c,
0 ~ z
400
300
200
100
NEXUS ultra fines
Sulfur 100 m N ~ Sulfur 1 0 m N
24 26 28 3 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1 23 25 27 29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2 -- Potassium 100 m N C:=J Potassium 10 m N
22 24 26 28 30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1 23 25 27 29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2 12 10
-- Zinc 100 m N c:J Zinc 10 m N
24 26 28 30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 1
23 25 27 29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2 September October

UNCORRECTED PROOF
Atmospheric Environment
xxx (2016) xxx-xxx 17
Fig.

21. Ultra-fine nickel, copper, iron and zinc at 10 m north during the 3 day intensive.

Table
4 Comparison of ultra-fine (D p < 0.09 μm) traffic-derived particles measured in NEXUS versus Watt Ave. in suburban Sacramento, CA during spring non-inversion conditions (Cahill et al., 2014). NEXUS concentrations are 5-week averages. na = not available.

For comparison, winter values which are impacted by strong surfaced-based inversions in the California Central Valley are included. The sources of the high nickel values seen at Watt Avenue in winter were totally absent 500 m upwind, so they must be associated with Watt Avenue traffic. Such levels, however, have never been seen in our other stud- ies.

Ultra-fine elements (ng/m3) S P Fe (non-soil) Ni Cu Zn Pb Non-inversion conditions
Detroit
10 m n 68.1 10.5 3.4 0.5 1.9 3.2 2.9
100 m n 61.4 9.5 3.4 0.5 1.8 2.6 3.3
Watt Ave (Spring) 129 na 4.8 0.3 0.4 2.0 1.4
Strong inversions conditions
Watt Ave (Winter) 42.8 na 29.3 12.1 9.3 13.7 1.6
Sacramento (Winter) 52 4.0 14.3 0.6 na 14.3 2.5
Bakersfield (Winter) 242 10.5 27.7 1.6 na 12.3 2.6
rent
models such as Emfac2007. Concentrations were strongly af- fected by meteorology such as wind velocity and rain fall. For the PM2.5 particles, the fall off versus distance was still close to the models, reaching a 90% reduction in freeway-derived particles in the 200–250 m range. Very fine particles, 0.26 > D p > 0.09 μm diameter, persisted all the way to the 300 m downwind site, with only a modest reduction in concentrations. Ultra-fine particles were only measured at two sites, 10 m north and 100 m north, and the concentrations at both sites were almost identical indicating a very slow fall off and/or strong upwind sources. The persistence of ultra-fine particles in urban areas away from freeways was also observed in the Los Angeles data (Zhu et al., 2002).

A 3-day period of southerly winds provided an opportunity to conduct an upwind-downwind highway transect to assess the impact of roadway emissions. For PM 2.5, the transect yielded a measured near-roadway PM2.5 concentration of 2.0 ± 0.7 μg/m3 versus a value of 1.6 ± 0.5 μg/m3 predicted from recent California vehicle emis- sion rates and ∼3 μg/m3 using current European freeway data. Thus, measured PM 2.5 concentrations at the near-road site showed excel- lent agreement with predicted PM 2.5 concentrations assuming recent published emission rates. By contrast, using 1973 emission rates one would predict roughly 16 μg/m3 of PM 2.5, thus illustrating the dra matic
progress that has been made in reducing roadway emissions in the US during the past four decades.

The NEXUS and European estimates of the contribution of brake drums were also similar, although numerous assumptions are needed to make the comparison. The European value is roughly 3.2 mg/km brake wear, assuming 6% trucks in their mix. The NEXUS iron val- ues were strongly enhanced over soil concentrations, indicating road- way sources including brake wear. The fine and very fine zinc parti- cles were too fine in size to be from tire erosion. With these assump- tions, the expected emission rate from brake wear is roughly 5.2 mg/ km, similar to the European values especially after some iron reduc- tion from vehicle debris and roadbed wear.

For ultra-fine particles, there was a dramatic increase in diesel soot, closely tied to traffic volumes, but only a modest increase in traf- fic-correlated zinc and other transition metals during the 3 day in- tensive. For the rest of the 5 week study, mean elemental concentra- tions were essentially identical at the 10 m and 100 m north sites and highly correlated in time. Further, this correlation existed independent of wind direction, including having both the 10 m north and 100 m north sites upwind of the freeway. From these results, we conclude that most of the time the freeway was a negligible enhancement of a regional ultra-fine background of transition metals.

The regional ultra-fine species closely resembled those seen in a recent study of stop-and-go traffic on a heavily traveled secondary road with heavy braking (Cahill et al., 2014). These particles are sim- ilar in composition to debris from brake pads, drums, and the zinc ad- ditive in motor oil (Cahill et al., 2011). These results reinforce the be- havior seen in the Los Angeles I-710 studies (Zhu et al., 2002) for ultra-fine particle number and black carbon, e.g., these components extended well upwind of the freeway and were soon reestablished at the same concentrations downwind beyond 150 m, thus showing a re- gional distribution of ultra-fines well away from the local freeway.

The NEXUS data also support the conclusions of Zhu et al. (2002) that little coagulation or particle size growth was seen in ultra-fines from freeways, as both Zhu's ultra-fine soot and particle number data were essentially identical versus distance with non-reactive CO.

The conclusion from this study is that for traffic on a freely flow- ing, high speed freeway with little braking and acceleration, almost all ultra-fine particles except for diesel soot, and most of the very fine particles, are generated from diverse vehicular and industrial sources Cahill -- 00000340
AR002347
Ultra fine aerosols < 0.09 mi crome ters
Nickel - Copper -- Zinc - lron/2
12 ,--------------- ------------------------------,
10 ..., 8 E .;
E
e 6
"' 0
C
"' z 4
2 0
23 noon 24 noon 25 noon 26 noon 27 noon 28 noon 29 noon October , 2010

UNCORRECTED PROOF
18 Atmospheric Environment xxx (2016) xxx-xxx
in the region and are not directly associated with local freeway traffic.

Uncited reference
Acknowledgements
The authors gratefully acknowledge the assistance in meteorolog- ical models of Ron Henry, as well as the detailed aethalometer, me- teorological and traffic data of the US EPA/FHWA study (US EPA, 2013). The United States Environmental Protection Agency through its Office of Research and Development funded and collaborated in the research described here under contract EP-12-D-000219 to the University of California, Davis. It has been subjected to Agency re- view and approved for publication.

Appendix A. Supplementary data
Supplementary data related to this article can be found at http://dx. doi.org/10.1016/j.atmosenv.2016.09.023.

References
ARB, 2007. Emfac2007. California Air Resources Board. http://www.arb.ca.gov/msei/ onroad/downloads/docs/user_guide_emfac2007.pdf.

Bukowiecki, N., Lienemann, P., Hill, M., Furger, M., Richard, A., Amato, F., Prévôt, A.S.H., Baltensperger, U., Buchmann, B., Gehrig, R., 2010. PM10 emission fac- tors for non-exhaust particles generated by road traffic in an urban street canyon and along a freeway in Switzerland. Atmos. Environ. 44, 2330–2340.

Cahill, T.A., Feeney, P.J., 1973. Contribution of Freeway Traffic to Airborne Particu- late Matter. Final Report of the California Air Resources Board, Contract ARB-502.

Cahill, T.A., Goodart, C., Nelson, J.W., Eldred, R.A., Nasstrom, J.S., Feeney, P.J., 1985. In: In: Ariman, Teoman, Nejat Veziroglu, T. (Eds.), Design and Evaluation of the Drum Impactor. Proceedings of International Symposium on Particulate and Multi-phase Processes. vol. 2. Hemisphere Publishing Corporation, Washington, D.C, pp. 319–325.

Cahill, T.A., Barnes, D.E., 2009. Comparison of Fine Mass, UC Davis DRUM versus FRM, at the ARB 13th and T Street Site. Final Report to the California ARB (2009).

Cahill, Thomas A., Barnes, David E., Spada, Nicholas J., Lawton, Jonathan A., Cahill, Thomas M., 2011. Very fine and ultra-fine metals and ischemic heart disease in the California Central Valley 1: 2003– 2007. Aerosol Sci. Technol. 45, 1125–1134.

Cahill, T.A., Barnes, D.E., Lawton, J.A., Miller, R., Spada, N.J., Cliff, S.S., Zhou, Y.J., Perry, K.D., Cahill, C.F., 2010. Continuous Aerosol Sampling and Analysis in Support of Health Based Studies. Final Report to W. Wilson, US EPA RFQ-RT-09–00122.

Cahill, T.A., 2010. How does Vegetation affect pollution removal. In: The Workshop on the Role of Vegetation in Mitigation Air Quality Impacts from Traffic Emis- sions EPA RTP North Carolina April 27, 28.

Cahill, Thomas M., Cahill, Thomas A., 2013. Seasonal variability of particle-associ- ated organic compounds near a heavily traveled secondary road. Aerosol Sci.

Technol. http://dx.doi.org/10.1080/02786826.2013.857757.

Cahill, T.A., Barnes, D.E., Spada, N.J., 2014. Seasonal variability of ultra-fine metals downwind of a heavily traveled secondary road. Atmos. Environ. 94, 173–179.

Chen, L.C., Lippmann, M., 2009. Effects of metals within ambient air particulate mat- ter (PM) on human health. Inhal. Toxicol. Int. Forum Resp. Res. 21, 1–31.

Courtney, W.J., Rheingrover, S., Pilotte, J., Kaufmann, H.C., Cahill, T.A., Nelson, J.W., 1978. Continuous observation of particulates during the general motors sul- fate dispersion experiment. J. Air Pollut. Control Assoc. 28, 225–228.

Denier Van der Gon, H., Gerlofs Nijland, M., Gehrig, R., Gustafsson, M., Janssen, N., Harrison, R., Hulskotte, J., Johansson, C., Jozwicka, M., Keuken, M., Krijgsheld, K., Ntziachristos, L., Riediker, M., Cassee, F., 2013. The policy relevance of wear emissions from road transport, now and in the Future– an international workshop report and consensus statement. J. Air Waste Manag. Assoc. 63, 136–149.

Feeney, P.J., Cahill, T.A., Flocchini, R.G., Eldred, R.A., Shadoan, D.J., Dunn, T., 1975. Effect of roadbed configuration on traffic derived aerosols. J. Air Pollut.

Control Assoc. 25, 1145–1147.

Gauderman, W.J., McConnell, R., Gilliland, F., London, S., Thomas, D., Avol, E., Vora, H., Berhane, K., Rappaport, E.B., Lurmann, F., Margolis, H.G., Peters, J., 2000. Association between air pollution and lung function growth in southern Cali- fornia children. Am. J. Respir. Crit. Care Med. 162, 1383–1390.

Gertler, Alan, Abu-Allaban, Mahmoud, Coulombe, William, Gillies, John A., Pierson, William R., Rogers Rogers Fred, C., Sagebiel, John C., Tarnay, Leland, Cahill, Thomas A., 2003. Measurements of mobile source particulate emissions in a high- way tunnel. Int. J. Veh. Des. 27, 86–93.

Habibi, Kamran, 1973. Characterization of particulate matter in vehicle exhaust. Envi- ron. Sci. Technol. 7 (3), 223–234. http://dx.doi.org/10.1021/es60075a001.

Health Effects Institute, 2009. Traffic Related Air Pollution: a Critical Review of the Literature on Emissions, Exposure, and Health Effects. HEI 17, Boston, MA.

Henry, Ronald C., Vette, Alan, Norris, Gary, Vedantham, Ram, Kimbrough, Sue, Shores, Richard C., 2011. Separating the air quality impact of a major highway and nearby sources by nonparametric trajectory analysis. Environ. Sci. Tech- nol. 45, 10471–10476. http://dx.doi.org/10.1021/es202070k.

Karner, A.A., Eisinger, D.S., Neimeier, D.A., 2010. Near roadway air quality: synthe- sizing the findings from real world data. Environ. Sci. Technol. 44, 5334–5344.

Lewis, T.C., Robins, T.G., Dvonch, J.T., Keeler, G.J., Yip, F.Y., Mentz, G.B., Lin, X.H., Parker, E.A., Israel, B.A., Gonzalez, L., Hill, Y., 2005. Air pollution-associ- ated changes in lung function among asthmatic children in Detroit. Environ.

Health Perspect. 113, 1068–1075.

Lin, C.C., Peng, C.K., 2010. Characterization of indoor PM10, PM2.5, and ultra-fine particles in elementary school classrooms: a review. Environ. Eng.

Sci. 27, 915–922.

Lippmann, Morton, 2009. Semi-continuous speciation analyses for ambient air particu- late matter: an urgent need for health effects studies. J. Expo. Sci. Epidemiol.

Stud. 19, 235–247.

Malm, W.C., Sisler, J.F., Huffman, D., Eldred, R.A., Cahill, T.A., 1994. Spatial and seasonal trends in particle concentration and optical extinction in the United States.

J. Geophys. Res. 99 (1), 1347–1370.

McGee, M.A., Kamal, A.S., McGee, J.K., Wood, C.E., Dye, J.A., Krantz, Q.T., Lan- dis, M.S., Gilmour, M.I., Gavett, S.H., 2015. Differential effects of particulate matter upwind and downwind of an urban freeway in an allergic mouse model. En- viron. Sci. Technol. 49 (6), 3930–3939.

Nichols, M., 2009. Letter and Appended Report from the Staff of the California Air Resources Board.

Oberdorster, Gunter, Stone, Vicki, Donaldson, Ken, 2007. Toxicology of nanoparti- cles: a historical perspective. Nanotoxicology 1 (1), 2_25.

Peters, John M., Avol, Edward, Navidi, William, London, Stephanie J., James Gauder- man, W., Lurman, Fred, Linn, William S., Margolis, Helene, Rappaport, Edward, Gong Jr., Henry, Thomas, Duncan C., 1999a. A study of Twelve southern Califor- nia communities with differing levels and types of air pollution I. Prevalence of respiratory morbidity. Am. J. Respir. Crit. Care Med. 159 (3), 760–767.

Peters, John M., Avol, Edward, Navidi, William, London, Stephanie J., James Gauder- man, W., Lurman, Fred, Linn, William S., Margolis, Helene, Rappaport, Edward, Gong Jr., Henry, Thomas, Duncan C., 1999b. A study of Twelve southern Califor- nia communities with differing levels and types of air pollution; II. Effects on pul- monary function. Am. J. Respir. Crit. Care Med. 159 (3), 768–775.

Raabe, Otto G., Braaten, David A., Axelbaum, Richard L., Teague, Stephen V., Cahill, Thomas A., 1988. Calibration studies of the DRUM impactor. J. Aerosol Sci. 19 (2), 183–195. Personal communication 1995.

US EPA, 2013. FHWA and EPA National Near-road Study. Detroit, MI. Report No.

DTFH61-07-X-30015 http://www.fhwa.dot.gov/environment/air_quality/air_ toxics/research_and_analysis/near_road_study/detroit_michigan/index.cfm.

Vette, A., Burke, J., Norris, G., Landis, M., Batterman, S., Breen, M., Isakov, V., Lewis, T., Gilmour, M.I., Kamal, A., Hammond, D., Vedantham, R., Bereznicki, S., Tian, N., Croghan, C., 2013. The near-road exposures and effects of urban air pollutants study (NEXUS): study design and methods. Sci. Total Envi- ron. 448, 38–47.

Zielinska, B., Goliff, W., McDaniel, M., Cahill, T., Kittelson, D., Watts, W., 2003.

Chemical Analyses of Collected Diesel Particulate Matter Samples in the E-43 Project. National Renewable Energy Laboratory.

Zielinska, B., Sagebiel, J., McDonald, J.D., Whitney, K., Lawson, D.R., 2004. Emis- sion rates and comparative Chemical composition from selected in-use diesel and gasoline fueled vehicles. J. Air Waste Manag. Assoc. 54, 1138–1150.

Zhu, Y., Hinds, W.C., Kim, S., Shen, S., Sioutas, C., 2002. Study of ultra-fine particles near a major highway with heavy duty diesel traffic. Atmos. Envi- ron. 36, 4323–4335.

Cahill -- 00000341AR002348