This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 140.247.231.25

This content was downloaded on 28/03/2017 at 22:44

Please note that terms and conditions apply.

Public health impacts of the severe haze in Equatorial Asia in September–October 2015:

demonstration of a new framework for informing fire management strategies to reduce

downwind smoke exposure

View the table of contents for this issue, or go to the journal homepage for more

2016 Environ. Res. Lett. 11 094023

(http://iopscience.iop.org/1748-9326/11/9/094023)

Home Search Collections Journals About Contact us My IOPscience

You may also be interested in:

Fire emissions and regional air quality impacts from fires in oil palm, timber, and logging

concessions in Indonesia

Miriam E Marlier, Ruth S DeFries, Patrick S Kim et al.

Regional air quality impacts of future fire emissions in Sumatra and Kalimantan

Miriam E Marlier, Ruth S DeFries, Patrick S Kim et al.

Contribution of vegetation and peat fires to particulate air pollution in Southeast Asia

C L Reddington, M Yoshioka, R Balasubramanian et al.

Impact of the June 2013 Riau province Sumatera smoke haze event on regional air pollution

Sheila Dewi Ayu Kusumaningtyas and Edvin Aldrian

Sensitivity of mesoscale modeling of smoke direct radiative effect to the emission inventory: a

case study in northern sub-Saharan African region

Feng Zhang, Jun Wang, Charles Ichoku et al.

Biomass burning drives atmospheric nutrient redistribution within forested peatlands in Borneo

Alexandra G Ponette-González, Lisa M Curran, Alice M Pittman et al.

Vegetation fires, absorbing aerosols and smoke plume characteristics in diverse biomass burning

regions of Asia

Krishna Prasad Vadrevu, Kristofer Lasko, Louis Giglio et al.

Industrial concessions, fires and air pollution in Equatorial Asia

D V Spracklen, C L Reddington and D L A Gaveau

Environ. Res. Lett. 11 (2016) 094023 doi:10.1088/1748-9326/11/9/094023

LETTER

Public health impacts of the severe haze in Equatorial Asia inSeptember–October 2015: demonstration of a new framework forinforming fire management strategies to reduce downwind smokeexposure

ShannonNKoplitz1,9, Loretta JMickley2,MiriamEMarlier3, Jonathan J Buonocore4, Patrick SKim1,Tianjia Liu5,Melissa P Sulprizio2, Ruth SDeFries3, Daniel J Jacob1,2, Joel Schwartz6,Montira Pongsiri7 andSamuel SMyers6,8

1 Department of Earth and Planetary Sciences, HarvardUniversity, Cambridge,MA,USA2 School of Engineering andApplied Sciences,HarvardUniversity, Cambridge,MA,USA3 Department of Ecology, Evolution, and Environmental Biology, ColumbiaUniversity, NewYork,NY,USA4 Center forHealth and theGlobal Environment, Harvard T.H. Chan School of PublicHealth,HarvardUniversity, Boston,MA,USA5 Department of Earth and Environmental Sciences, ColumbiaUniversity, NewYork,NY,USA6 Harvard T.H. Chan School of PublicHealth,HarvardUniversity, Boston,MA,USA7 Wildlife Conservation SocietyHEALProgram,Washington, DC,USA8 HarvardUniversity Center for the Environment, HarvardUniversity, Cambridge,MA,USA9 Author towhomany correspondence should be addressed.

E-mail: skoplitz@fas.harvard.edu

Keywords: land use changefires, smoke exposure, GEOS-chem adjoint

Supplementarymaterial for this article is available online

AbstractIn September–October2015,ElNiñoandpositive IndianOceanDipole conditions set the stage formassivefires in Sumatra andKalimantan (IndonesianBorneo), leading topersistently hazardous levels of smokepollution acrossmuchofEquatorialAsia.Herewequantify the emission sources andhealth impacts of thishaze episode and compare the sources and impacts to an event of similarmagnitudeoccurringundersimilarmeteorological conditions in September–October 2006.Using the adjoint of theGEOS-Chemchemical transportmodel,wefirst calculate the influenceof potentialfire emissions across thedomainonsmoke concentrations in three receptor areas downwind—Indonesia,Malaysia, andSingapore—duringthe 2006 event. This stepmaps the sensitivity of each receptor tofire emissions in eachgrid cell upwind.Wethen combine these sensitivitieswith 2006 and2015fire emission inventories from theGlobal FireAssimilation System (GFAS) to estimate the resultingpopulation-weighted smoke exposure.Thismethod,which assumes similar smoke transport pathways in 2006 and2015, allowsnear real-time assessmentofsmokepollution exposure, and therefore the consequentmorbidity andprematuremortality, due to severehaze.Our approachalsoprovides rapid assessmentof the relative contributionoffire emissions generatedin a specificprovince to smoke-relatedhealth impacts in the receptor areas.Weestimate that haze in 2015resulted in 100 300 excess deaths across Indonesia,Malaysia andSingapore,more thandouble those of the2006 event,withmuchof the increasedue tofires in Indonesia’s SouthSumatraProvince. Themodelframeworkwe introduce in this study can rapidly identify those areaswhere landusemanagement toreduce and/or avoidfireswould yield the greatest benefit to humanhealth, bothnationally and regionally.

1. Introduction

The thick smoke that blanketed Equatorial Asia duringSeptember–October of 2015 was likely the worst haze

episode since 1997, when land use fires caused billionsof dollars in damage and thousands of prematuredeaths (Johnston et al 2012, Marlier et al 2013). Thedegraded peatlands that typically burn during such

OPEN ACCESS

RECEIVED

29 January 2016

REVISED

20 July 2016

ACCEPTED FOR PUBLICATION

29 July 2016

PUBLISHED

19 September 2016

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.

© 2016 IOPPublishing Ltd

episodes contain significant combustible organic mat-erial and so release large amounts of fine particulatematter (PM2.5), the leading cause of global pollution-related mortality (World Health Organization 2009,Lelieveld et al 2015). As in previous episodes, theprevailing winds in 2015 transported the smoke todensely populated areas across Indonesia and theMalay Peninsula, including Singapore and KualaLumpur. In this work, we identify provinces whereland use policies and management strategies couldmost effectively mitigate the downwind smoke expo-sure and consequent costs to human health duringsuch haze events. We also compare the 2015 event toanother large smoke episode in 2006 in order todetermine how fire emission patterns driven by landuse and land cover changemay be evolving.

Across Indonesia, fires are frequently used to burnagricultural residue, clear forest, or prepare land forplantations and smallholder farms. Fires also occurfrom vandalism and accidental ignitions (Denniset al 2005, Gaveau et al 2014a). Fire emission levels aregreatest from degraded peatlands, especially in dryyears (Marlier et al 2015a, 2015b). In 2006, burning inindustrial concessions to clear land for oil palm andtimber plantations accounted for ∼40% of total fireemissions in Sumatra and∼25% inKalimantan (Indo-nesian Borneo) (Marlier et al 2015c). As on oil palmplantations, fire on timber plantations is used to clearnative vegetation quickly and cheaply in order toestablish commercial wood pulp species. Such speciesinclude fast growing trees such as Acacia, whose turn-over rate is about three times faster than oil palm (∼7years compared to ∼25 years), potentially resulting inmore frequent post-harvest burning (Effendy andHardono 2001, Feintrenie et al 2010).

As in 1997 and again in 2006, the severe haze inSeptember–October 2015 was enabled by a combina-tion of El Niño and positive Indian Ocean Dipole(pIOD) conditions, both of which promote droughtand greatly enhance fire activity in the region, eitherbecause fuel loads are drier allowing fires to escape andburn out of control or because farmers take advantageof the dry weather and clear more land than usual(Field and Shen 2008, van der Werf et al 2008, van derWerf et al 2010, Reid et al 2012). Though the under-lying meteorological triggers are similar across theseextreme haze events, the spatial distributions of landcover and fire emissions are evolving rapidly, drivenlargely by expanding globalmarkets for oil palm, pulp-wood and timber, and by increases in small-scale agri-culture (Field et al 2009, Miettinen et al 2011,Margono et al 2014, Gaveau et al 2014b, Marlieret al 2015a). A comparison between the 2006 and 2015events is critical for (1) quantifying the contributionsof specific fire source locations to smoke exposure indownwind population centers during extreme hazeevents, and (2) identifying possible trends in the mag-nitudes of these differentiated contributions over thelast decade. We chose 2006 and not 1997 for this

comparison since modern satellite technology (e.g.,themulti-wavelength instruments on board Terra andAqua that can capture burned area and AOD) was notavailable until 1999 (King et al 2003). Diagnosis of thefire emission locations that result in the greatest smokeexposures downwind can guide the design of moreinformed strategies to prevent or minimize recurrenthaze events.

In this study, we demonstrate the potential of a newanalytical framework to rapidly assess in near real-timethe emission sources and health impacts of an ongoingsmoke episode in Equatorial Asia. The framework, pre-sented here all together for thefirst time, integrates infor-mation on (1) fire emissions related to land cover andland use (Marlier et al 2015a, 2015b, 2015c), (2)meteor-ological drivers of smoke transport, (3) domestic andtransboundary source–receptor relationships, whichquantify the sensitivity of PM2.5 concentrations in recep-tor areas to the specific locations of fire emissions (Kimet al 2015, Marlier et al 2015a, 2015c), and (4) healthimpact functions incorporating regionally specific dataon mortality rates, age structure, and population. Pre-vious efforts to quantify health impacts from biomassburning in Equatorial Asia have proven computationallyexpensive (e.g., Johnston et al2012,Marlier et al2013). Incontrast, once the source–receptor relationships havebeenmapped, our novel framework can readily quantifysmoke exposure fromany given distribution offire emis-sions without the need for additional computationallyexpensive model simulations. In this way, stakeholderscan quickly identify the key emission areas contributingto that exposure and estimate the resulting morbidityand premature mortality in downwind populations,even as an extreme smoke event unfolds.

2.Methods

Below we describe the fire emissions estimates and thesurface and satellite observations used in this analysis.In the supplement, we describe the GEOS-Chemadjoint used to calculate smoke exposures and thehealth impact calculations.

2.1. Fire emissionsWe use near-real time fire emission estimates for the2015 haze event from the GFAS, a product that wasreadily available for next-day processing at the time ofour analysis (GFAS, http://join.iek.fz-juelich.de/macc/; Kaiser et al 2012). GFAS emissions are derivedat 0.5°×0.5° horizontal resolution from observa-tions of fire radiative power (FRP) from the ModerateResolution Imaging Spectroradiometer (MODIS)instruments on board the Terra and Aqua satellites.The emissions are based on observed relationshipsbetween FRP and combusted dry matter for eight landcover types based on those used in the Global FireEmissions Database (GFED; van der Werf et al 2010),supplementedwithmaps of organic soil andpeatlands.

2

Environ. Res. Lett. 11 (2016) 094023

Here we define smoke as the sum of organic carbon(OC) and black carbon (BC) aerosol, the primarycomponents of smoke-related PM2.5. Based on com-parisons with satellite AOD, Kaiser et al (2012)recommended scaling GFAS OC and BC emissions bya global factor of 3.4. However, the Kaiser et al (2012)underestimate in modeled AOD may be partly due tomodel treatment of carbon aerosol in the atmosphererather than emissions (Andela et al 2013). Somewhatarbitrarily, we instead scale the GFAS emissions up by50%, which produces model results that better matchsurface observations during 2015 (section 2.2). Scalingup by 50% also yields PM2.5 emission totals forIndonesia in 2006 that match those obtained from adifferent emission inventory (Marlier et al 2015c). Toidentify sources of fire emissions, we rely on estimatesof the spatial distributions of peatlands and industrialconcessions in 2010, the only year for which such dataare available (figure S1). We overlay these estimatesonto 1×1 km2 MODIS FRP detections to attributefire activity to these sources (table S1). This finerspatial scale information is necessary given the multi-ple types of land use and land cover within each0.5°×0.5°GFAS grid cell.

2.2. Surface PMobservationsWe compare the GEOS-Chem adjoint estimates forsmoke exposure in Singapore during 2015 againstobserved PM2.5 concentrations for five stations inSingapore operated by the National EnvironmentAgency (NEA; http://nea.gov.sg/). In order to isolateenhancements in surface PM2.5 due only to smokeaerosol, we first calculate the average observed PM2.5

during June (13.6 μg m−3), the earliest available mea-surements before the onset of the haze event. We then

subtract the June average non-smoke concentrationfrom the full time series.

2.3. Satellite observationsWe also compare the adjoint smoke exposures fromthe 2006 and 2015 haze events to two satelliteproducts: ultraviolet aerosol index (AI) from theOzone Monitoring Instrument (OMI; Torreset al 2002, Torres et al 2007) and aerosol optical depth(AOD) at 550 nm from MODIS aboard the Terrasatellite (Levy et al 2010). We use the Level 3 quality-assured products from both instruments, processed at1° horizontal resolution. Both products are frequentlyemployed to characterize smoke aerosol in SoutheastAsia (Reid et al 2013, Chang et al 2015). By using twodifferent aerosol products, we overcome some of theuncertainty associated with representing smoke opti-cal properties in the retrievals (Zender et al 2012). Wealso maximize the amount of usable satellite dataretrieved during the time periods of interest in anenvironment that is challenging to observe from spacedue to frequent cloudiness, lengthy coastlines, andmountainous terrain (Reid et al 2013).

3. Results

3.1. Fire emissionsFigure 1 shows the 1995–2015 time series of theNINO3.4 index and Dipole Mode Index (DMI), thestandard indices used to represent the phases of the ElNiño Southern Oscillation (ENSO) and the IndianOcean Dipole (IOD; www.stateoftheocean.osmc.noaa.gov). Both El Niño and positive phases of theIOD result in suppressed convection over Indonesia,leading to drought and increased fire activity. Septem-ber 2015 was the strongest El Niño on record since

Figure 1. Indices of large scale circulation patterns in Equatorial Asia. The blue line shows theNINO3.4 index for the ElNiño-SouthernOscillation for January 1995–December 2015. The red line shows theDipoleMode Index (DMI) for the IndianOceanDipole (IOD). Dotted lines indicate the thresholds for El Niño (+1) or LaNina (−1) in theNINO3.4 series, and for strongly positive(+1) or strongly negative (−1) IOD conditions in theDMI series. Data are from theNational Oceanic andAtmosphericAdministration (stateoftheocean.osmc.noaa.gov).

3

Environ. Res. Lett. 11 (2016) 094023

1997 and much stronger than September 2006.September 2006, however, saw a slightly strongerpositive IOD. Over the entire 20 year record, there areonly three instances when both theNINO3.4 andDMIindices simultaneously exceed a value of +1: 1997,2006, and 2015. Prior to 2015, the 1997 regional hazeevent was the most severe on record. However, due toa lack of available satellite data for 1997, we focus ourcomparison on 2006, the second most extreme hazeevent prior to 2015.

South Sumatra and Central Kalimantan bothburned strongly during the 2006 event, contributing30% and 31% respectively of total IndonesianOC+BC emissions released (figures 2 and 3). TheJuly–October 2015 Indonesian fire emissions were 2.1Tg higher compared to the 2006 emissions over thesame months, an increase of 110% (figure 3). SouthSumatra contributed 62% (1.3 Tg) of the emission dif-ference between 2015 and 2006, while Central Kali-mantan contributed only 18% (0.4 Tg). JambiProvince, which was responsible for less than 5% ofthe haze during 2006, contributed the third highestemissions of any individual province in 2015 and 12%to the emission difference between 2015 and 2006.Contributions fromWest Kalimantan decreased from16% in 2006 to 6% in 2015.

3.2. Smoke exposure during haze eventsTo estimate smoke exposure at each receptor areaduring the two haze events, we multiply the GFASemissions for July–October 2006 and July–October2015 by the adjoint sensitivities simulated for July–October 2006 (figure S3). Although the most intensehaze occurred during September–October in bothyears (65% of total annual fire emissions in 2006, 80%in 2015), we extend our time horizon to include theentire burning season from July–October (83%of totalannual fire emissions in 2006, 93% in 2015). Asdescribed above, we assume that the 2006 smoketransport patterns are similar to those in 2015. Smoketransport for all three receptors shows a strongsensitivity to the prevailing southeasterly winds gov-erned by the location of the Intertropical ConvergenceZone (ITCZ) during September–October (Changet al 2005). To test our results using meteorology fromother years we also calculate smoke exposures inSingapore due to the 2015 fire emissions with adjointsensitivities for 2005 and 2007–2009 (figure S4).Seasonal mean exposures are similar across the2005–2008 sensitivities, while 2009 is ∼25% greaterthan the 2006 average.

Figure 4 shows a combined time series of dailyNEA PM2.5 concentrations at Singapore for1 July–31October 2015. Smoke concentrations returned to nor-mal levels in late October with the onset of monsoonalrains (Cochrane 2015). The NEA data reveal an aver-age observed smoke exposure of 30 μg m−3 in Singa-pore for July–October 2015. Our approach yields an

average July–October 2015 smoke exposure at Singa-pore of 27 μg m−3, consistent with the NEA surfaceobservations. Estimated population-weighted smokeexposures in Indonesia and Malaysia during 2015 are19 μg m−3 and 14 μg m−3, respectively. Mean July–October non-smoke PM2.5 concentrations in theseareas are ∼10–15 μg m−3 (section 2.2; GEOS-Chemforward model simulation, not shown), yielding totalannual average PM2.5 exposures below 50 μg m−3 andwithin the range of linearity (supplement section 2).

During 2006, we find that July–October exposuresare lower by at least a factor of 2 at all three receptorscompared to 2015, with values of 10 μg m−3 in Singa-pore, 8 μg m−3 in Indonesia, and 6 μg m−3 in Malay-sia. Figure 5 compares OMI AI and MODIS AOD forSeptember–October 2006 to those quantities in Sep-tember–October 2015. Also shown for 2015 is a com-parison of MODIS AOD to AOD measurements atseveral sites in the Aerosol Robotic Network (AERO-NET; Holben et al 1998); no data are available at thesesites for September–October 2006. Both satelliteinstruments reveal an approximate doubling in aero-sol levels in 2015 compared to 2006 acrossmuch of thedomain, consistent with the adjoint exposure resultsand confirming the utility of our approach.

3.3. Emission sourcesDuring 2006, South Sumatra and Central Kalimantancontributed roughly equally to regional emissions(30%–31%), but across all three receptor areas, SouthSumatra accounted formore than three times asmuchsmoke exposure than did Central Kalimantan(figure 3). Also in 2006, West Kalimantan contributedonly 16% to regional emissions, but was responsiblefor the second highest exposure levels in Malaysia,after South Sumatra. In 2015, the percent contribu-tions to smoke exposure from South Sumatraincreased by 10%–15% in the absolute sense across allreceptors, compared to South Sumatra’s contributionto smoke exposures across the same areas in 2006.Exposure contributions from Jambi doubled in 2015compared to 2006, while contributions from the otherregions decreased. The trends in smoke exposure fromSouth Sumatra and Jambi in 2015 suggest significantchanges in land use occurred in these regions duringthe intervening years as discussed below.

Emission source regions for both the 2006 and2015 haze events differ from those contributing to asevere haze event in June 2013 that severely affectedSingapore and the Malay Peninsula (Gaveauet al 2014a). The 2013 haze has been traced to agri-cultural burning in Riau, a province in northernSumatra covered extensively by peatlands and oil palmconcessions (figure S1; Gaveau et al 2014a). In con-trast, we find that Riau did not contribute significantlyto smoke exposure at any receptor during either the2006 or 2015 haze events. This difference is partly dueto the longer burning season in Riau than further

4

Environ. Res. Lett. 11 (2016) 094023

south in Indonesia, with increased burning earlier inthe year (Reid et al 2013). Also, the prevailing windsover Riau shift from westerly in June to southeasterlyduring September–October as the ITCZmoves south-ward across Equatorial Asia, transporting what smokethere is later in the season toward the northwest ratherthan towards the populatedMalay Peninsula.

To estimate the contributions of different types ofland use and land cover to fire activity, we used satelliteFRP observations at high-resolution (1 km×1 km)for July–October 2006 and 2015 (table S1). While FRPtotals are not directly comparable to emissions, theyare available in near real-time and can reveal the spa-tial distribution of fire activity and land use types affec-ted by fires, as demonstrated by Marlier et al (2015c).For example, although oil palm concessions have pre-viously been implicated as amajor driver of peat burn-ing in Indonesia (Koh et al 2011), we find that burningin oil palm concessions in 2006 accounted for only11% of total FRP in Sumatra and 32% in Kalimantan.In 2015, these contributions declined to just 5% and

20%, respectively. In Sumatra, FRP in timber conces-sions increased from 27% in 2006 to 55% in 2015,while in Kalimantan the timber contribution was<10% in both years. In contrast, the percentage oftotal FRP occurring in peatlands increased in bothSumatra (44%–72%) and Kalimantan (32%–43%)from 2006 to 2015. Reasons for the apparent increasein fire activity in peatlands are not clear. Draining ofpeatlands to prepare for agricultural use in the inter-vening years may have made the peat more vulnerableto fires (Carlson et al 2012, Turetsky et al 2015). Fur-ther analysis of the 2015 event with updated land usemaps is needed to fully understand these patterns at aspatial scale that is useful for stakeholders.

3.4.Health impact estimatesThe US Environmental Protection Agency primarystandard for unhealthy levels of annual average PM2.5 is12 μgm−3. According to the World Bank, much ofEquatorial Asia is close to this standard in non-haze years—e.g., annual mean values reported for 2011 are

Figure 2. September–October total emissions of organic and black carbon (OC+BC) fromGFAS during 2006 and 2015. The top panelshows 2006, themiddle panel shows 2015, and the bottompanel shows the difference (2015–2006). Province boundaries are shown inthe bottompanel by colored lines, corresponding to Jambi in coral, South Sumatra andBangka–Belitung in green,West Kalimantanin blue, andCentral Kalimantan in purple.

5

Environ. Res. Lett. 11 (2016) 094023

13–14 μgm−3 for Indonesia, Malaysia, and Singapore(http://data.worldbank.org/indicator/EN.ATM.PM25.MC.M3). Based on smoke exposures for 2006(section 3.2), we estimate the following excess deaths forthat year, with 95% confidence intervals calculated as inDriscoll et al (2015): 34 600 (9000–60 100) in Indonesia,2300 (600–4000) in Malaysia, and 700 (200–1200) inSingapore. For 2015, we estimate these excess deaths tobe: 91 600 (24 000–159 200) in Indonesia, 6500(1700–11 300) in Malaysia, and 2200 (600–3800) inSingapore. (See supplement section 2 for description of

the health impact calculations and comparison toprevious estimates for the 1997 event.) Our resultssuggest that regional smoke-related mortality was 2.7times higher in 2015 than in 2006, an increase whosecauseswe summarize in thediscussion and conclusions.

4.Discussion and conclusions

A combination of El Niño and pIOD conditions inJuly–October 2015 led to dry conditions that exacer-bated agricultural and land clearing fires in southern

Figure 3.Contributions by province to average regional population-weighted smoke exposures (left) and total Indonesianfireemissions ofOC+BC (right) during July–October 2006 and 2015. The island province of Bangka–Belitung is includedwith SouthSumatra. Province boundaries are shown in figure 2 and supplement figure S2.

Figure 4.Time series of observed daily smoke PM2.5 concentrations in Singapore from theNational Environment Agency (NEA;www.nea.org) for 1 July–31October 2015.Gray dots representmean 24 h concentrations averaged acrossfive stations: Central, North,South,West, and East Singapore. Unfilled black circles atmid-month representmonthlymean observed smoke PM2.5 fromNEAaveraged across allfive stations. Non-smoke PM2.5 has been removed from the time series by subtracting the average observed PM2.5

concentrations during June, before the onset of the haze event. Triangles showmonthlymean smoke exposure at Singapore estimatedby this work for July–October 2015 using unscaledGFAS emissions (blue), GFAS emissions scaled by the recommended factor of 3.4(red) andGFAS emissions scaled by 50% (green).

6

Environ. Res. Lett. 11 (2016) 094023

Sumatra and Kalimantan. The resulting dense hazepersisted across much of Equatorial Asia for weeks,imposing adverse public health impacts on popula-tions in Indonesia, Singapore, andMalaysia. Using theadjoint of the GEOS-Chem global chemistry modeltogether with health response functions, we estimate∼60 μg m−3 of population-weighted smoke PM2.5

exposure and 100 300 premature deaths across Indo-nesia, Malaysia, and Singapore due to extreme haze in2015. These values are more than double the25 μg m−3 of smoke PM2.5 and 37 600 prematuredeaths that we estimate for a similar haze event in theregion in 2006. The approximate doubling of regionalsmoke exposure in 2015 compared to 2006 is con-sistent with observations of haze from both OMI AIandMODISAODduring the two events.

Smoke exposures in downwind population centersin both 2006 and 2015 stemmed mainly from burningin South Sumatra, an area that contributed more than30% of the regional emissions and more than 50% ofthe regional smoke exposure during both events. Wealso find that Jambi Province, which did not burn sig-nificantly during 2006, contributed about∼20%of theincreased smoke exposure between 2006 and 2015.High resolution FRP data suggests that fire activity onSumatra in 2015 was dominated by burning on timberconcessions (55%) and peatlands (72%). An updatedanalysis of the sectors and land types contributing tosmoke exposure would build on Marlier et al (2015c)and test our hypothesis.

There are several limitations to the model frame-work we present. First, peatland emissions are difficult

Figure 5.Average satellite aerosol observations for September–October during the 2006 and 2015 haze events. Observations duringthe 2006 event are shown in the top two panels; the 2015 event is shown in themiddle panels. Values of Aerosol Index (AI) from theOzoneMonitoring Instrument (OMI) are shown at left; aerosol optical depths (AOD) at 550 nm from theModerate ResolutionImaging Spectroradiometer (MODIS) on the Terra satellite are on the right. Gray pixels indicatemissing data. Colored circles in theMODISAOD2015 panel indicatemean September–October AERONETobservations. These data are level 1.5, which include cloudfiltering but not thefinal calibrations of the quality-assured level 2.0 product. For consistencywith the TerraMODISmorningoverpass time, the AERONETdata has been averaged during 9am–12pm local time. The bottom two panels compare the 2015 valuesto those in 2006 for the two satellite products, calculated as the ratio of 2015/2006. For clarity, ratios in those gridboxes with lowaerosol content in both products (AOD<0.5 andAI<1.0) are not shown.

7

Environ. Res. Lett. 11 (2016) 094023

to quantify and thus lead to a wide range of emissionestimates across fire inventories. Peat fires can smolderin the subsurface for weeks to months after ignition,often at temperatures too low to be detected accuratelyfrom space (Tansey et al 2008). The contribution ofthis low temperature burning to total smoke emissionsin peatlands is uncertain. The amount of peat fuel con-sumed during each fire event is also uncertain and cur-rently not explicitly represented in fire emissioninventories (Konecny et al 2016). Second, the conces-sion maps we use in our attributions correspond to2010 rather than 2015, and a subset of the concessionsrecorded in 2010 may have been converted to otherland use types prior to 2015. Also, since our analysiswas conducted on a 0.5°×0.67° grid, there is someuncertainty in our attributions of emissions and expo-sure contributions to specific provinces. Third, ourmethod produces population-weighted averagesmoke exposures, which neglect the spatial variabilityin exposures within a receptor region. Populations liv-ing close to fires may experience annual average PM2.5

exposures above the 50 μg m−3 linearity threshold inthe concentration response function we used. We alsofocus only on adult mortality due to lack of knowledgeon the effects of air pollution on child mortality, eventhough impacts on children are likely significant.Additionally, the concentration response functionused here is based on studies done in high-incomecountries with different baseline health characteristicsand air pollution sources. To date, there is little evi-dence quantifying the relationship between PM2.5

composition and toxicity, and so we do not considerthis factor (Levy et al 2012). Our approach also doesnot explicitly eliminate the potential influence of con-founding health effects from conditions related to ElNiño-pIOD meteorology that are independent fromfire pollution, such as heat waves or excess airbornedust from the drier conditions. Ideally wewould valid-ate our results with a multi-year epidemiologicalstudy, but such an evaluation is beyond the scope ofour analysis. Finally, our approach relies on theassumption that the prevailing transport patternsbetween 2015 and 2006 are likely similar given thebackground meteorological conditions (sections 3.1–3.2). Predictive real-time tools to forecast haze inSoutheast Asia exist and have recently been improvedupon (Hertwig et al 2015), but these methods requiresubstantial computational investment and do notreadily provide the geographical source attributioninformation inherent in our adjoint approach.

These limitations in our approach are largely offsetby the benefits of (1) identifying in near real-time thekey fire locations contributing significantly to down-wind smoke exposure during haze events in EquatorialAsia, and (2) rapidly producing estimates of the asso-ciated health impacts when the policymakers and civilsociety groups are seeking ways to effectively addressseasonal burning in Equatorial Asia. In particular, ourapproach provides an early-response diagnosis of

those provinces where effective fire and land use man-agement would yield the greatest benefits to humanhealth, even as the haze event is still unfolding. (Seesupplement section 6 for discussion of challengesfacing existing fire management strategies in Indone-sia.) Applying the framework presented in this workrequires only spatially explicit estimates of carbonaerosol emissions from fires, the archived GEOS-Chem adjoint sensitivities (available upon request),and a health impact model. The exposure calculationsare effectively a one-stepmultiplication combining thesensitivities with the fire emissions, taking only sec-onds to perform and requiring minimal computa-tional resources. In the immediate aftermath of largehaze events, when the incentive for constructive deci-sion-making among stakeholders is greatest, our fra-mework supplies detailed information on the airquality consequences of agricultural fires, a commonland use practice in Equatorial Asia.

Our modeling approach quantifies the public healthimpacts of smoke pollution, including haze crossinginternational borders. The capacity of our framework toquickly identify the provinces where fire emissions aremost severely affecting downwind populations and toquantify the resulting PM2.5-related health impacts canhelp government agencies prioritize forested and peat-land areas for protection and restoration. Furthermore,our integrated modeling approach can help policy-makers reduce health impacts from haze and strengthenlong-term efforts like Reducing Emissions from Defor-estation and Forest Degradation (REDD+) to decreasegreenhouse gas emissions, maximizing the climate co-benefits of preventing the release of carbon from trees,peat, and other soil. Finally, the framework’s ability tocausally link specific fire events to public health impactsin domestic and transboundary locations could supportimplementation of laws that hold responsible those indi-viduals and entities involved in illegal burning of landand forests in this region.

Acknowledgments

Code and supporting material for the GEOS-Chemmodel is available at www.geos-chem.org. Analysis ofMODIS and OMI satellite data used in this paper wereprocessed with the Giovanni online data system,developed and maintained by the NASA GES DISC.This work was funded by the Rockefeller Foundationand the Gordon and BettyMoore Foundation throughtheHealth& Ecosystems: Analysis of Linkages (HEAL)program. Miriam Marlier was funded through theWinslow Foundation. The authors would like to thankHan Sheng Quek and collaborators at the SingaporeNational Environment Agency for providing PMmeasurements.We also thank Santo V Salinas Cortijo,Susana Dorado, Brent Holben, Soo-Chin Liew, Mas-tura Mahmud, Maznorizan Mohamad, Lim HweeSan, and the rest of the AERONET team for their

8

Environ. Res. Lett. 11 (2016) 094023

efforts in establishing and maintaining the Singapore,Kuching, Pontianak, Palangkaraya, ND Marbel Uni-versity, and USM Penang sites. We also thank Guidovan der Werf for helpful comments on fire emissioninventories. Finally, we thank the Monitoring atmo-spheric composition and climate (MACC-II) projectfor processing, maintaining, and distributing theGFAS data. The GFAS data was taken from a websitemanaged by the ECMWF. Neither the EuropeanCommission nor the ECMWF accepts any responsi-bility or any liability for any use which is made ofthis data.

References

AndelaN, Kaiser JW,Heil A, van LeeuwenTT, van derWerf GR,WoosterM J, Remy S and SchultzMG2013Assessment oftheGlobal Fire Assimilation System (GFASv1)TechnicalMemorandum702ECMWF, Reading, UK

CarlsonKM,Curran LM, Ratnasari D, PittmanAM,Soares-Filho B S, AsnerGP, Trigg SN, GaveauDA,LawrenceD andRodriguesHO2012Committed carbonemissions, deforestation, and community land conversionfromoil palmplantation expansion inWest Kalimantan,IndonesiaProc. Natl Acad. Sci. 109 7559–64

ChangCH,Hsiao Y L andHwangC 2015 Evaluating spatial andtemporal variations of aerosol optical depth and biomassburning over southeast asia based on satellite data productsAerosol AirQual. Res. 15 2625–40

ChangC-P,WangZ,McBride J and LiuCH2005Annual cycle ofSoutheast Asia—Maritime continent rainfall and theasymmetricmonsoon transition J. Clim. 18 287–301

Cochrane J 2015Rain in Indonesia dampens forestfires that spreadtoxic hazeTheNewYork Times (http://nytimes.com/2015/10/29/world/asia/indonesia-forest-fire-toxic-haze.html?_r=0)

Dennis RA et al 2005 Fire, people and pixels: linking social scienceand remote sensing to understand underlying causes andimpacts offires in IndonesiaHum. Ecol. 33 465–504

Driscoll CT, Buonocore J J, Levy J I, Lambert K F, BurtrawD,Reid S B, FakhraeiH and Schwartz J 2015US power plantcarbon standards and clean air and health co-benefitsNat.Clim. Change 5 535–40

EffendyA andHardonoDS 2001The large scale private investmentof timber plantation development in Indonesia Proc. Int.Conf. on Timber Plantation (http://fao.org/docrep/005/ac781e/AC781E08a.htm)

Feintrenie L, ChongWKand Levang P 2010Why do farmers preferoil palm? Lessons learnt fromBungo district, IndonesiaSmall-Scale Forestry 9 379–96

Field RD and Shen S S P 2008Predictability of carbon emissionsfrombiomass burning in Indonesia from1997 to 2006J. Geophys. Res.: Biogeosci. 113G04024

Field RD, van derWerf GR and Shen S S P 2009Humanamplification of drought-induced biomass burning inIndonesia since 1960Nat. Geosci. 2 185–8

GaveauDLA et al 2014aMajor atmospheric emissions frompeatfires in Southeast Asia during non-drought years: evidencefrom the 2013 Sumatran fires Sci. Rep. 4 6112

GaveauDLA et al 2014b Four decades of forest persistence,clearance and logging on BorneoPLoSOne 9 e101654

HertwigD,Burgin L,GanC,HortM, JonesA, ShawF,WithamCandZhangK2015Development anddemonstrationof aLagrangian dispersionmodeling system for real timepredictionof smokehaze pollution frombiomassburning in SoutheastAsia J. Geophys.Res.: Atmos.120 12605–30

Holben BN et al 1998AERONET—a federated instrument networkand data archive for aerosol characterizationRemote Sens.Environ. 66 1–16

Johnston FH,Henderson SB, ChenY, Randerson J T,MarlierM,DeFries R S, Kinney P, BowmanDMJ S andBrauerM2012Estimated globalmortality attributable to smoke fromlandscapefiresEnviron. Health Perspect. 120 695–701

Kaiser JW et al 2012 Biomass burning emissions estimatedwith aglobalfire assimilation systembased on observed fireradiative powerBiogeosciences 9 527–54

KimP S, JacobD J,Mickley L J, Koplitz SN,MarlierME,DeFries R S,Myers S S, ChewBN andMaoYH2015Sensitivity of population smoke exposure to fire locations inEquatorial AsiaAtmos. Environ. 102 11–7

KingMD,MenzelWP, KaufmanY J, TanréD, GaoBC, Platnick S,Ackerman SA, Remer LA, Pincus P andHubanks PA 2003Cloud and aerosol properties, precipitable water, and profilesof temperature andwater vapor fromMODIS IEEETrans.Geosci. Remote Sens. 41 442–58

Koh L P,Miettinen J and Liew SC 2011Remotely sensed evidence oftropical peatland conversion to oil palm Proc. Natl Acad. Sci.108 5127–32

KonecnyK, BallhornU,Navratil P, Jubanski J, Page S E, Tansey K,Hooijer A, VernimmenR and Siegert F 2016Variable carbonlosses from recurrent fires in drained tropical peatlandsGlob.Change Biol. 22 1469–80

Lelieveld J, Evans J S, FnaisM,Giannadaki D and Pozzer A 2015Thecontribution of outdoor air pollution sources to prematuremortality on a global scaleNature 525 367–71

Levy J I, DiezD,DouY, Barr CDandDominici F 2012Ameta-analysis andmultisite time series analysis of the differentialtoxicity ofmajor fine particulatematter constituentsAm. J.Epidemiol. 175 1091–9

LevyRC, Remer LA andKleidmanRG2010Global evaluation ofthe collection 5MODIS dark-target aerosol products overlandAtmos. Chem. Phys. 10 10399–420

MargonoBA, Potapov PV, Turubanova S, Stolle F andHansenMC2014 Primary forest cover loss in Indonesia over 2000–2012Nat. Clim. Change 4 730–5

MarlierME,DeFries R, PenningtonD,Nelson E,Ordway EM,Lewis J, Koplitz SN andMickley L J 2015b Future fireemissions associatedwith projected land use change inSumatraGlob. Change Biol. 21 345–62

MarlierME,DeFries R S, KimP S,GaveauDLA,Koplitz SN,JacobD J,Mickley L J,Margono BA andMyers S S 2015aRegional air quality impacts of futurefire emissions inSumatra andKalimantan Environ. Res. Lett. 10 054010

MarlierME,DeFries R S, KimP S, Koplitz SN, JacobD J,Mickley L J andMyers S S 2015c Fire emissions and regionalair quality impacts fromfires in oil palm, timber, and loggingconcessions in Indonesia Environ. Res. Lett. 10 085005

MarlierME,DeFries R S, Voulgarakis A, Kinney P L, Randerson J T,Shindell DT, ChenY and Faluvegi G 2013 ElNiño and healthrisks from landscapefire emissions in Southeast AsiaNat.Clim. Change 3 131–6

Miettinen J, Shi C and Liew SC2011Deforestation rates in insularSoutheast Asia between 2000 and 2010Glob. Change Biol. 172261–70

Reid J S et al 2013Observing and understanding the Southeast Asianaerosol systemby remote sensing: an initial review andanalysis for the seven Southeast Asian studies (7SEAS)programAtmos. Res. 122 403–68

Reid J S, Xian P,Hyer E J, FlatauMK,Ramirez EM,Turk F J,SampsonCR, ZhangC, Fukada EMandMaloney ED2012Multi-scalemeteorological conceptual analysis of observedactivefire hotspot activity and smoke optical depth in themaritime continentAtmos. Chem. Phys. 12 2117–47

TanseyK, Beston J,Hoscilo A, Page S E and ParedesHernándezCU2008Relationship betweenMODIS fire hot spot count andburned area in a degraded tropical peat swamp forest inCentral Kalimantan, Indonesia J. Geophys. Res.: Atmos. 113D23112

TorresO,Decae R, Veefkind P and de LeeuwG2002OMI aerosolretrieval algorithmOMIAlgorithmTheoretical BasisDocument Volume III: Clouds, Aerosols, and SurfaceUVIrradiance ed P StammesATBD-OMI-O3, version 2 (http://

9

Environ. Res. Lett. 11 (2016) 094023

projects.knmi.nl/omi/documents/data/OMI_ATBD_Volume_3_V2.pdf)

TorresO,TanskanenA,VeihelmannB,AhnC,BraakR,BhartiaPK,VeefkindPandLevelt P 2007Aerosols and surfaceUVproductsfromozonemonitoring instrumentobservations: anoverviewJ.Geophys. Res.: Atmos.112D24S47

TuretskyMR,BenscoterB,PageS andReinG2015Global vulnerabilityof peatlands tofire and carbon lossNat.Geosci.811–4

vanderWerfGR et al2008Climate regulationoffire emissions anddeforestation in equatorialAsiaProc.NatlAcad. Sci.10520350–5

van derWerf GR, Randerson J T andGiglio L 2010Globalfireemissions and the contribution of deforestation, savanna,forest, agricultural, and peatfires (1997–2009)Atmos. Chem.Phys. 10 11707–35

WorldHealthOrganization 2009Global Health Risks:Mortality andBurden of Disease Attributable to SelectMajor Risks (Geneva,Switzerland:WorldHealthOrganization)

ZenderC S, Krolewski AG, ToscaMGandRanderson J T 2012Tropical biomass burning smoke plume size, shape,reflectance, and age based on 2001–2009MISRimagery of BorneoAtmos. Chem. Phys. 12 3437–54

10

Environ. Res. Lett. 11 (2016) 094023