GLOBAL FIRE ACTIVITY

A human-driven decline in globalburned areaN Andela12 D C Morton1 L Giglio3 Y Chen2 G R van der Werf4

P S Kasibhatla5 R S DeFries6 G J Collatz1 S Hantson7 S Kloster8 D Bachelet9

M Forrest10 G Lasslop8 F Li11 S Mangeon12 J R Melton13

C Yue14 J T Randerson2

Fire is an essential Earth system process that alters ecosystem and atmosphericcomposition Here we assessed long-term fire trends using multiple satellite data sets Wefound that global burned area declined by 243 plusmn 88 over the past 18 years Theestimated decrease in burned area remained robust after adjusting for precipitationvariability and was largest in savannas Agricultural expansion and intensification wereprimary drivers of declining fire activity Fewer and smaller fires reduced aerosolconcentrations modified vegetation structure and increased the magnitude of theterrestrial carbon sink Fire models were unable to reproduce the pattern and magnitude ofobserved declines suggesting that they may overestimate fire emissions in futureprojections Using economic and demographic variables we developed a conceptual modelfor predicting fire in human-dominated landscapes

Fires play an integral role in shaping ecosys-tem properties (1) and have widespreadimpacts on climate biogeochemical cyclesand human health (2ndash4) Frequent fires areessential for maintaining savanna ecosys-

tems (5) whereasmore episodic events in temper-ate and boreal forests create a mosaic of habitatsin different stages of postfire succession (6) In-troduction or exclusion of fire from the landscapemay lead to rapid shifts in vegetation structureand composition (5) carbon stocks (7) and bio-diversity (8) Globally fire emissions are re-sponsible for 5 to 8 of the 33 million annualpremature deaths frompoor air quality and fireis the primary cause of elevated mortality fromair pollution across much of the tropics (3) Firesaffect global climate through changes in veg-etation and soil carbon (7) surface albedo (9)and atmospheric concentrations of aerosols andgreenhouse gases (10) Climate feedbacks onfire activity are complex and vary by biome andlevel of fire suppression (11) Given projectedincreases in fire risk from climate change (12)fire management will be increasingly importantfor maintaining ecosystem function air qualityand other services that influence human well-being (13)Climate is a dominant control on fire activity

regulating vegetation productivity and fuel mois-ture Over short time scales rainfall during thedry season suppresses fire activity whereas over

longer time scales fuel build-up duringwet yearsin more arid ecosystems can increase burnedarea in subsequent years (11) The redistributionof precipitation in response to El NintildeondashSouthernOscillation (ENSO) and other climate modestherefore has a large and sometimes contrastingeffect on the interannual variability of biomassburning across continents (6 14 15) Oscillationsin Pacific and Atlantic sea surface temperaturesalso influence trends in fire activity on longertime scales (6 15) Climate change may increasefire risk inmany regions (12 16) given projectedwarming and drying in forests and other biomeswith sufficient fuel loads to support fire activ-ity Ultimately the interactions among climatevegetation and ignition sources determine thespatial and temporal pattern of biomass burn-ing (17)In addition to natural processes humans have

shaped patterns of global burning for millennia(4) and human activity is now the primarysource of ignitions in tropical forests savannasand agricultural regions (14 18) Human influ-ence on fire dynamics depends on populationdensity (17) socioeconomic development (19)landscape fragmentation (18) and land man-agement (15) as people introduce or suppressfires (4) andmanipulate the timing and fuel con-ditions of fires in human-dominated landscapes(11 14) During the past two decades humanpopulation increased by about 25 or 15 billion

(20) and agricultural production increased bymore than 40 (21) Today about 36 of theworldrsquos land surface is used for pasture or crop-lands (22) directly affecting the way fire ismanaged within these ecosystems Earlier workhas demonstrated that cropland expansion ordeforestation rates are closely linked to regionalfire trends (14 15) and for many regions chang-ing fire activity in recent decades extends a long-term transition fromnatural to human-dominatedfire regimes (23ndash25) However global implicationsof changingagriculturalmanagementand themech-anisms that regulate fires in human-dominatedlandscapes remain poorly understood Even inareas dominated by human sources of ignitionvariations in precipitation and other weather con-ditions may modulate year-to-year variations inignition efficiency fire spread rate and fire sizeThe interactions among fire weather fuels andignition therefore make it challenging to sepa-rate climate andhuman controls on fire dynamicsat regional and global scales Nevertheless thisseparation is necessary to build and improve pre-dictive fire modelsSatellite-derived burned area data provide a

consistent global perspective on changingpatternsof fire activityHerewe analyzed long-term trendsin burned area from 1998 through 2015 using theGlobal Fire Emissions Database version 4 productthat includes small fires (GFED4s) (26 27) We con-ducted several analyses to assess the drivers andimplications of long-term trends in burned area(28) First we estimated the influence of precip-itation on burned area variability and trends ineach 025deg grid cell using a statistical model toseparate climate and human contributions to firetrends (fig S1) Second we separated burned areainto contributions from the number and sizeof individual large fires using 500-m-resolutionburnedareadata fromNASArsquosModerateResolutionImaging Spectroradiometer (MODIS) sensors(29) during 2003 to 2015 (fig S2) Third wecompared observed trends with prognostic firemodel estimates from the Fire Model Inter-comparison Project (FireMIP) with the aim ofunderstanding limits to fire prediction Fourthwe examined spatial and temporal relationshipsbetween burned area and socioeconomic data toinvestigate patterns of human influence on fireactivity On the basis of these data we developeda conceptual model of fire use that considers theroles of land management and socioeconomic de-velopment Finally we assessed the impact ofthe observed decreasing burned area trend onecosystem structure the magnitude of the ter-restrial carbon sink and atmospheric composi-tion in biomass burning regions Together these

RESEARCH

Andela et al Science 356 1356ndash1362 (2017) 30 June 2017 1 of 6

1Biospheric Sciences Laboratory NASA Goddard Space Flight Center Greenbelt MD 20771 USA 2Department of Earth System Science University of California Irvine CA 92697 USA3Department of Geographical Sciences University of Maryland College Park MD 20742 USA 4Faculty of Earth and Life Sciences Vrije Universiteit Amsterdam Amsterdam Netherlands5Nicholas School of the Environment Duke University Durham NC 27708 USA 6Department of Ecology Evolution and Environmental Biology Columbia University New York NY 10027 USA7Karlsruhe Institute of Technology Institute of Meteorology and Climate Research Atmospheric Environmental Research 82467 Garmisch-Partenkirchen Germany 8Max Planck Institute forMeteorology Bundesstraszlige 53 20164 Hamburg Germany 9Biological and Ecological Engineering Oregon State University Corvallis OR 97331 USA 10Senckenberg Biodiversity and ClimateResearch Institute (BiK-F) Senckenberganlage 25 60325 Frankfurt am Main Germany 11International Center for Climate and Environmental Sciences Institute of Atmospheric Physics ChineseAcademy of Sciences Beijing China 12Department of Physics Imperial College London London UK 13Climate Research Division Environment Canada Victoria BC V8W 2Y2 Canada14Laboratoire des Sciences du Climat et de lrsquoEnvironnementndashInstitute Pierre Simon Laplace Commissariat agrave lrsquoEacutenergie Atomique et aux Eacutenergies Alternatives (CEA)ndashCentre National de laRecherche Scientifique (CNRS)ndashUniversiteacute de Versailles Saint Quentin Universiteacute Paris-Saclay 91198 Gif-sur-Yvette FranceCorresponding author Email nielsandelanasagov

on Septem

ber 23 2020

httpsciencesciencemagorg

Dow

nloaded from

analyses underscore the pervasive influence ofhuman activity on global burned area includ-ing the potential for further declines in savannafires from ongoing agricultural development ac-ross the tropics

Trends in burned area

Global burned area declined by nearly one-quarter between 1998 and 2015 (ndash243 plusmn 88or ndash135 plusmn 049 yearminus1) Large decreases oc-curred in tropical savannas of South Americaand Africa and grasslands across the Asian steppe(Fig 1) Globally decreases were concentrated inregions with low and intermediate levels of treecover whereas an increasing trend was observedin closed-canopy forests Declining trends wererobust when assessed with different burned areadata sets and time intervals (Table 1 and fig S3)Burned area from GFED4s and the 500m MODISproduct showed a similar decline during 2003 to2015 (ndash128 plusmn 096 yearminus1 and ndash115 plusmn 121 yearminus1respectively) and satellite-based active fire detec-tions fromMODIS provided an independent con-firmation of the patterns of decreasing global fireactivity (fig S4) Regional increases in burnedarea were also observed but areas with a sig-nificant decline (P lt 005) in burned area out-numbered areas with a significant increase in

burned area for all continents except Eurasia(fig S5) For tropical savannas and grasslandsdeclines outnumbered increases by 31 Withinindividual continents strong contrasting trendswere observed between northern and southernAfrica and between Central America and tem-perate North America (table S1)Rainfall patterns explained much of the inter-

annual variability in burned area but little of thelong-term decline (Table 1 and Fig 2A) Buildingon previous work (15) we developed a linearmodel to adjust for the influence of precipitationvariability on burned area (28) (fig S6) Long-term trends weremore significant after reducingprecipitation-driven variability For example theglobal decline in burned area of ndash115 plusmn 121yearminus1 (2003 through 2015 P lt 01) in the 500mMODIS time series strengthened to ndash123 plusmn 044yearminus1 (P lt 001) after adjusting for precipitation(Table 1)Regionally precipitation-adjustedburnedarea time series showed significant declines inCentral America Northern Hemisphere SouthAmerica Europe Northern Hemisphere Africaand Central Asia (table S1)A decrease in the number of fires explained

most of the global decline in burned area with asmaller contribution from decreasing mean firesize (Fig 2) The relative contributions from fire

number and fire size to observed trends variedconsiderably among regions (Table 1 and fig S7)In northern Africa the number and mean size offires contributed nearly equally to the net declinein burned area By contrast a decrease in thenumber of fires was the primary factor causing adecline in burned area in South America CentralAmerica and Central Asia (table S1) Regionaland interannual variability in the size distribu-tion of individual fires provided new informationabout the combined influence of climate land-scape fragmentation andmanagement on burnedarea this information is essential for improvingrepresentation of fire in Earth system modelsCurrent global fire models were unable to

predict the magnitude or spatial pattern of theobserved decline in global burned area (Fig 3and figs S8 to S10) During 1997 to 2013 FireMIPmodels (n = 9) predicted a mean trend in globalburned area of ndash013 plusmn 056 yearminus1 compared tothe observed trend of ndash109 plusmn 061 yearminus1 forthis interval (28) (Table 1 and table S2) Focusingon the three models that account for humancontributions to the number and size of fires(table S3) two models predicted a small declinein global burned area but often poorly simulatedthe spatial structure of trends across differentcontinents (Fig 3 and fig S8) Despite including

Andela et al Science 356 1356ndash1362 (2017) 30 June 2017 2 of 6

Burned area ( yrndash1)

Burned area trend ( yrndash2)

1

5

10

25

50

100

10 20 30 40 50 60 70 80 90Fractional tree cover ()

0123456

BA

(yr

minus1)

minus 250

minus 100minus 050minus 025

minus 010

000

010

025050100

250

10 20 30 40 50 60 70 80 90Fractional tree cover ()

-3-2-10123

BA

tren

d(

yrminus

2)

Fig 1 Satellite observations show a declining trend in fire activity across the worldrsquos tropical and temperate grassland ecosystems and land-usefrontiers in the Americas and Southeast Asia (A) mean annual burned area and (B) trends in burned area (GFED4s 1998 through 2015) Line plots (inset)indicate global burned area and trend distributions by fractional tree cover (28)

RESEARCH | RESEARCH ARTICLEon S

eptember 23 2020

httpsciencesciencem

agorgD

ownloaded from

land use and population as input variables sev-eral models predicted increasing burned area(28) (table S4) consistent with global trends infire weather (16) These model-data differenceshighlight the importance of human activity in re-ducing burning despite growing climate-drivenfire risk

Humans as a driver of the long-term trend

Population cropland area and livestock densitywere important factors constraining landscapepatterns of burning yet the sign and magnitudeof the spatial correlation coefficient between thesevariables and burned area varied across biomesand along gradients of tree cover (Fig 4) Allthree indicators had negative spatial correlationswith burned area in savannas and grasslandsAlthough these three variables had similar globalstructure we found that the distribution of agri-cultural activity clearly modified burned areabeyondpopulation alone For examplewidespreadagricultural waste burning in large parts of Asiagenerated a strong positive correlation betweencropland and burned area Similarly livestockdensity and burned area were negatively cor-related in the Brazilian Cerrado as livestockmaydirectly suppress fire activity by reducing fuelloads or altering fire management decisions Intropical forests population density and croplandwere positively correlated with the spatial pat-tern of burned area as humans have introducedfires for deforestation and agricultural manage-ment (7 27) In boreal forests we found a stron-

ger positive relationship between populationand burned area in Eurasia than North Americaconsistent with past work documenting highlevels of human-driven fire activity inRussia (30)Trends in agricultural production and fire ac-tivity were also consistent at the national scaleThe largest relative declines in GFED4s burnedarea occurred in countries with the largest in-creases in agricultural extent and production val-ue (fig S11)We developed a conceptual model of changes

inburned areawith increasingdevelopment basedon spatial patterns of burned area land use pop-ulation density and gross domestic product (GDP)data (Fig 5) Our analysis showed that the evolu-tion of human-dominated fire regimes follows pre-dictable patterns with the transition from naturalto managed landscapes in forest and savanna re-gions generating markedly different burned areatrajectories (28) (figs S12 and S13) For humidtropical forests frequent fires for deforestationand agricultural management yielded a sharprise in fire activity with the expansion of settledland uses providing quantitative evidence forrapid ecosystem transformation during earlyland-use transitions described in previous work(31 32) However in semi-arid savannas andgrasslands the transition from natural land-scapeswith common landownership to agricultureon private lands generated a nonlinear decreasein fire activity even in areas without large-scaleland cover conversion The reorganization of landcover and fire use on the landscape also altered

the contributions fromdifferent fire types to totalburned area (Fig 5) For both forested and sa-vanna regions the most rapid changes in bothland cover and total burned area occurred fortransitions at very low levels of per capita GDP(lt$5000 kmminus2 yearminus1 figs S12 and S13)With an expanding human presence on the

landscape increasing investment in agricul-tural areas reduced fire activity in both savannasand forests (Fig 5) In highly capitalized re-gions burned areawas considerably lower likelyas a consequence of both mechanized (fire-free)management and fire suppression to protect high-value crops livestock homes infrastructureand air quality (13) (Figs 4 and 5 and fig S11)Livelihoods change drastically along this tra-jectory of fire use as does the perception of fireand smoke (23) Regulation to improve air qualityhas significantly reduced cropland burning in thewestern United States (33) By contrast fire ac-tivity increased in some densely populated agri-cultural regions of India andChina (Figs 1 and 4)suggesting that without investments in air qual-itymanagement agricultural intensificationmayincrease fire activity in regions where crop res-idue burning is the dominant fire type Agricul-tural expansion and intensification are likely tocontinue in coming decades (21) with the largestchanges expected in the tropics as developmentshifts vast areas of common land or extensiveland uses toward more capital-intensive agricul-tural production for regional or global markets(21 32) These changes in land use suggest that

Andela et al Science 356 1356ndash1362 (2017) 30 June 2017 3 of 6

Table 1 Relative trends in burned area number of fires and mean fire size for different regions of the worldTrends are shown for different timeperiods as indicated to directly compare burned area estimates from different sources All trends were calculated by using fire season estimates of burned

area with the exception of the FireMIP data which were produced per calendar year (28) Increases (regular type) and decreases (bold) in burned area are

indicated for each region and time period significant trends are denoted by asterisks (P lt 01 P lt 005 and P lt 001)

Fire productTime

period

Full or

residual

Trend ( yearminus1) with 95 confidence limits

WorldNorth

America

South

AmericaEurasiadagger

Southeast

AsiaAfrica Australia

Burned area 1998ndash2015Full ndash135 (049) ndash037 (266) ndash168 (275) ndash080 (198) 023 (194) ndash127 (032) ndash253 (423)

(GFED4s) PA ndash099 (029) 040 (213) ndash051 (168) ndash026 (134) 025 (132) ndash126 (025) ndash067 (191)

Burned area

2003ndash2015

Full ndash127 (095) ndash008 (217) ndash266 (538) ndash218 (298) ndash030 (323) ndash151 (051) 148 (795)

(GFED4s)

during the

MODIS era

PA ndash117 (039) 033 (177) ndash175 (314) ndash124 (196) ndash013 (202) ndash145 (042) 039 (316)

Burned area from2003ndash2015

Full ndash115 (121) 061 (276) ndash140 (699) ndash223 (413) ndash062 (392) ndash160 (056) 153 (821)

500m MODIS

MCD64A1PA ndash123 (044) 078 (212) ndash109 (364) ndash102 (235) ndash061 (246) ndash168 (046) 056 (340)

Fire number 2003ndash2015Full ndash098 (073) ndash144 (292) ndash267 (498) ndash356 (439) ndash110 (336) ndash087 (038) 150 (659)

PA ndash100 (035) ndash141 (237) ndash213 (255) ndash346 (306) ndash103 (196) ndash088 (030) 073 (283)

Fire size 2003ndash2015Full ndash039 (038) 047 (173) 139 (232) ndash009 (203) ndash068 (099) ndash078 (029) ndash029 (169)

PA ndash043 (018) 032 (128) 129 (128) ndash016 (136) ndash069 (085) ndash081 (021) 008 (111)

Burned area

predicted

by FireMIPDagger

1997ndash2013 Full ndash013 (056) ndash036 (137) ndash006 (077) 006 (053) ndash189 (195) ndash025 (070) ndash050 (242)

Residual time series after adjusting for the influence of precipitation variability (PA) were estimated by using the approach described in the supplementarymaterials daggerEurasia excludes regions in Southeast Asia DaggerIn this instance numbers in parentheses are the standard deviation of the trend averaged across thedifferent FireMIP models (n = 9)

RESEARCH | RESEARCH ARTICLEon S

eptember 23 2020

httpsciencesciencem

agorgD

ownloaded from

observed declines in burned area may continueor even accelerate in coming decadesSuccessful prediction of fire trends on decadal

time scales requires a mechanistic description offire use during the different phases of develop-ment shown in Fig 5 Considering the observa-tional constraints described here we identifiedthree primary reasons why the FireMIP modelswere unable to reproduce the observed decline inglobal burned area First all FireMIP modelsunderestimated the magnitude of burned areadeclines in areas with moderate and high den-

sities of population and per capita GDP (fig S14)suggesting that the models were not sensitiveenough to the influence of economic develop-ment on fire activity Second although many ofthe FireMIP models included pasture area as avariable describing human modification of landcover burning in pasture areas was often treatedthe same as burning in grasslands (table S3) andin many tropical countries most land areas avail-able for grazing had been converted to pasture bythe 1970s The relative saturation of changingpasture area during the past two decades con-

trasted sharply with very large increases in live-stock density (fig S15) highlighting the importanceof better integrating drivers of land-use intensi-fication within prognostic models Third firemodels overestimated burned area in semi-aridtropical ecosystems and underestimated burnedarea in mesic and humid tropical ecosystems(fig S14) This bias in the spatial distribution ofburned area may have weakened the modelsrsquooverall sensitivity to human development driversbecause population and wealth changes weremore pronounced in areas with higher levels of

Andela et al Science 356 1356ndash1362 (2017) 30 June 2017 4 of 6

Fig 3 Comparison of burned areatrends from satellite observations(GFED4s) and prognostic firemodels fromFireMIP (A) Time seriesof global burned area (B) A compar-ison of global mean annual burnedarea versus the relative trend in globalmean burned area from the observa-tions and models GFED4s observa-tions are shown in black and FireMIPmodels are shown with differentcolors FireMIP model estimates wereavailable from 1997 through 2013for six models from 1997 through2012 for the CTEM fire module andJULES-INFERNO and from 1997through 2009 for MC-FireThe FireMIPmodels are described in more detail inthe supplementary materials and byRabin et al (34) (tables S3 and S4 andfig S8)

Fig 2 A decrease in the number of fires wasthe primary driver of the global decline inburned area Normalized variation (2003 = 1)and linear trends in (A) burned area(B) number of fires and (C) mean fire sizederived from the MODIS 500m product(MCD64A1) Shading denotes 95 predictionintervals Adjusting for precipitation-driventrends in burned area isolated residual trendsassociated with other factors including humanactivity (28) (D) Summary of trends in globalburned area calculated as the product of thenumber and size of fires after adjusting for theinfluence of precipitation Regional trends in firenumber and fire size are provided in Table 1table S1 and fig S7

RESEARCH | RESEARCH ARTICLEon S

eptember 23 2020

httpsciencesciencem

agorgD

ownloaded from

rainfall Whereas the spatial pattern of burnedarea has been widely used as a target for firemodel development in past work (34) (fig S9)our analysis highlights the importance of usingtrend and fire size observations to constrainscenarios of future fire activity (fig S10 andtable S5)

Implications of declining globalfire activity

The observed large-scale decline of burned areain the worldrsquos grasslands savannas and tropicalland-use frontiers had broad consequences forvegetation dynamics carbon cycling air qualityand biodiversity Fires play an important role inregulating the competition between herbaceousand woody vegetation (5) In grid cells whereburned areawas equal to or exceeded 10 yearminus1we found that the spatial pattern of trends inboth dry season enhanced vegetation index(EVI) and vegetation optical depth (VOD) wasnegatively correlated with trends in burnedarea consistent with woody encroachment inareas with declining burned area (28) (table

S6 and fig S16) However further analysis ofhigher-resolution satellite imagery is necessaryto quantify the magnitude of encroachment inareas with observed fire trends as well as othermechanisms that may influence vegetationindicesLess-frequent burning also allowed biomass

litter and soil organic matter stocks to accumu-late contributing to a 02 Pg yearminus1 carbon sinkby 2015 in tropical and temperate savannas andgrasslands (28) (fig S17 40degN to 40degS) Placingthis estimate in the context of the global carboncycle declining savanna and grassland fires ac-counted for about 7 of the contemporary globalnet land flux (35) and were likely an importantdriver of the large and variable terrestrial carbonsink previously reported in semi-arid ecosystems(36) Our estimate of the fire contribution to theterrestrial carbon sink is likely conservative be-cause the biogeochemicalmodel we used did notaccount for burned area changes prior to 2001declining emissions in deforestation zones orwoody encroachment in regions with decliningfire frequency (27 28)

Analysis of satellite observations of aerosoland carbon monoxide concentrations providedindependent evidence of declining fire emis-sions Declining fire emissions lowered aerosolconcentrations in themajor tropical biomass burn-ing regions during the fire season (fig S4b) lead-ing to improved air quality and regional changesin radiative forcing and atmospheric compositionAlthough atmospheric transport distributes fireemissions across large areaswe identified a stronglocal effect of declining burned area on aerosolabsorption optical depth in frequently burninggrid cells (correlation coefficient r=026Plt001table S6) Similarly declining fire activity alsolowered regional carbon monoxide concentra-tions (r = 011 P lt 001 fig S4c) suggestingthat decreasing biomass burning emissions mayhave partly offset other drivers of increasingatmospheric methane (37)Declining fire frequency supports climatemiti-

gation efforts but may run counter to conserva-tion objectives in fire-dependent ecosystemsFrequent fires are a key aspect of many ancientgrassland ecosystems that support a range ofendemic species (38) and a large portion of theworldrsquos remaining wild large mammals (39) Themagnitude of habitat and biodiversity losses fromdeclining burned area in savanna and grasslandecosystems may equal or exceed other humanimpacts in the tropics (Fig 1) but these impactshave been largely neglected by the internationalcommunity (40) Challenges for conserving savannaecosystems abound trade-offs among conserva-tion climatemitigation human health and agri-cultural productionwill ultimately determine thebalance of fire activity in savannas and grasslands(Figs 4 and 5 and figs S4 S16 and S17)The strong and sustained burned area declines

in grasslands and savannas documented hererepresent a first-order impact on the Earth sys-tem with consequences for ecosystems and cli-mate thatmay be comparablewith those of otherlarge-scale drivers of global change A shift towardmore capital-intensive agriculture has led to fewerand smaller fires driven by population increasessocioeconomic development and demand foragricultural products from regional and globalmarkets Together these factors influence fireuse in predictable ways with a strong inverserelationship between burned area and economicdevelopment (Fig 5) The pervasive influence ofhuman activity on burned area was not capturedby state-of-the-art fire models improving thesemodels in the futuremay require amore sophis-ticated representation of land-use intensificationand its influence on fire dynamics Despite poten-tial increasing fire risk fromclimate change (12 16)ongoing socioeconomic development will likelysustain observed declines in fire in savanna andgrassland ecosystems in coming decades alteringvegetation structure and biodiversity Fire is oneof the oldest tools for human landscape manage-ment yet the use of fire is rapidly changing inresponse to the expansion of global agricultureAchieving a balance between conservation of fire-dependent ecosystems and increasing agricul-tural production to support growing populations

Andela et al Science 356 1356ndash1362 (2017) 30 June 2017 5 of 6

Fig 4 Population and agriculture influence the spatial pattern of burned area with contrast-ing impacts in different biomes Maps of the spatial correlation between burned area and(A) population density per km2 (B) fractional cropland area and (C) livestock density per km2Map panels indicate the spatial correlation between burned area (GFED4s) and human landuse for the 36 025deg pixels within each 15deg grid cell Line plots (inset) show the mean correlation asa function of fractional tree cover (28)

RESEARCH | RESEARCH ARTICLEon S

eptember 23 2020

httpsciencesciencem

agorgD

ownloaded from

will require careful management of fire activity inhuman-dominated landscapes

REFERENCES AND NOTES

1 W J Bond F I Woodward G F Midgley New Phytol 165525ndash538 (2005)

2 P J Crutzen M O Andreae Science 250 1669ndash1678(1990)

3 J Lelieveld J S Evans M Fnais D Giannadaki A PozzerNature 525 367ndash371 (2015)

4 D M J S Bowman et al Science 324 481ndash484 (2009)5 R J Scholes S R Archer Annu Rev Ecol Syst 28 517ndash544

(1997)6 T W Swetnam J L Betancourt Science 249 1017ndash1020 (1990)

7 S E Page et al Nature 420 61ndash65 (2002)8 M A Cochrane M D Schulze Biotropica 31 2ndash16 (1999)9 J T Randerson et al Science 314 1130ndash1132 (2006)10 D S Ward et al Atmos Chem Phys 12 10857ndash10886

(2012)11 S Archibald A Nickless N Govender R J Scholes V Lehsten

Glob Ecol Biogeogr 19 794ndash809 (2010)12 O Pechony D T Shindell Proc Natl Acad Sci USA 107

19167ndash19170 (2010)13 M A Moritz et al Nature 515 58ndash66 (2014)14 L E Aragatildeo et al Philos Trans R Soc London B Biol Sci

363 1779ndash1785 (2008)15 N Andela G R van der Werf Nat Clim Chang 4 791ndash795

(2014)16 W M Jolly et al Nat Commun 6 7537 (2015)

17 I Bistinas S P Harrison I C Prentice J M C PereiraBiogeosciences 11 5087ndash5101 (2014)

18 S Archibald D P Roy B W van Wilgen R J ScholesGlob Change Biol 15 613ndash630 (2009)

19 E Chuvieco C O Justice in Advances in Earth Observation ofGlobal Change E Chuvieco J Li X Yang Eds (2010)pp 187ndash199

20 United States Census Bureau Historical estimates of worldpopulation wwwcensusgov

21 N Alexandratos J Bruinsma Food and AgricultureOrganization World agriculture towards 20302050 The 2012revision ESA Work Pap No 12-03 (2012)

22 N Ramankutty A T Evan C Monfreda J A Foley GlobalBiogeochem Cycles 22 GB1003 (2008)

23 S J Pyne Fire in America A Cultural History of Wildland andRural Fire (Princeton Univ Press 1982)

24 M J E van Marle et al Historic global biomass burningemissions based on merging satellite observations with proxiesand fire models (1750-2015) Geosci Model Dev Discuss(2017) doi 105194gmd-2017-32

25 J R Marlon et al Nat Geosci 1 697ndash702 (2008)26 J T Randerson Y Chen G R van der Werf B M Rogers

D C Morton J Geophys Res 117 G04012 (2012)27 G R van der Werf et al Global fire emissions estimates

during 1997-2015 Earth Syst Sci Data Discuss 105194essd-2016-62 (2017)

28 Materials and methods are available as supplementary materials29 L Giglio T Loboda D P Roy B Quayle C O Justice

Remote Sens Environ 113 408ndash420 (2009)30 D Mollicone H D Eva F Achard Nature 440 436ndash437 (2006)31 J A Foley et al Science 309 570ndash574 (2005)32 T K Rudel R Defries G P Asner W F Laurance Conserv

Biol 23 1396ndash1405 (2009)33 H-W Lin et al J Geophys Res Biogeosci 119 645ndash660 (2014)34 S S Rabin et al Geosci Model Dev 10 1175ndash1197 (2017)35 C Le Queacutereacute et al Earth Syst Sci Data 7 47ndash85 (2015)36 A Ahlstroumlm et al Science 348 895ndash899 (2015)37 A L Rice et al Proc Natl Acad Sci USA 113 10791ndash10796

(2016)38 W J Bond Science 351 120ndash122 (2016)39 G P Hempson S Archibald W J Bond Science 350

1056ndash1061 (2015)40 C L Parr C E R Lehmann W J Bond W A Hoffmann

A N Andersen Trends Ecol Evol 29 205ndash213 (2014)

ACKNOWLEDGMENTS

NA YC and JTR received funding from the Gordon and Betty MooreFoundation (grant GBMF3269) DCM was supported by NASArsquosInterdisciplinary Science and Carbon Monitoring System ProgramsGRvdW was supported by the Netherlands Organisation for ScientificResearch (NWO) SH by the EU FP7 projects BACCHUS (grant603445) and LUC4C (grant 603542) FL by the National ScienceFoundation of China (grant 41475099) and CY by the EuropeanSpace Agency Fire_CCI project We thank M N Deeter for helpfulsuggestions on the CO analysis The authors declare that they have nocompeting interests Data used in this study are available atwwwglobalfiredataorg httpsreverbechonasagov httpgpcpumdedu wwwfaoorg and httpwebornlgovscilandscan andare described in more detail in the supplementary materials FireMIPmodel simulation output is archived with the supporting informationand full data sets are available on request

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent35663451356supplDC1Materials and MethodsFigs S1 to S18Tables S1 to S6References (41ndash90)

22 November 2016 accepted 2 June 2017101126scienceaal4108

Andela et al Science 356 1356ndash1362 (2017) 30 June 2017 6 of 6

Fig 5 Conceptual model showing changes in fire use along the continuum from commonland ownership to highly capitalized agricultural management on private lands In humidtropical regions (A and B) (precipitation ge1200 mm yearminus1) deforestation fires for agriculturalexpansion (A) lead to peak burned area during an early land-use transition phase to more settledland uses (B) In the semi-arid tropics (C and D) (precipitation 500 to 1200 mm yearminus1) burned areais highest under common land ownership (D) as intact savanna and grazing lands allow for thespread of large fires Conversion of savanna and grassland systems for more permanent agriculture(C) drives a nonlinear decline in burned area from landscape fragmentation and changing fire use foragricultural management (D) The conceptual model is based on the spatial distribution of burnedarea land use population and GDP (28) (figs S12 and S13) Similar patterns are observed across allcontinents but absolute burned area differs as a function of culture climate and vegetation

RESEARCH | RESEARCH ARTICLEon S

eptember 23 2020

httpsciencesciencem

agorgD

ownloaded from

A human-driven decline in global burned area

Kloster D Bachelet M Forrest G Lasslop F Li S Mangeon J R Melton C Yue and J T RandersonN Andela D C Morton L Giglio Y Chen G R van der Werf P S Kasibhatla R S DeFries G J Collatz S Hantson S

DOI 101126scienceaal4108 (6345) 1356-1362356Science

this issue p 1356Sciencechanges to the atmosphere vegetation and the terrestrial carbon sinkof agricultural expansion and intensification The decline of burned area has consequences for predictions of futurethe past 18 years despite the influence of climate The decrease has been largest in savannas and grasslands because

25 oversim use satellite data to show that unexpectedly global burned area declined by et alwarming world Andela Humans have and always have had a major impact on wildfire activity which is expected to increase in our

Burn less baby burn less

ARTICLE TOOLS httpsciencesciencemagorgcontent35663451356

MATERIALSSUPPLEMENTARY httpsciencesciencemagorgcontentsuppl2017062835663451356DC1

REFERENCES

httpsciencesciencemagorgcontent35663451356BIBLThis article cites 80 articles 14 of which you can access for free

PERMISSIONS httpwwwsciencemagorghelpreprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAASScienceScience 1200 New York Avenue NW Washington DC 20005 The title (print ISSN 0036-8075 online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science No claim to original US Government WorksCopyright copy 2017 The Authors some rights reserved exclusive licensee American Association for the Advancement of

on Septem

ber 23 2020

httpsciencesciencemagorg

Dow

nloaded from