Environmental Research Letters

LETTER • OPEN ACCESS

Air pollution accountability of energy transitions: the relative importanceof point source emissions and wind fields in exposure changesTo cite this article: Lucas RF Henneman et al 2019 Environ. Res. Lett. 14 115003

 

View the article online for updates and enhancements.

This content was downloaded from IP address 140.247.231.25 on 11/02/2020 at 16:20

Environ. Res. Lett. 14 (2019) 115003 https://doi.org/10.1088/1748-9326/ab4861

LETTER

Air pollution accountability of energy transitions: the relativeimportance of point source emissions and wind fields in exposurechanges

Lucas RFHenneman1 , Loretta JMickley2 andCorwinMZigler3

1 Department of Environmental Health,Harvard T.H.Chan School of PublicHealth, Boston,Massachusetts, United States of America2 JohnA. Paulson School of Engineering andApplied Sciences, HarvardUniversity, Cambridge,Massachusetts, United States of America3 Department of Statistics andData Sciences andDepartment ofWomen’sHealth, University of Texas andDellMedical School, Austin,

United States of America

E-mail: lhenneman@gmail.com

Keywords: energy transitions, coal, air pollution accountability, HYSPLIT, air pollution exposure

Supplementarymaterial for this article is available online

AbstractRecent studies have sought epidemiological evidence of the effectiveness of energy transitions. Suchevidence often relies on so-called ‘natural experiments’, wherein environmental and/or healthoutcomes are assessed before, during, and after the transition of interest. Often, these studies attributeair pollution exposure changes—eithermodeled ormeasured—directly to the transition.Weformalize a framework for separating the fractions of a given exposure change attributable tometeorological variability and emissions changes. Using this framework, we quantify relative impactsof wind variability and emissions changes from coal-fired power plants on exposure to SO2 emissionsacross theUnited States under three unique combinations of spatial-temporal and source scales.Wefind that the large emissions reductions achieved byUnited States coal-fired power plants after 2005dominated population exposure changes. In each of the three case studies, however, we identifiedperiods and regions inwhichmeteorology dampened or accentuated differences in total exposurerelative to exposure change expected from emissions reductions alone. The results evidence a need forseparatingmeteorology-induced variability in exposure when attributing health impacts to specificenergy transitions.

1. Introduction

Energy transitions in the United States utility sectorare the result of multiple influences, including envir-onmental regulations, variable fuel prices, and chan-ging demand [1]. Such transitions in regionalelectricity generationmanifest themselves as actions—e.g. shuttering, scaling output, and installing emissionscontrols—taken at individual generating units [1].Utility sector transitions influence emissions, airquality, and health at temporal scales from immediate(in the case of shutterings or emissions controlinstallations) to decadal and at spatial scales from localto global.

A growing body of research—air pollutionaccountability—has sought to quantify the

downstream air quality and health impacts of reg-ulatory actions in particular [2], and many of themethods developed for air pollution accountabilityhave the potential to extend to air quality and healthbenefits of energy transitions in general.Whether con-sidering energy transitions explicitly or other abruptchanges in pollution derived from opportunistic‘quasi’ or ‘natural’ experiments, accountability assess-ment is made difficult by co-varying factors in timeand space [2, 3].

One such co-varying factor is the propensity ofemitted pollutants to transport through and react inthe atmosphere; downstream air quality relationshipswith emissions changes are highly nonlinear [4]. Pre-vious air pollution accountability studies have addres-sed the potential for atmospheric conditions and

OPEN ACCESS

RECEIVED

1August 2019

REVISED

5 September 2019

ACCEPTED FOR PUBLICATION

26 September 2019

PUBLISHED

29October 2019

Original content from thisworkmay be used underthe terms of the CreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2019TheAuthor(s). Published by IOPPublishing Ltd

transport patterns to confound the results in healthanalyses of emissions-reducing events. In one of mul-tiple studies linking health and social outcomes to airquality changes concurrent with a 1980s strike at asteel mill near Salt Lake City, Utah, Pope (1989) notedthat ‘One concern aboutmaking observations pertain-ing to these time periods is that the winter when theGeneva steel mill was closed may have had relativelygood weather conditions and limited conditions ofstagnant air [5].’

In fact, accounting (or not) for concurrentmeteorology in accountability studies has repeatedlybeen documented as an explanation for confusing ormisinterpreted results [6]. Initial studies of the airquality changes spanning the 1996 Summer Olympicsin Atlanta (and attendant interventions to reduce traf-fic during the games) suggested subsequent declines inambient ozone and pediatric asthma acute care events[7], but follow-up analyses that explicitly identifiedconcurrent regional patterns inmeteorology indicatedthat the Olympics-related interventions were likelynot the dominant cause of the air quality improve-ments [8]. Similarly, multiple studies identifiedimproved air quality [9–13] and health [14–17] fol-lowing pollution reduction efforts during the BeijingOlympics; however, Wang et al (2009) used back tra-jectories during the period and identified that theapparent improvements were not due solely to theOlympics-related interventions, but cleaner air thatwas transported into the Beijing area during the periodof the Games [18]. They concluded that 40% of thevariability in ambient particulate matter concentra-tions was due to meteorology, while only 16% wasattributable to emissions reductions.

While not necessarily studies of energy transitionsper se, the above studies entail relevant examples inwhich researchers sought epidemiological evidence ofthe benefits to human health from emissions reduc-tions, in much the same way one might investigate apast intervention amounting to an explicit energytransition. The common feature is a clear action (oractions) taken to impact pollution emissions andinvestigation of changes in air quality and health out-comes spanning the intervention that must considerconcurrent changes in meteorology and pollutiontransport in order to accurately reflect consequencesof the transition. In addition to their opportunisticnature to generate quasi-experimental changes in pol-lution exposure, such studies have potential to informfuture policies under the implicit assumption that theobserved results would generalize amid future deploy-ment of a similar intervention in a new setting [19, 20].Misattribution of pollution (and health) impacts fol-lowing a transition to the transition itself, instead of toconcurrent meteorological or transport patterns, lim-its the generalizability to future similar actions.

Previous studies have employed dynamical airquality models to quantify meteorological and emis-sions influences on air pollution concentrations

[21, 22] and source category sensitivities [23]. Othershave employed empirical statistical models (e.g.meteorological detrending techniques) to quantifymeteorology-induced variability in observed ambientpollutant concentrations [24–28]. Towards the goal offormalizing these issues in the specific context ofenergy transitions, we seek to extend the generalizedframeworks implied in these previous works for quan-tifying emissions and meteorological influences onchanging exposures resulting from transitions atgroups of individual sources.

Specifically, we describe a framework under whichmeteorological variability (specifically, wind fieldvariability) impacts accountability studies of energytransitions. Using a recently developed exposuremodel and a national database of coal power plantemissions, we investigate the implications of arbitrarybefore/after periods on three different spatial-tem-poral scales. We conclude by addressing the implica-tions of the results on the interpretation of past studiesand the potential for improved design of futurestudies.

2. Theory

Population exposure (Exp) to emissions from sourcesJ in period t at location i can be formulated as afunction f of the meteorological conditions (includingatmospheric initial and boundary conditions) Mt andpollutant emissions EJ t, , i.e. f M EExp ,tot i J t t J t, , ,= ( )∣ .Similarly, changes in exposure Exptot J t, , betweenperiods t− and t+ depend on changing meteorologyand emissions across the same period (we drop the J[source] notation because it is well-defined for eachscenario below, and we drop i [exposure location]because we calculate exposure changes for each loca-tion):

f M E f M EExp , , . 1tot t t t tD = -+ + - -( ) ( ) ( )

Exposure (Exp) in this case is defined broadly as thesource contribution of well-defined emissions sourceson air quality in a given area. Sulfate concentrationsattributable exclusively to SO2 emissions from coalpower plants provide an appropriate example of thistype of exposure. Importantly, variability in suchexposures are nonlinear with emissions. For example,sulfate concentrations depend on many atmosphericconditions, including concentrations of existing con-stituents such as water vapor andOH [29, 30]. In addi-tion, exposures to emissions of individual pollutantsare not independent of emissions of other pollutants.These nonlinearities and interactions may be more orless important in relating exposure changes to emis-sions changes depending on the precise emissionssources and exposure of interest. Wemaintain a broaddefinition of exposure so as not to restrict the theor-etical methods to the specific examples presentedbelow.

2

Environ. Res. Lett. 14 (2019) 115003

For illustration, consider a single source thatinstalls an emissions control device in period t0

between periods t− and t+, an action that might pro-vide an opportunity for a ‘quasi-experimental’accountability study. Figure 1 presents a schematic ofchanging exposure attributable tometeorological con-ditions Mt- and Mt+ over an arbitrary region near asource. In an exposure model wheref m e m e, = *( ) , M 1t =+ , and M 1t =- under theirrespective plumes and 0 outside, the change in expo-sure ExptotD takes three discrete values for regions A,B, and C. Population living in region A, which experi-ences a decrease in exposure between periods t− andt+, benefits only frommeteorological conditions (con-ceptualized here as a change in wind direction). Popu-lation living in region C, conversely, experiences anincreased exposure between the two periods due tometeorology. Only the population living in region B,which is covered by both Mt- and Mt+, experiences anexposure reduction attributable to the emissionscontrol.

To extend this example to a more realistic casewith potentially many sources, we define two terms:the change in exposure attributable to meteorologicalvariability ( ExpmetD ) and the change in exposure attri-butable to emissions changes ( ExpemissD ) across peri-ods t− and t+. ExpmetD is calculated as the differencebetween exposure in t− and exposure in a hypotheticalscenario with meteorology in t+ and emissions in t−.Conversely, ExpemissD is calculated as the difference inexposure between two scenarios with meteorologyfixed at Mt+ and emissions varied between Et- andEt+.

f M E f M EExp , , , 2met t t t tD = -- - + -( ) ( ) ( )

f M E f M EExp , , . 3emiss t t t tD = -+ - + +( ) ( ) ( )

The convention basing ExpmetD on the emissionsin t− and ExpemissD on the meteorology in t+ opposedto vice versa was chosen for two reasons. First, varyingperspectives between t+ and t− ensures that ExpmetD

and ExpemissD sum to ExptotD for linear functions f.Second, we base ExpmetD on emissions in t− to ensurethe magnitude of ExpmetD remains interpretable rela-tive to Exptot in t

−. Three previous studies used varyingcombinations of base and future years in chemicaltransport models for meteorological and emissionschanges, and do not set a consistent precedent forwhich combination ismost appropriate [21–23].

Knowledge of ExptotD for a given energy trans-ition does not in and of itself reveal information aboutthe relative contributions of ExpmetD or ExpemissD .Regimes separated according to relativemagnitudes of

ExpmetD and ExpemissD (figure 2) provide a convenientinterpretive framework of the relative influences of

ExpmetD and ExpemissD on ExptotD . For a givenobserved positive ExptotD (i.e. increase in total expo-sure), the only requirement is that the combinedimpact ofmeteorological and emissions changes led toan increase, as denoted in regimes a M E { },b M E { }, c M E { }, or d M E { } (up anddown arrows denote positive and negative influence,respectively; double arrows denote greater magnitudeinfluence of one factor relative to the other). Similarly,for a negative ExptotD , it is possible to exist in any ofregimes e M E { }, f M E { }, g M E { },or h M E { }.

3.Methods

In this section, we describe the exposure modelemployed to simulate pollutant transport emissionsfrom coal-fired power plants in the United States.While the framework above could be applied withmodels of varying complexity, we describe anapproach for calculating ExptotD , ExpmetD , and

ExptotD using the HYSPLIT average dispersionmodel,HyADS, a recently developed model for estimatingpopulation exposure to point source emissions [31].Finally, we introduce three source, time, and distancescale combinations for analysis.

3.1.Measuring exposure to coal power plantemissions usingHyADSHyADS employs HYSPLIT, an air parcel transport anddispersion model [32, 33], to model 100 parceltrajectories originating from each source every sixhours (i.e. beginning at 12:00 a.m., 6:00 a.m., 12:00 p.m., and 6:00 p.m.)—100 was chosen to allow reason-able plume dispersion while enabling computationalefficiency. The trajectories are tracked for 10 days.Three-dimensional hourly parcel locations beginningat hour 2 are trimmed if they are below 0 elevation orabove the planetary boundary layer and are thenassigned to a fine grid and aggregated by month.Gridded monthly aggregated parcel totals are thenscaled by the source’s monthly SO2 emissions. HyADSwas described and evaluated in detail at annual [31]and monthly [34] time scales and has been applied inhealth analyses [35, 36].

Figure 1. Schematic showing changing exposure coverage intime periods t− and t+ under atmospheric conditions Mt-

and Mt+. The orange circle at left represents a single pointemission source.

3

Environ. Res. Lett. 14 (2019) 115003

Weusemonthly SO2 emissions, location data, andSO2 emissions control type and installation dates fromover 1000 coal-fired power plants in theU.S.Environmental Protection Agency’s Air MarketsProgram Database [37]. Unit stack heights wereretrieved from the National Emissions Inventory [38];the databases were merged on unit ID as previouslydescribed byHenneman et al (2018) [31].

HyADS, a reduced complexity model, does notcapture all atmospheric processes important in deter-mining emissions exposure. There are, however, twoimportant benefits that motivate its use here. First, itsreduced complexity nature allows for the multipleactual and counterfactual runs needed for this analy-sis; these would require orders of magnitude morecomputational time in a traditional chemical trans-port model. Second, HyADS includes two primaryinputs—wind fields and emissions—thereby allowingus to isolate the influence of these inputs on exposure.

3.2. Emissions andmeteorology exposure variabilitySince HyADS uses linear combinations of transportedair pollution coverage (i.e. unit-less HyADS parceldensity in period t; Mt) and emissions (Et [tons]), wetake f m e m e, = *( ) and simplify equations (1)–(3):

M E M EExp , 4tot t t t tD = -+ + - - ( )

M E M EExp , 5met t t t tD = -- - + - ( )

M E M EExp . 6emiss t t t tD = -+ - + + ( )

For each period, we calculate ExptotD , ExpmetD ,and ExpemissD at each ZIP code for each coal unit andsum across units to yield single values for each of thethree terms at each ZIP code. To improve interpret-ability, we then calculate Expp totD , Expp metD , and

Expp emissD as the percent change from the base period,i.e.

M EExp

Exp100%, 7p tot

tot

t t

D =D

´- -

( )

M EExp

Exp100%, 8p met

met

t t

D =D

´- -

( )

M EExp

Exp100%. 9p emiss

emiss

t t

D =D

´- -

( )

We quantify these terms for energy transitions atcoal power plants on three combinations of sourcegroups and spatial-temporal scales. Two of the threescale combinations were selected to mirror recent epi-demiological accountability studies:

1.Annual, national exposure changes from allunits (figure SI-1 is available online at stacks.iop.org/ERL/14/115003/mmedia) in 2005, 2006,2011, and 2012; the scales mirror a recent studyrelating coal emissions exposure changes tochanges in various health outcome rates in theMedicare population [36].

2.Monthly, national exposure changes from 108units at 49 facilities that installed SO2 emissionscontrols in 2008–2009 (figure SI-1).

3.Quarterly, city-level exposure changes from 15units at 4 facilities that contributed large frac-tions of Lousiville, KY total exposure in 2012 andinstalled SO2 emissions controls or shutteredbetween 2012 and 2017; the scales mirror a recentstudy investigating these transitions’ impacts onasthma outcomes [35].

At each combination of source/receptor scales, wecalculate the percent change in total exposure andexposure attributable to meteorology and emissionschanges ( Expp metD and Expp emissD to Expp totD ) from

Figure 2.Relative combinations of ExpmetD and ExpemissD to ExptotD , which is negative at lower left (cool colors) and positive atupper right (warmcolors). Letters and arrow combinations denote the relative positive () and negative () contributions of ExpmetD(M) and ExpemissD (E), with double arrows denoting larger relative contribution than single arrows. Lighter colors signifymorepositive ExpmetD than ExpemissD .

4

Environ. Res. Lett. 14 (2019) 115003

a base period defined as the first period in eachanalysis.

4. Results

4.1. Annual exposure changes fromall emissionsreductionsRelative to 2005, nationwide coal power plant SO2

emissions decreased by 18.4%, 51.9%, and 65.4%across 2006, 2011, and 2012, respectively. Mostgenerating units decreased emissions, but not all—in2012, for example, 201 out of 1152 operating unitsincreased their emissions relative to 2005 (figure SI-2).Annual nationwide exposure to coal power plant SO2

emissions relative to 2005 ( Expp totD ) decreased22.2%±19% (standard deviation), 65.0%±8%,and 53.3%±17% in 2006, 2011, and 2012 (figure 3and table SI-2). Spatially, the largest changes occurredin the eastern-most third of the country (figure SI-3),consistent with the location of the highest densities ofcoal power plants (figure SI-1).

Emissions reductions from coal power plantsdrove annual average nationwide ZIP code exposuresreductions ( Expp emissD ) of 8.6%±11% in 2006,32.1%±16% in 2011, and 67.5%±26% in 2012(figure SI-3 and table SI-2). In 2006 and 2011, meteor-ological differences from 2005 led to average

Expp metD of−13.6%±16% and−32.9%±18%. In2012, however, meteorological changes from 2005 ledto positive Expp metD of 14.2%±24%.

These results are further reflected in the differ-ences in dominating change regimes between 2011and 2012. In 2011, most of the country fell intoregimes f M E { } (63.8%) and g M E { }(33.4%, primarily in the eastern third of the country),which are characterized by emissions and meteorol-ogy-attributable decreases in exposure (table SI-1). In2012, most (72.9%) of the country was in regimeh M E { }, characterized by Expp metD and

Expp emissD of opposite signs, and 74.0% of the countryexperienced positive influence of meteorology differ-ences on their exposure. Therefore, wind variabilityled to an annual average Expp totD in 2011 that wasmore negative than Expp totD in 2012, even as emis-sions reductions in 2012 were greater than those in2011 (we investigate reasons for these differences inthe discussion section). Without accounting formeteorological differences, total exposure changesbetween 2005 and 2011 appear more effective thanthey actually were, whereas exposure changes in 2012appear less effective.

4.2.Monthly exposure changes from scrubberinstallationsThe 108 units that installed SO2 emissions controls in2008 and 2009 represented many of the largest in thecountry, and they were among the highest emittingbefore installing controls. In 2005–06, these unitsemitted 29% of all SO2 from coal power plants; in2011–12, they emitted only 4%. Total SO2 emissionsfrom these units decreased 93% between 2005–6 and2011–12.

Emissions reductions at the 108 units contributedto average national Expp totD reductions relative toJanuary 2005 of between 87% and 95% in all monthsin 2012 (figure 4). Expp emissD decreased substantiallybetween 2005–06 and 2011–12, averaging near 0% inthe earlier time period and near −100% in the latter.Recall that Expp emissD can be below−100% because itdepends on the current month’s meteorology. Varia-bility in Expp totD in 2005 and 2006 is primarily drivenby meteorological changes; Expp metD is more positivein summer months and more positive overall in 2005than 2006.

Monthly exposure variability is further reflected inexposure change regimes (figure 4). In 2005 and 2006,area throughout the country in most months is domi-nated by regimes a M E { }, b M E { },e M E { }, and f M E { }, which are all

Figure 3.Top: annual evolution ofmeteorology and emissions exposure changes in theU.S. relative toHyADS exposure in 2005.Bottom: ZIP codes covered by each exposure regime.

5

Environ. Res. Lett. 14 (2019) 115003

characterized by smaller magnitude Expp emissD thanExpp metD . Months in 2005 are slightly more domi-

nated by regimes a M E { } and b M E { } than2006 (which is more represented by e M E { } andf M E { }), suggesting meteorology contributed tomore exposure increases in 2005 and decreases in 2006relative to January 2005.

In all months in 2011–12, nearly 100% of thecountry is covered by regimes f M E { },g M E { }, and h M E { }, showing the dominantimpact of emissions reductions on Expp totD (figures 4and SI-4). In both years, the country was covered simi-larly (between 16% and 40%) by regime g M E { },which is characterized by more negative Expp emissDthan Expp metD . In large portions of the country,therefore, total exposure decreased more than wouldbe expected based on emissions changes alone.

The greater coverage of regime f M E { } in2011 and h M E { } in 2012 corroborates results ofthe annual evaluation above (figure 4). Each month in2012 saw more coverage of regime h M E { } com-pared to the same month in 2011, suggesting lowerobserved benefits than would be expected based onemissions changes alone.

4.3.Quarterly exposure changes in Louisville, KY,2011–2017Through unit retirements and emission controlsinstallations, the 15 units at four facilities near Louis-ville, KY reduced their annual SO2 emissions by 80%between 2012 and 2017. Annual Expp totD from theseunits in the 39 Louisville ZIP codes decreased by76.2%±12%across the same period (figure 5).

In 2012, 2013, and 2014, quarterly variability inExpp totD across Louisville was dominated byExpp metD , as shown by the relatively small variability

in Expp emissD (figure 5). Exposure change regimesa M E { }, b M E { }, e M E { }, andf M E { }—each of which is characterized by a

larger meteorology-related influence than emissions-related one—dominated the area in these years. Start-ing in the fourth quarter of 2015, emissions reductionsovercame meteorological variability to reduce overallexposure relative to quarter 1, 2012. Regimeh M E { } dominated the Louisville area beginningin 2014, fourth quarter. Only in the first quarters of2016 and 2017 did a marked difference occur—inthese quarters, meteorological differences from thefirst quarter of 2012 increased the impact of emissionsreductions on Exptot .

The results show the potential for wind field varia-bility on a local scale to affect the relationshipsbetween emissions reductions and exposure changes.Even while facilities nearby Louisville installed emis-sions control devices and reduced their emissionsfrom 2012–2014, the reductions had little impact onvariability in total exposure because of meteorologicalvariability. Only with a large enough decrease in emis-sions did Expp totD remain below zero consistently.Even so,meteorological differences betweenmost per-iods from 2015 to 2017 yielded smaller (closer to zero)

Expp totD ʼs than would be expected based on emis-sions changes alone.

These results have implications on results of epide-miological studies of such interventions, such as theone undertaken by Casey et al (in review). It would beunreasonable to assume, for instance, that emissionsreductions of similar magnitudes across periods withvastly differing meteorology lead to similar exposurereductions (and subsequent health improvements).All interventions on the four facilities investigated byCasey et al (in review), for example, occurred beforethe end of 2016. Comparisons of quarterly Expp totDafter 2016, however, would lead to overestimates (firstquarter, 2017) or underestimates (second-fourthquarters, 2017) of the change in exposure due exclu-sively to emissions changes (and, therefore, impactevaluations of the effectiveness of the controls).

Figure 4.Top:monthly evolution of emissions,meteorology, and total exposure changes in national ZIP codes relative toHyADSexposure in January 2005.Outliers have been removed in the plot. Bottom: percent of total area in the lower 48 states covered by eachregime.

6

Environ. Res. Lett. 14 (2019) 115003

One potential approach to address large intra-annual meteorological variability would be to com-pare the same quarters across years instead of employ-ing the first quarter of 2012 as the base period. Thissolution is imperfect, however, because of the pre-sence of inter-annual variability in Expp metD(figure 5). The variability is greatest in the summer-time—3rd-quarter Expp metD ranges from 115% in2014% to 215% in 2016.

5.Discussion

5.1. LimitationsHyADSmeasures exposure to emissions using a linearcombination of spatial impacts and SO2 emissions,not exposure to atmospheric pollutants such as PM2.5

and its precursors, SO2, NOx, or other atmosphericconstituents previously linked to adverse healtheffects. Previously, we have shown that national andindividual unit exposures measured by HyADS corre-late well with chemical transport model simulatedPM2.5 in both direct sensitivity [31] and adjointsensitivity [39] frameworks. In addition, Foley et al(2015) found that atmospheric nonlinearities led tosmaller, more homogeneous changes in annual ozoneconcentrations across the country between 2002 and2005 relative to changes attributable to meteorologyand emissions changes. Based on these previousresults, the linearity assumption employed by HyADS(i.e. f m e m e, = *( ) ) is sufficient for quantifyingexposure variability on the spatial and temporal scalesinvestigated here, but future studies may employchemical transport models within the proposed fra-mework to investigate the importance of nonlinea-rities in energy transitions impacts on exposure.

A wide body of research has sought to quantifyeffects of meteorological variability on ambient pri-mary and secondary pollutant concentrations usingmodel [40, 41] and observation-based approaches

[24, 30]. These studies have highlighted the impor-tance of a range of atmospheric conditions, such astemperature, relative humidity, precipitation, andconcentrations of other pollutants in determiningobserved ambient concentrations (which are related,but distinct from the exposures we seek to measurehere). While surface winds can both transport pollu-tion away from polluted areas and transport emissionstoward otherwise cleaner areas [24, 30, 40], variabilityin wind speed and direction does not capture all of theprocesses important in determining exposure varia-bility. This point is developed further in the sub-sequent section.

As discussed in the theory section above, the rela-tive contributions of ExpmetD and ExpemissD to

ExptotD (and congruently Expp metD , Expp emissD , andExpp totD ) are dependent on the base and future years

selected for calculation of each. We calculatedExpmetD based on the base year’s emissions (Et-) to

maintain a consistentmagnitude. Therefore, to ensureExpmetD + ExpemissD = ExptotD , ExpemissD is required

to be based on future year meteorology (Mt+). Thisconvention results in meteorological and emissionsimpacts that are somewhat more difficult to interpretindividually (e.g. ExpemissD represents reductions ofmore than 100%); taken together, however, ExpmetDand ExpemissD gain interpretability with the introduc-tion of concentrations regimes (figure 2).

5.2.Meteorology and exposure variabilityIn this section, we contextualize the annual, nationalresults with published literature addressing the pro-pensity of meteorology to impact sulfate concentra-tions (SO2 emissions readily convert to sulfate in theatmosphere). We discuss the results above in thecontext of annual NCEP/NCARReanalysis 10 mwindfields across theUnited States [42, 43].

Tai et al (2010) showed that nationwide sulfateconcentrations were enhanced in stagnant conditionsand with increased temperature, relative humidity,

Figure 5.Top: quarterly evolution of emissions,meteorology, and total exposure changes in Louisville, KY relative toHyADS exposurein 2012, quarter 1.Outliers have been removed. Bottom: percent of total area in the Louisville, KY covered by each regime.

7

Environ. Res. Lett. 14 (2019) 115003

and—importantly for the present study—wind ema-nating from the Ohio River Valley area, which con-tains the greatest concentration of coal emissions inthe country (figure SI-1). Similarly, others have foundthat sulfate (and other PM2.5 constituents) are nega-tively correlated with wind speed [44–46]. In relationto the present study, these previous results suggestthat, in periodswith relatively greater wind speed ema-nating from areas other than theOhio River Valley, wewould expect a negative ExpmetD , and vice versa.

Nationwide, average annual 10 m wind speedsincreased slightly from 0.94 m s 1- in 2005 to 1.02,1.19, and 1.25 m s 1- in 2006, 2011, and 2012, respec-tively (figure SI-6). Wind field anomalies in both 2011and 2012 show increased flux of marine air from theGulf ofMexico, implying greater ventilation, at least intheGulf States. However, 2012 saw greater increases insoutherly winds in the Midwest and smaller increasesin southerly winds along the East coast compared to2011 (figure SI-7). The wind changes in 2012 implygreater recirculation of air from the East toward theUS interior, less ventilation of continental air, and thusincreased exposure to power plant emissions. Theseanomalies in the 2012wind pattern are consistent withthe anticyclonic conditions and severe heat observedover the central and eastern U.S. that spring and sum-mer [44].

While this evaluation implies multiple simplifica-tions—e.g. including only annual 10 mwind fields—itdoes confirm differences in meteorologies of the yearsunder investigation, particularly in 2011 and 2012.Our results show the importance of only wind indetermining exposure, without accounting for othermeteorological differences that have been shown to beimportant to determining ambient air pollution.These factors warrant further investigation under thepresent framework requiring more complex atmo-spheric models; however, such models should beinterpreted in acknowledgement of their limited capa-cities to fully capture observed sensitivities of air pollu-tion tometeorological variability [47–49].

5.3. RecommendationsIn three scenarios (two that mirrored epidemiologicalstudies), we investigated how energy transitions’impacts on exposure to coal emissions were impactedby wind field and emissions variability. Results showthat, with large enough emissions reductions,

Expp emissD decreased in all spatial and temporal scalesto a magnitude greater than any observed positive

Expp metD .Meteorology, however, still plays a substan-tial role in determining the effectiveness of emissionsreductions on reducing exposure, with increasingimportance at finer temporal and spatial scales.

Despite relying on HyADS’s linear exposuremodel ( M EExp= * ), the results we present agree inprinciple with a wide body of work describing meteor-ological influences on ambient air quality. Complexair quality models (e.g. chemical transport models) are

able to more fully resolve relationships betweenmeteorology, emissions, and exposure, but are limitedby their increased computational cost. Quantifyingthese relationships remains an active area of research,and future studies of energy transitions should weighthe application of a variety of models to their exposureand time and distance scales of interest.

Variability in ExpmetD is greater with increasingspatial and temporal resolution, suggesting that anapproach for mitigating the influence of meteorologyin a future study is to use annually aggregated expo-sures or compare the same periods across years. Win-ter wind fields tend to decrease exposure, whereassummer fields tend to increase exposure. However,differences between years remain (notably, between2011 and 2012). Exposure change regime coveragebecomes more homogeneous with large emissionsdecreases and atfiner spatial scales (e.g. in Louisville).

Air pollution accountability studies often employ apotential outcomes framework, wherein implicationsof interventions on exposure/health outcomes areinferred by estimating a counterfactual—i.e. the out-come assuming no intervention. In such a framework,the outcome assumes that everything else except theintervention (e.g. meteorology) is constant betweenthe actual and counterfactual. While this approachwould seem to preclude the relevance of varyingmeteorological influence, it remains important toconsider because meteorology-independent estimatesof intervention impacts are more relevant for inform-ing future policies, particularly as a changing climateincreases uncertainty surrounding future meteor-ological conditions and their impacts on ambient airquality [47–49].

Our approach has utility for other studies focusedon exposure. For example, our use of percent changeand our definition of the change regimes a-h allow foreasy comparison with results from other studies thatmay use different techniques to characterize exposure.Our description of exposure in terms of area coveragepermits the extension of our results to other spatialscales.

Importantly, our results do not imply that pre-viously calculated associations between observed ormodeled ExptotD and health outcomes are invalid (forinstance, two that motivated the present manuscript[35, 36]). Evaluating impacts of changes in total expo-sure following an energy transition is a valid approachwhen interpreted cautiously. Most work in thisdomain, however, implicitly assumes that

ExptotD == ExpemissD , i.e. the observed exposurechange is caused entirely by a change in emissions.Analyses that do seek to attribute exposure changes toemissions (versus total exposure) must confront thepotential entanglement withmeteorology.

8

Environ. Res. Lett. 14 (2019) 115003

Acknowledgments

This work was supported by research funding fromNIHR01ES026217, EPA 835872, and HEI 4953. Itscontents are solely the responsibility of the grantee anddo not necessarily represent the official views of theUSEPA. Further, USEPA does not endorse the pur-chase of any commercial products or services men-tioned in the publication.

Data availability

The data and code that support the findings of thisstudy are openly available at DOI 10.176 05/OSF.IO/B8PA6.

ORCID iDs

Lucas RFHenneman https://orcid.org/0000-0003-1957-8761

References

[1] National ResearchCouncil 2004AirQualityManagement intheUnited States (Washington, DC:National Academies Press)(https://doi.org/10.17226/10728)

[2] HEIAccountabilityWorkingGroup 2003AssessingHealthImpact of AirQuality Regulations: concepts andMethods forAccountability ResearchCommunication 11 Health EffectsInstitute, BostonMA

[3] Henneman LRF, LiuC,Mulholland J A andRussell AG2016Evaluating the effectiveness of air quality regulations: a reviewof accountability studies and frameworks J. AirWasteManage.Assoc. 67 144–72

[4] Seinfeld JH and Pandis SN2006Atmospheric Chemistry andPhysics: fromAir Pollution to Climate Change (Hoboken,NJ:Wiley)

[5] PopeCA1989Respiratory disease associatedwith communityair pollution and a steelmill, UtahValleyAm. J. Public Health79 623–8

[6] van ErpAM,Kelly F J, Demerjian K L, PopeCA andCohenA J2011 Progress in research to assess the effectiveness of airquality interventions towards improving public healthAirQual. Atmos. Health 5 217–30

[7] FriedmanMS, Powell K E,Hutwagner L, GrahamLMandTeagueWG2001 Impact of changes in transportation andcommuting behaviors during the 1996 summer olympic gamesinAtlanta on air quality and childhood asthma Jama 285897–905

[8] Peel J L, KleinM, FlandersWD,Mulholland J A andTolbert P E 2010 Impact of improved air quality during the1996 summer olympic games in atlanta onmultiplecardiovascular and respiratory outcomesTechnical Report 148Health Effects Institute

[9] WangB, ShaoM, Lu SH, YuanB, ZhaoY,WangM,Zhang SQ andWuD2010Variation of ambient non-methanehydrocarbons in Beijing city in summer 2008Atmos. Chem.Phys. 10 5911–23

[10] WangM, ZhuT, Zheng J, Zhang RY, Zhang SQ, Xie XX,HanYQ and Li Y 2009Use of amobile laboratory to evaluatechanges in on-road air pollutants during the Beijing 2008SummerOlympicsAtmos. Chem. Phys. 9 12857–98

[11] Wang S, ZhaoM,Xing J,WuY, ZhouY, Lei Y,HeK, Fu L andHao J 2010Quantifying the air pollutants emission reductionduring the 2008 olympic games in Beijing Environ. Sci. Technol.44 2490–6

[12] WangT et al 2010Air quality during the 2008 BeijingOlympics: secondary pollutants and regional impactAtmos.Chem. Phys. 10 7603–15

[13] WangT andXie S 2009Assessment of traffic-related airpollution in the urban streets before and during the 2008BeijingOlympic Games traffic control periodAtmos. Environ.43 5682–90

[14] HouQ,AnXQ,Wang Y andGuo J P 2010An evaluation ofresident exposure to respirable particulatematter and healtheconomic loss in Beijing during Beijing 2008OlympicGamesSci. Total Environ. 408 4026–32

[15] RichDQ et al 2012Association between changes in airpollution levels during the BeijingOlympics and biomarkers ofinflammation and thrombosis in healthy young adult J. Am.Med. Assoc. 307 2068–78

[16] RichDQ et al 2015Differences in birthweight associatedwiththe 2008 beijing olympic air pollution reduction: results from anatural experiment Environ.Health Perspect. 117 1713–7

[17] SuC,Hampel R, FranckU,Wiedensohler A, Cyrys J, PanX,WichmannHE, Peters A, Schneider A andBreitner S 2015Assessing responses of cardiovascularmortality to particulatematter air pollution for pre-, during- and post-2008Olympicsperiods Environ. Res. 142 112–22

[18] WangX,Westerdahl D, Chen LC,WuY,Hao J, PanX,GuoX andZhangKM2009Evaluating the air quality impactsof the 2008 BeijingOlympicGames: on-road emission factorsand black carbon profilesAtmos. Environ. 43 4535–43

[19] BellML,Morgenstern RD andHarringtonW2011Quantifying the humanhealth benefits of air pollutionpolicies: review of recent studies and newdirections inaccountability researchEnviron. Sci. Policy 14 357–68

[20] CropperM, Fraas A andMorgensternR 2017 Lookingbackward tomove forward Science 355 1375–6

[21] FoleyKM,Hogrefe C, Pouliot G, Possiel N, Roselle S J,SimonH andTiminB 2015Dynamic evaluation of CMAQ:I.Separating the effects of changing emissions and changingmeteorology on ozone levels between 2002 and 2005 in theeasternUSAtmos. Environ. 103 247–55

[22] Godowitch JM, Pouliot GA andTrivikramaRao S 2010Assessingmulti-year changes inmodeled and observed urbanNOXconcentrations from a dynamicmodel evaluationperspectiveAtmos. Environ. 44 2894–901

[23] Henneman LRF, LiuC,HuY,Mulholland J A andRussell AG2017Air qualitymodeling for accountability research:operational, dynamic, and diagnostic evaluationAtmos.Environ. 166 551–65

[24] Camalier L, CoxWandDolwick P 2007The effects ofmeteorology on ozone in urban areas and their use in assessingozone trendsAtmos. Environ. 41 7127–37

[25] CoxWMandChu S-H 1993Meteorologically adjusted ozonetrends in urban areas: a probabilistic approachAtmos. Environ.27 425–34

[26] Henneman LRF,HolmesHA,Mulholland J A andRussell AG2015Meteorological detrending of primary and secondarypollutant concentrations:method application and evaluationusing long-term (2000–2012) data in atlantaAtmos. Environ.119 201–10

[27] Kuebler J, van denBerghH andRussell AG 2001 Long-termtrends of primary and secondary pollutant concentrations inSwitzerland and their response to emission controls andeconomic changesAtmos. Environ. 35 1351–63

[28] Rao S T andZurbenko IG 1994Detecting and trackingchanges in ozone air qualityAirWaste 44 1089–92

[29] BrockCA et al 2002Particle growth in the plumes of coal-firedpower plants J. Geophys. Res. 107 9–1

[30] Tai A PK,Mickley L J and JacobD J 2010Correlations betweenfine particulatematter (PM2.5) andmeteorological variablesin theUnited States: implications for the sensitivity of PM2.5to climate changeAtmos. Environ. 44 3976–84

[31] Henneman LRF, Choirat C, IveyCE, CummiskeyK andZigler CM2019Characterizing population exposure to coalemissions sources in theUnited States using theHyADSmodelAtmos. Environ. 203 271–80

9

Environ. Res. Lett. 14 (2019) 115003

[32] Draxler RR andHess GD1998An overview of the hysplit_4modelling system for trajectories, dispersion, and depositionAust.Meteorol.Mag. 47 295–308

[33] Stein A F,Draxler RR, RolphGD, Stunder B J B,CohenMDandNgan F 2015Noaaʼs hysplit atmospherictransport and dispersionmodeling systemBull. Am.Meteorol.Soc. 96 2059–77

[34] Henneman LRF,Dedoussi I C, Casey J, Choirat C andZigler CComparisons of simple and complexmethods for quantifyingexposure to point source air pollution emissions In review

[35] Casey J A et alReduced coal-fired power plant emissionsrelated to improved community asthma-related healthcareutilization and individual symptoms in Louisville, Kentucky,USA In review

[36] Henneman LRF, Choirat C andZigler CM2019Accountability Assessment ofHealth Improvements in theUnited States AssociatedwithReducedCoal EmissionsBetween 2005 and 2012Epidemiology accepted 30 477–85

[37] U.S. EPA 2016AirMarkets ProgramDataMart [internetdatabase] available at: (http://www.epa.gov/ttn/airs/aqsdatamart)

[38] U.S. EPA 2016b 2014National Emissions Inventory Plan.Technical report. (https://www.epa.gov/air-emissions-inventories/national-emissions-inventory-nei)

[39] Henneman LRF, LiuC, ChangH,Mulholland J, Tolbert P andRussell A 2019Air quality accountability: developing long-termdaily time series of pollutant changes and uncertainties inatlanta, georgia resulting from the 1990 clean air actamendments Environ. Int. 123 522–34

[40] Shen L,Mickley L J andTai A PK2015 Influence of synopticpatterns on surface ozone variability over the easternUnitedStates from 1980 to 2012Atmos. Chem. Phys. 15 10925–38

[41] ZhongQ et al 2018Distinguishing emission-associatedambient air PM2.5 concentrations andmeteorological factor-inducedfluctuations Environ. Sci. Technol. 52 10416–25

[42] CompoGP et al 2011The twentieth century reanalysis projectQ. J. R.Meteorolog. Soc. 137 1–28

[43] CompoGP,Whitaker J S and Sardeshmukh PD2006Feasibility of a 100-year reanalysis using only surface pressuredataBull. Am.Meteorol. Soc. 87 175–90

[44] PetersonTC,HoerlingMP, Stott S C andHerring PA 2014Explaining extreme events of 2013 from a climate perspectiveBull. Am.Meteorol. Soc. 95 S1–96

[45] Shen L,Mickley L J andMurray L T 2017 Influence of2000–2050 climate change on particulatematter in theUnitedStates: results from anew statisticalmodelAtmos. Chem. Phys.17 4355–67

[46] Westervelt DM,Horowitz LW,NaikV, Tai A PK,Fiore AMandMauzerall D L 2016Quantifying PM2.5-meteorology sensitivities in a global climatemodelAtmos.Environ. 142 43–56

[47] Shen L andMickley L J 2017 Seasonal prediction ofUSsummertime ozone using statistical analysis of large scaleclimate patterns Proc. Natl Acad. Sci. 114 2491–6

[48] Tai A PK,Mickley L J and JacobD J 2012Atmosphericchemistry and physics Impact of 2000–2050 climate change onfine particulatematter (PM2.5) air quality inferred from amulti-model analysis ofmeteorologicalmodesAtmos. Chem.Phys. 12 11329–37

[49] Tai A PK,Mickley L J, JacobD J, Leibensperger EM, Zhang L,Fisher J A and PyeHOT2012Meteorologicalmodes ofvariability for fine particulatematter (PM2.5) air quality in theUnited States: implications for PM2.5 sensitivity to climatechangeAtmos. Chem. Phys. 12 3131–45

10

Environ. Res. Lett. 14 (2019) 115003