Environ. Res. Lett. 11 (2016) 125008 doi:10.1088/1748-9326/11/12/125008
LETTER
Variability of fire emissions on interannual tomulti-decadaltimescales in two Earth Systemmodels
DSWard1, E Shevliakova2, SMalyshev2, J-F Lamarque3 andATWittenberg4
1 Program inAtmospheric andOceanic Sciences, PrincetonUniversity, Princeton, NJ,USA2 Department of Ecology and Evolutionary Biology, PrincetonUniversity, Princeton, NJ,USA3 National Center for Atmospheric Research, Boulder, CO,USA4 NOAA/Geophysical FluidDynamics Laboratory, Princeton,NJ,USA
Keywords: climate variability, fire emissions, Earth systemmodels
Supplementarymaterial for this article is available online
AbstractConnections betweenwildfires andmodes of variability in climate are sought as ameans for predictingfire activity on interannual tomulti-decadal timescales. Severalfire drivers, such as temperature andlocal drought index, have been shown to vary on these timescales, and analysis of tree-ring datasuggests covariance betweenfires and climate oscillation indices in some regions. However, theshortness of the satellite record of globalfire events limits investigations on larger spatial scales. Herewe explore the interplay between climate variability andwildfire emissions with the preindustrial longcontrol numerical experiments and historical ensembles of CESM1 and theNOAA/GFDLESM2Mb.Wefind that interannual variability infires is underpredicted in both Earth Systemmodels (ESMs)compared to present dayfire emission inventories.Modeled fire emissions respond to the ElNiño/southern oscillation (ENSO) andPacific decadal oscillation (PDO)with increases in southeast Asiaand borealNorthAmerica emissions, and decreases in southernNorthAmerica and Sahel emissions,during the ENSOwarmphase in both ESMs, and the PDOwarmphase inCESM1. Additionally,CESM1produces decreases in boreal northern hemisphere fire emissions for thewarmphase of theAtlanticMeridionalOscillation. Through analysis of the long control simulations, we show that the20th century trends in both ESMs are statistically significant,meaning that the signal of anthropogenicactivity on fire emissions over this time period is detectable above the annual to decadal timescalenoise. However, the trends simulated by the two ESMs are of opposite sign (CESM1decreasing,ESM2Mb increasing), highlighting the need for improved understanding, proxy observations, andmodeling to resolve this discrepancy.
Introduction
Extreme fire emissions from Indonesia in 2015were inthe global news and were linked to major degradationof regional air quality (Chisholm et al 2016), which hasbeen suggested to contribute to increased mortality inSoutheast Asia (Marlier et al 2012). These fires werebrought on by dry conditions associated with thestrong 2015–2016 El Niño and, preliminarily, arethought to be the most extreme episode of fires in thisregion since the 1997–1998 El Niño and could bringeven higher economic costs (Chisholm et al 2016).
Factors underlying fires in this region, especiallyexposure of large amounts of peat following forestclearing, exacerbate the potential severity of theburning and subsequent emissions (Marlieret al 2014, 2015), but it is natural climate variabilitythat drives the timing and scale of these events (Chenet al 2011, 2016). Fire variability in other regions hasalso been connected to the El Niño/southern oscilla-tion (ENSO) (e.g. Heyerdahl et al 2008, Le Pageet al 2008, Monks et al 2012) and the Pacific decadaloscillation (PDO) (e.g. Duffy et al 2005, Kitzbergeret al 2007). The impacts of these variations are felt
OPEN ACCESS
RECEIVED
29April 2016
REVISED
14November 2016
ACCEPTED FOR PUBLICATION
22November 2016
PUBLISHED
2December 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.
globally; for example, fires are the main cause ofinterannual variability (IAV) in global carbon mon-oxide concentrations and carbonaceous aerosol opti-cal depth (Voulgarakis et al 2015), with implicationsfor global atmospheric chemistry, cloud physics, andradiative forcing (e.g. Naik et al 2005, Randersonet al 2006, Ward et al 2012, Tosca et al 2015), and alsoplay an important role in the variable annual growthrate of CO2 (Nevison et al 2008).
Despite the clear connection between Pacific seasurface temperature (SST) anomalies and fires inequatorial Asia, our understanding of these relation-ships on a global scale and for regions without longobservational records of fires (including much of thetropics and subtropics) is limited by the length of thesatellite record in which active fires can be sensedremotely, roughly 18 years (Giglio et al 2013). More-over, there are reasons to suspect that global fires exhi-bit decadal and multidecadal variability becauseimportant fire drivers have been shown to vary onthese timescales, including precipitation and soil water(Ault et al 2012, 2013, Dai 2012, Chikamotoet al 2015), temperature (Meehl 2015), and ENSOitself, whichmay undergo periods of extreme behavioron a wide range of timescales (Wittenberg 2009, 2015,Emile-Geay et al 2013, McGregor et al 2013, Witten-berg et al 2014, Capotondi et al 2015). While the satel-lite record of fires is too short to fully characterizevariability on decadal timescales, charcoal sedimentrecords, used as a proxy for fire emissions, covermuchlonger time periods but typically with century-aver-aged values that do not provide information onmulti-decadal aspects (e.g. Marlon et al 2008, Daniauet al 2012). An alternate proxy for fire activity, burnscars on tree rings, has been used to suggest a connec-tion between decadal climate oscillations and fires inwestern North America (Kitzberger et al 2007, Tayloret al 2008, Trouet et al 2010). Ice core records of blackcarbon deposition and trace gas tracers for fire activity(e.g. Zennaro et al 2014) have not been utilized for thispurpose to our knowledge, but could provide highertime resolution than charcoal sediments (Baueret al 2013).
The gap in our knowledge of interactions betweenclimate variability and fires on a global scale spanstimescales that are important for near-term predic-tions, and limits our ability to address questions ofdetection and attribution. With the lack of availableobservational data of sufficient length, Earth Systemmodels (ESMs), which account for interactions amongmultiple fire drivers, may be used to provide informa-tion about how fires respond regionally and globally tovariations in climate across timescales. Numericalexperiments with CO2-concentration driven ESMsallow a separation between forced signal (e.g. histor-ical warming or land use and land cover change) andinternal climate variability. In this study we investigatefires in longmodel integrations with preindustrial for-cing, as well as ensemble simulations of the historical
time period using the Community Earth SystemModel (CESM1) (Hurrell et al 2013, Kay et al 2015)and the Geophysical Fluid Dynamics Laboratory(GFDL) ESM2Mb (Dunne et al 2012, Malyshevet al 2015). We aim to characterize the role of internalclimate variability (Baede et al 2001) in driving fires inorder to inform future modeling investigations of fireregime changes, extreme events and trend detection.
Methods
The CESM1 Large Ensemble Community Project(LENS) comprises historical and future simulationswith many ensemble members, with an aim ofimproving our ability to distinguish between climatechange and climate variability. Kay et al (2015) providea detailed description of the LENS simulation proto-col. In this project, the CESM1 land, ocean, atmos-phere and sea-ice components have approximately 1°resolution and it was forced with 1850 solar andradiative forcing and 1850 land cover for the preindus-trial control. We use the last 1800 years of the CESM1control simulation and also examine the 40 membersof the CESM1 historical ensemble from years 1920 to2005. The ensemble members are initialized withslightly perturbed initial conditions to provide asampling of internal climate variability under histor-ical climate forcing. Land cover change in the 20thcentury ensemble is represented by plant functionaltype (PFT) transitions on a yearly basis and followsHurtt et al (2011), adjusted to match the CESM1 landmodel PFT scheme by Lawrence et al (2012).
The CESM1 fire model is based on the Thonickeet al (2001) scheme, which simulates fires on a dailybasis. In this scheme, probability of fire occurrence isparameterized as a function of soil moisture in the top0.5 m of the soil, with separate ‘moisture of extinction’values for woody and herbaceous PFTs above whichfires will not occur. Fire occurrence in this modelrequires dead litter availability above 100 gCm−2 andalso a ground temperature above zero Celsius (Tho-nicke et al 2001). Fire season length and annual areaburned are computed from an empirically-derivedfuel moisture function (Thonicke et al 2001). Carbon(C) emissions from fires are determined by applyingPFT-specific combustion completeness and mortalityfactors to the available biomass within theburned area.
ESM2Mb is based on ESM2M (Dunneet al 2012, 2013) with updated parameter settings forthe land model LM3 (Malyshev et al 2015), approxi-mately 2° horizontal resolution for the atmosphereand land, and roughly 1° horizontal resolution for sea-ice and ocean components. The control run was con-tinued for 6000model years after reaching quasi-equi-librium with solar and radiative forcing representativeof year 1860. This simulation used potential vegetation(i.e. undisturbed by human land use) instead of
2
Environ. Res. Lett. 11 (2016) 125008
preindustrial vegetation cover and includes dynamicvegetation, important distinctions from the CESM1control run. In addition to the preindustrial control,we analyze the corresponding 10-member ensemble ofhistorical forcing simulations running from 1861 to2005 with 20th century land cover changes followingHurtt et al (2011). The ESM2Mb transitions frompotential vegetation in the preindustrial control run to1861 land cover via a 160 year ‘bridge’ experimentwith 1860 radiative forcing and 1700–1860 land coverhistorical reconstructions (Hurtt et al 2011).
The ESM2Mb fire model, described by Shevlia-kova et al (2009), predicts fires as a function of soilmoisture but fires occur annually. Under this scheme,individual months are designated as drought or non-drought, based on the soil moisture deficit. The yearlysum of drought months and the available aboveground fuel are used to modulate a PFT-specific cli-matological fire return-interval factor that yieldsannual fire emissions of C at each grid point (Shevlia-kova et al 2009). In contrast to CESM1, this scheme isbound more to the seasonal cycle of soil moisture andfuel availability and has no direct air temperature-dependence.
The fire models used in this study do not capturesubtle ecosystem shifts that impact fire dynamics, suchas different species make-up within similar PFTs(Kelly et al 2013, Rogers et al 2015), or landscape frag-mentation (Nepstad et al 2006, Chen et al 2013). Inaddition, neither fire model explicitly simulates fireignition from lightning or human activity, which isoften represented in global fire models (e.g. Klosteret al 2010, Li et al 2013, Pfeiffer et al 2013, Yueet al 2014, Le Page et al 2015). Recent studies indicate adiminished role for natural (Bistinas et al 2014) andanthropogenic (Prentice et al 2011) ignitions in globalfire prediction. Lightning, however, may be a betterpredictor of fire on a regional basis (Abatzoglouet al 2016). Moreover, lightning has been shown tovary substantially with tropical Pacific SST anomalies(Sátori et al 2009). We are unable to explore any con-nections between IAV in lightning and fires with thecurrent model setup. Models that include lightningignitions typically use a lightning climatology and are,therefore, also unable to explore these connections.
Our analysis of fire emissions from both ESMs isdone on an annual basis. Fires in some arid regionsrespond to the robustness of the previous wet seasonand associated vegetation growth (Van der Werfet al 2008). We shift the start month of the annualaverages for CESM1 emissions at each grid point toeightmonths prior to themodel climatological peak infire emissions, determined by harmonic analysis, toaccount for the different regional seasonality. Annualwindows for locations with biannual fire cycles werenot shifted. This shifting is applied in the ENSO com-posite analysis only.
We use the GFED4s (Giglio et al 2013), GFASv1(Kaiser et al 2012) and FINN (Wiedinmyer et al 2011)
fire emission inventories (details in supplementarymaterial) for comparison to present-day model fireemission variability for 14 regions (figure S1). In addi-tion, we compiled charcoal sedimentation recordsfrom the Global Charcoal Database (GCD) for 11 sitesin westernNorth America (table S1) (Power et al 2008)as a proxy for past fire emissions in this region (detailsin supplementary material). We note that as a metricof fire activity, C emissions emphasize forest fires,while another commonly used measure, area burned,is dominated by savannah and grassland fires (van derWerf et al 2008).
To compute ENSO indices we use SSTs averagedover the NINO3 region (150W–90W, 5S–5N), as inWittenberg (2009), for which positive values indicateEl Niño-like conditions and negative values indicateLaNiña-like conditions. Both ESMs generate a reliableENSO with ESM2Mb producing a stronger ENSOcompared to CESM1 and the HadISST1 timeseries(Rayner et al 2003) (figure S2). Precipitation tele-connection patterns are similar between the twomod-els for North America and equatorial Asia but showsubstantial differences in both the sign and seasonaltiming of the response inAfrica (figure S3).
We quantify the interdecadal Pacific oscillation(IPO) index as the leading principal component ofPacific basin SSTs after applying a 13 year low-pass fil-ter to the unforced long-control model run (Meehlet al 2009). We split the model timeseries into periodsof 300 years before filtering and computing theempirical orthogonal functions. For CESM1, the firstEOF explains 39% of the variance and exhibits a spa-tial pattern very similar to the IPO associated withobserved SSTs (not shown) suggesting that the modelcaptures this mode of interdecadal variability. ForESM2Mb, the first EOF explains less than 25% of thevariance. Therefore, we do not use the IPO index com-puted from the ESM2Mb in the remainder of thestudy. Hereafter we refer to this index as the PDO, not-ing that the IPO and PDOare highly correlated and theterms are often used interchangeably (Meehl 2015).
The Atlantic Meridional Oscillation index is heredefined as the SST anomalies in the North AtlanticOcean (80W–0W, 0N–60N)minus global SST anoma-lies (60S–60N), following Trenberth and Shea (2006)as recommended by Phillips et al (2014).We also com-pute 10 year and 50 year low-pass filtered AMO indextimeseries, as in Knight et al (2006), to remove varia-bility related to ENSO and the PDO, respectively.
Results
We assess the IAV in fire emissions globally and byregion using the coefficient of variation (CV; quotientof the standard deviation and the mean) (figure 1).Globally, CESM1 and ESM2Mb underpredict the CVin comparison to the GFED4s natural fires. The largestregional biases occur in boreal regions where fires are
3
Environ. Res. Lett. 11 (2016) 125008
more episodic than in the tropics (Clark et al 2015).Precipitation is not underestimated in boreal regions(figure S4), suggesting the bias is caused by poorrepresentation of other fire drivers, such as ignitionand temperature, which are more important in borealzones. Temperature, for example, drives boreal fireseason length and severity (Flannigan et al 2009) andthe depth of the active layer of permafrost, whichregulates area burned and fire severity (e.g.Venevzky 2001, Frolking et al 2011). Central Americanfires increased strongly with the 1997–98 El Niño (vander Werf et al 2004) but here both models showpositive and negative precipitation responses to ENSOin this region (figure S3), dampening any modelresponse in fire emissions. The full GFED4s recordhas generally higher CV than the 2003–2013 record(figure 1), likely due to the influence of the 1997–98 ElNiño. Also, the addition of anthropogenic fires (largelydeforestation and agricultural fires) leads to lower CVin the GFED4s and ostensibly in the GFASv1 andFINNv1.5 as well (figure 1). Both the CESM1 andESM2Mb capture the high variability in equatorialAsia, attributable largely to extreme responses to thedry conditions associated with the 1997–98 El Niño
event (van derWerf et al 2004, 2006, Giglio et al 2013),suggesting that the models demonstrate someresponse to ENSO.
The power spectrum of CESM1 fire emissionsshows substantial variance at periods of 3–7 years,which is significantly different from red noise at a 95%confidence level (figure 2(a)). The variance ofESM2Mb fire emissions is not clearly distinguishablefrom red noise at any frequency, although the signal isstrongest for periods of 3–7 years (figure 2(b)). Indivi-dual regions show statistically significant variance atthese timescales, including temperate North Americain both ESMs, and all Asian regions in CESM1(figures 2(a), (b)). In the CESM1 output, variance isdistinguishable from red noise in southern hemi-sphere Africa and the Middle East on timescales of thePDO (10–30 years), and inCentral America and borealN America on AMO timescales (75–100 years)(figures 2(a), (d)).
The shortness of the GFED4s record of fire emis-sions makes it difficult to interpret its power spectrum(figure 2(c)). There is a suggested connection to ENSOwith higher variance at a period of 5 years but this peri-odicity cannot be confirmed statistically with the small
Figure 1. Interannual variability of global and regional fire emissions represented by the coefficient of variation for the last 18 years ofeach ensemblemember from (a)CESM1 and (b)ESM2Mb compared to the coefficient of variation for theGFED4s natural fires, years1997–2014 (black circles), GFED4s allfires 1997–2014 (orange circles), GFED4s allfires 2003–2013 (red circles), GFASv1 2003–2013(yellow circles), FINNv1.5 2003–2013 (green circles). The distributions of coefficients computed from themodel ensembles are shownas box andwhiskers plots (box:median and interquartile range, whiskers: range).
4
Environ. Res. Lett. 11 (2016) 125008
sample size. GCD timeseries from western NorthAmerican sites exhibit strong variability at longer peri-ods of about 40 years and 70 years (figure 2(c)). How-ever, given our knowledge of variability associatedwith ENSO, it is possible that aliasing of higher fre-quency oscillations is responsible for some of the var-iance. The potential for aliasing is difficult to state withcertainty given that the GCD data are accumulationrates, each encompassing several years, and have dif-ferent original timesteps depending on the site. Theproblematic nature of spectral analysis on the availabledatasets of fire emissions and proxies stands to high-light the gaps in our knowledge of fire/climate interac-tions, especially on decadal timescales.
We assess the variability of fire emissions with theaforementioned climate oscillation indices using com-posite analysis. We group the long control run annualfire emissions by positive or negative index, ENSO,PDO and AMO, and compute the difference in meansbetween the data subsets at each model grid point. Asimple statistical test of difference between means isthen applied to identify locations where the mean fireemissions shift depending on the sign of the oscillationindex. The fire emissions timeseries are standardizedsuch that the differences in mean are plotted as frac-tions of the standard deviation characteristic of eachlocation. We present analysis of fire emissions rela-tionship to the decadal and multi-decadal climateoscillations for CESM only because ESM2Mb does notgenerate a reliable PDO and does not produce anAMO with variability on multidecadal timescales(figure 2(e)).
The spatial pattern of the CESM1 response toENSO (figure 3(a)) compares well to the fire emissionanomalies observed during the 1997–1998 El Niñowith anomalously high emissions in boreal NorthAmerica, the Amazon, southeast Asia, and Australia,and decreased emissions in northern hemisphereAfrica (van der Werf et al 2004)) (figure S5). TheESM2Mb fire emissions exhibit a similar spatial pat-tern of response to ENSO but the signal is weak(figure 3(b)), as suggested by the spectral analysis(figure 2(b)).
The response of CESM1 fire emissions to the PDOindex is weaker than the response to ENSO but followsa similar spatial pattern (figure 3(c)) except in borealNorth America where fire emissions are negativelycorrelated with the PDO index. In this region, theaverage soil moisture increases for warm-phase PDO(not shown), contributing to a decrease in fires duringthis phase.
CESM1 fire emissions also exhibit a statisticallysignificant response to different phases of the AMO inseveral regions (figure 3(d)). The abrupt change in signof the response across the equator in South America,positive southward and negative northward(figure 3(d)), is consistent with analysis of satellite-derived fire counts from Chen et al (2011) who attri-bute the change to shifts in the ITCZ and precipitationclimatology that are associated with the different pha-ses of the AMO. In general, the CESM1 fire emissionsresponse to the AMO is opposite in sign to the PDOresponse (figure 1(c)), which is consistent withdrought and pluvial relationships between the PDOand AMO (e.g. Findell and Delworth 2010). However,
Figure 2.Power spectra offire emissions (black line)with variance required for statistical difference from red noisewith 95%confidence (black dashed line) for (a) theCESM1 long control run, (b) the ESM2Mb long control run and (c) theGFED4.1 s. Thespectrumof charcoal influx data from11westernNorth America sites interpolated to 20 year time resolution are also included in (c).Regional spectra are shown as gray lines for a)CESM1 and (b)ESM2Mbwith frequencies forwhich the spectra are significantlydifferent from red noisewith 95% confidence highlighted by color. Spectra and 95% confidence level curves are plotted for the PDO(brown) andAMO (black) indices for (d)CESM1 and (e)ESM2Mb. There is no PDOplotted for ESM2Mb since the long control rundid not produce a reliable PDOpattern.
5
Environ. Res. Lett. 11 (2016) 125008
Figure 3.Composite analysis showing differences in themeans of standardized anomalies of annual fire emissions for years withpositiveNINO3 indicesminus years with negativeNINO3 indices from (a)CESM1 and (b)ESM2Mb, years with positive PDOminusyears with negative PDO from (c)CESM1, and years with positive low-pass filteredAMO indicesminus years with negative low-passfilteredAMO indices fromd)CESM1. The values of the difference inmeans that are significant at a 95%confidence level (two-tailedtest) are shown in the colorbars as pink lines. Degrees of freedom are computed using an effective sample size that accounts for the lag-1 autocorrelation characteristic to thefire emissions timeseries.
Figure 4.Change in annual fire emissions shown as a timeseries of globalmeans for (a)CESM1 and (c)ESM2Mb, and plotted spatiallyas the difference inmean fire emissions between 1991–2005 and 1920–1934.Hatching indicates grid points for which the difference inmeans is significant with 95%confidence.
6
Environ. Res. Lett. 11 (2016) 125008
several studies have shown that a positive AMO canenhance environmental conditions brought on by apositive phase of the PDO (e.g. Kitzberger et al 2007,Mo et al 2009). Applying a 50 year low-pass filter to theAMO index here reduces the strong tropical responsein fire emissions to AMO (figure S6) implying a con-nection exists between the AMO and PDO in themodel results.
The ESM2Mb ensemble mean fire emissionsincrease by about 400 TgC yr−1 between 1920 and2005 while the CESM1 produces a small decrease(∼25 TgC yr−1) in 10 year mean over the same timeperiod (figures 4(a), (c)). The divergence of the trendsresults from differences in the fire model response toenvironmental factors, such as temperature and pre-cipitation, and differences in the ESM predictions ofthese environmental factors. Efforts to determine 20thcentury trends in fire activity from observations havealso led to mixed results. Mouillot and Field (2005)compiled a detailed historical reconstruction of areaburned by fires, combining satellite retrievals, fieldrecords, and tree-ring data. They conclude that globalfires have increased since 1900, with a mid-centuryminimum in fire activity. However, it is difficult tocompare their dataset with the model results reportedhere which do not account for deforestation fires orhumanfire ignition and suppression.
CESM1 exhibits a small global trend in fire emis-sions but there are major regional increases in the bor-eal northern hemisphere, equatorial Asia andmuch ofthe Amazon (figure 4(b)) that appear to be driven byland cover changes. Parts of southeast Asia underwenta transition from largely primary forest vegetation toherbaceous vegetation during the 20th century (Hurttet al 2011), which reduces the moisture of extinction
for fires in the CESM1 fire scheme and also shrinksfuel availability, leading to reduced fire. In boreal for-ests, increasing temperatures lengthen the fire seasonin CESM1, enhancing fire emissions despite highersoilmoisture.
Historical trends in emissions in ESM2Mb are lar-gely driven by soil moisture changes. Where regionaltrends are statistically significant from 1920 to 2005they are uniformly positive (figure 4(d)). The changesare related to increases in drought months over thistime period, and enhanced in the western UnitedStates, central Asia, and southern South America byco-located increases in fuel availability.
Discussion and conclusions
Anthropogenic impacts on climate and land covercould have lead to positive trends in 20th centuryglobal fire activity (Mieville et al 2010, Lasslop andKloster 2015), negative trends (Marlon et al 2008, Yanget al 2014, Knorr et al 2016), or trends that change sign(Mouillot and Field 2005, Kloster et al 2010, Pechonyand Shindell 2010). Here we are able to show thathistorical trends in fire emissions for CESM1 andESM2Mb, while opposite in sign, are statisticallysignificant at a 95% confidence level with a nullhypothesis of zero trend. This is to say that a 20thcentury anthropogenic signal, including humanimpacts on climate, CO2, and land cover, is detectablein natural fires above the year-to-year noise in fireemissions in these two ESMs. The significance testingused here can be generalized to any size trend andnumber of ensemblemembers (table 1).
The magnitudes of significant trends differ sub-stantially between the CESM1 and ESM2Mb due to
Table 1.Threshold for significance of trends in fire emissions. This is expressed as thedifference in globalfire emissions (TgC yr−1) averaged over a given number of years that isdetectable at a 95% statistical confidence level with the given number ofmodel ensemblemembers. Values are based on the variance of the ESM2Mb andCESM1 long controlsimulations.
ESM2Mb
EnsembleNumber of years
members 3 5 10 15 18 20 50 100
3 542 418 296 246 223 215 141 116
5 247 191 135 112 102 98 65 53
10 134 104 73 61 55 53 35 29
20 86 66 47 39 35 34 22 18
30 68 52 37 31 28 27 18 14
40 48 37 26 22 20 19 13 10
CESM1
3 301 223 157 132 114 108 52 32
5 137 102 71 60 52 49 24 15
10 75 55 39 33 28 27 13 8
20 48 35 25 21 18 17 8 —
30 38 28 20 16 14 14 6 —
40 27 20 14 12 10 10 — —
7
Environ. Res. Lett. 11 (2016) 125008
the dissimilar natural variability in fire emissionsinherent to each model. The model dissimilaritiescould result from biases in fuel load and climate varia-bility in the ESMs (figures S2, S3), and differences inmodel processes such as dynamic vegetation inESM2Mb compared to static vegetation in CESM1. Inmore recent versions of CESM, the Thonicke et al(2001) firemodel has been replaced with that of Li et al(2013), which represents peat and deforestation fires,as well as natural and human ignition and fire suppres-sion, and is able to improvemodel reproduction of firespatial patterns, total emissions, and IAV (Liet al 2013). Small-scale land cover effects on wildfires,such as ecosystem edge effects and local variations insurface hydrology, are still not well represented in glo-bal fire models (Ward and Mahowald 2015). Theseeffects could be especially important in regions ofintense land conversion, including equatorial Asiawhere, as mentioned in the Introduction, deforesta-tion can expose massive amounts of peat and increasethe sensitivity of regional fire emissions to climatevariability (Marlier et al 2015). Greater attention tothese small-scale, often sub-grid-scale, processescould improve our understanding of anthropogenicimpacts onfire variability.
The spatial pattern of the fire emissions responseto ENSO in both CESM1 and ESM2Mb is roughlyconsistent with observational records of fires inCanada (Skinner et al 2006) and Alaska (Hesset al 2001), and studies of tree-ring burn-scar syn-chrony. Greater synchrony, an indicator of increasedfire activity (Falk et al 2011), occurs during the coldphase of ENSO in the southwest and interior west Uni-ted States, and during the warm phase of ENSO in thePacific northwest (Kitzberger et al 2007, Trouetet al 2010). Burn-scar synchrony studies of connec-tions between fire activity and decadal oscillations,such as the PDO, are generally inconclusive due touncertainties in the proxies of climate variability(Kipfmueller et al 2012), but possible connectionswould havemajor implications for decadal fire predic-tion on a global scale. For example, best estimates ofglobalfire emissions, such as theGFED4s, are based ondata collected almost entirely during a negative phaseof the PDO, and there is evidence that a phase changemay have occurred in 2014 (Meehl 2015).
Our results suggest that, with this phase change inthe PDO, southeast Asia, Australia, the northern Ama-zon region, and eastern equatorial Africa will see a shiftto higher fire emissions contributed by natural climatevariability, and the southwestern United States, north-ernMexico, and the southwestern Amazonwill shift tolower fire emissions contributed by natural variability.Increasing confidence in long-term forecasts such asthese will require further progress in our under-standing of the natural variability in fires from addi-tional observations and proxy data, as well asimprovedmodeling of globalfires.
Acknowledgments
We would like to acknowledge assistance from Jenni-fer Marlon for providing data from the GCD. Com-puting resources (ark:/85065/d7wd3xhc) wereprovided by the Climate Simulation Laboratory atNCAR’s Computational and Information SystemsLaboratory, sponsored by the National Science Foun-dation and other agencies. The CESM project issupported by theNational Science Foundation and theOffice of Science (BER) of the US Department ofEnergy. The National Science Foundation sponsorsNCAR. This report was prepared by D Ward underaward NA14OAR4320106 from the National Oceanicand Atmospheric Administration (NOAA), USDepartment of Commerce. The statements, findings,conclusions, and recommendations are those of theauthors and do not necessarily reflect the views ofNOAA, or theUSDepartment of Commerce.
References
Abatzoglou J T, KoldenCA, Balch J K andBradley BA 2016Controls on interannual variability in lightning-caused fireactivity in thewesternUS Environ. Res. Lett. 11 1–11
Ault TR, Cole J E andGeorge S St 2012The amplitude of decadal tomultidecadal variability in precipitation simulated by state-of-the-art climatemodelsGeophys. Res. Lett. 39 L21705
Ault TR, Cole J E,Overpeck J T, PedersonGT,George S St,Otto-Bliesner B,Woodhouse CA andDeser C 2013Thecontinuumof hydroclimate variability inWesternNorthAmerica during the lastmillennium J. Clim. 26 5863–78
Baede APM,Ahlonsou E,Ding Y and Schimel D 2001TheClimateSystem: AnOverviewClimate Change 2001: The Scientific Basised BBolin and S Pollonais (Cambridge: CambridgeUniversity Press)
Bauer S E, BauschA,Nazarenko L, Tsigaridis K, XuB, Edwards R,BisiauxMandMcConnell J 2013Historical and future blackcarbon deposition on the three ice caps: ice coremeasurements andmodel simulations from1850 to 2100J. Geophys. Res.-Atmos. 118 7948–61
Bistinas I, Harrison S P and Prentice JMC2014Causal relationshipsversus emergent patterns in the global controls offirefrequencyBiogeosciences 11 5087–101
Capotondi A et al 2015Understanding ENSOdiversityBull. Am.Meteorol. Soc. 96 921–38
ChenY,MortonDC, AndelaN,Giglio L andRanderson J T 2016Howmuch global burned area can be forecast on seasonaltime scales using sea surface temperatures? Environ. Res. Lett.11 045001
ChenY,MortonDC, Jin Y, Collatz G J, Kasibhatla P S,van derWerf GR,DeFries R S andRanderson J T 2013 Long-term trends and interannual variability of forest, savanna andagriculturalfires in SouthAmericaCarbonManage. 4 617–38
ChenY, Randerson J T,MortonDC,DeFries R S, CollatzG J,Kasibhatla P S, Giglio L, Jin Y andMarlierME2011Forecastingfire season severity in SouthAmerica using seasurface temperature anomalies Science 334 787–91
ChikamotoY, TimmermannA, Stevenson S,DiNezio P andLangford S 2015Decadal predictability of soil water,vegetation, andwildfire frequency overNorthAmericaClim.Dyn. 45 2213–35
ChisholmRA,Wijedasa L S and Swinfield T 2016The need for long-term remedies for Indonesia’s forestfiresConservation Biol.30 5–6
Clark SC,WardD S andMahowaldNM2015The sensitivity ofglobal climate to the episodicity offire aerosol emissionsJ. Geophys. Res. Atmos. 120 11589–607
Dai A 2012The influence of the inter-decadal Pacific oscillation onUSprecipitation during 1923–2010Clim.Dyn. 41 633–46
DaniauA L et al 2012 Predictability of biomass burning in responseto climate changesGlob. Biogeochem. Cycles 26GB4007
Duffy PA,Walsh J E, Graham JM,MannDHandRuppT S 2005Impacts of large-scale atmospheric-ocean variability onAlaskan fire season severity Ecol. Appl. 15 1317–30
Dunne J P et al 2012GFDL’s ESM2 global coupled climate–carbonEarth systemmodels: I. Physical formulation and baselinesimulation characteristics J. Clim. 25 6646–65
Dunne J P et al 2013GFDL’s ESM2 global coupled climate–carbonearth systemmodels: II. Carbon system formulation andbaseline simulation characteristics J. Clim. 26 2247–67
Emile-Geay J, CobbK,MannMandWittenberg AT 2013Estimating central equatorial Pacific SST variability over thepastmillennium: II. Reconstructions and implicationsJ. Clim. 26 2329–52
Findell K L andDelworth TL 2010 J. Clim. 23 485–503FlanniganMD,KrawchukMA, deGrootW J,Wotton BMand
GowmanLM2009 Implications of changing climate forglobal wildlandfire Int. J.Wildland Fire 18 483–507
Frolking S, Talbot J, JonesMC, Treat CC,Kauffman J B,Tuittila E-S andRoulet N 2011 Peatlands in the Earth’s 21stcentury climate systemEnviron. Rev. 19 371–96
Giglio L, Randerson J T and van derWerf GR 2013Analysis of daily,monthly, and annual burned area using the fourth-generation globalfire emissions database (GFED4)J. Geophys. Res.: Biogeosci. 118 317–28
Hess J C, Scott CA,HuffordGL and FlemingMD2001 ElNiño andits impact on fireweather conditions inAlaska Int. J.WildlandFire 10 1–13
Heyerdahl EK,McKenzieD,Daniels LD,Hessl A E, Littell J S andMantuaN J 2008Climate drivers of regionally synchronousfires in the inland northwest (1651–1900) Int. J.Wildland Fire17 40–9
Hurrell JW et al 2013The community Earth systemmodel: aframework for collaborative researchBull. Am.Meteorol. Soc.94 1339–60
Hurtt GC et al 2011Harmonization of land-use scenarios for theperiod 1500–2100: 600 years of global gridded annual land-use transitions, wood harvest, and resulting secondary landsClim. Change 109 117–61
Kaiser JW et al 2012 Biomass burning emissions estimatedwith aglobalfire assimilation systembased on observed fireradiative powerBiogeosciences 9 527–54
Kay J E et al 2015The community Earth systemmodel (CESM) largeensemble project: a community resource for studying climatechange in the presence of internal climate variabilityBull. Am.Meteorol. Soc. 96 1333–49
Kelly R, ChipmanML,Higuera P E, Stefanova I, Brubaker LB andHuF S 2013Recent burning of boreal forests exceeds fireregime limits of the past 10 000 yearsProc. Natl Acad. Sci. 11013055–60
Kipfmueller K F, Larson ER andGeorge S St 2012Does proxyuncertainty affect the relations inferred between the PacificDecadalOscillation andwildfire activity in thewesternUnited States?Geophys. Res. Lett. 39 L04703
Kitzberger T, BrownPM,Heyerdahl EK, SwetnamTWandVeblenTT2007Contingent Pacific–AtlanticOceaninfluence onmulticenturywildfire synchrony over westernNorthAmerica Proc. Natl Acad. Sci. 104 543–8
Kloster S,MahowaldNM,Randerson J T, Thornton PE,Hoffman FM, Levis S, Lawrence P J, Feddema J J,OlesonKWand LawrenceDM2010 Fire dynamics duringthe 20th century simulated by the Community LandModelBiogeosciences 7 1877–902
Knight J R, FollandCK and Scaife AA 2006Climate impacts of theAtlanticmultidecadal oscillationGeophys. Res. Lett. 33L17706
KnorrW, Jiang L andArnethA 2016Climate, CO2 and humanpopulation impacts on global wildfire emissionsBiogeosciences 13 267–82
LasslopG andKloster S 2015 Impact of fuel variability onwildfireemission estimatesAtmos. Environ. 121 93–102
Lawrence P J et al 2012 Simulating the biogeochemical andbiogeophysical impacts of transient land cover change andwood harvest in the community climate systemmodel(CCSM4) from1850 to 2100 J. Clim. 25 3071–95
Le Page Y,MortonD, Bond-Lamberty B, Pereira JMCandHurtt G2015HESFIRE: a global firemodel to explore the role ofanthropogenic andweather driversBiogeosciences 12 887–903
Le Page Y, Pereira JMC, Trigo R, daCamaraC,OomDandMota B2008Globalfire activity patterns (1996–2006) and climaticinfluence: an analysis using theWorld Fire AtlasAtmos.Chem. Phys. 8 1911–24
Li F, Levis S andWardDS 2013Quantifying the role offire in theEarth system: I. Improved global firemodeling in theCommunity Earth SystemModel (CESM1)Biogeosciences 102293–314
Malyshev S, Shevliakova E, Stouffer R J and Pacala SW2015Contrasting local versus regional effects of land-use-change-induced heterogeneity on historical climate: analysis with theGFDLEarth SystemModel J. Clim. 28 5448–69
MarlierME et al 2012 ElNiño and health risks from landscape fireemissions in southeast AsiaNat. Clim. Change 3 131–6
MarlierME,DeFries R, PenningtonD,Nelson E,Ordway EM,Lewis J, Koplitz SN andMickley L J 2014 Future fireemissions associatedwith projected land use change inSumatraGlob. Change Biol. 21 345–62
MarlierME,DeFries R S, KimP S, Koplitz SN, JacobD J,Mickley L J andMyers S S 2015 Fire emissions and regional airquality impacts from fires in oil palm, timber, and loggingconcessions in Indonesia Environ. Res. Lett. 10 085005
Marlon J R, Bartlein P J, Carcaillet C,GavinDG,Harrison S P,Higuera P E, Joos F, PowerM J and Prentice I C 2008Climateand human influences on global biomass burning over thepast twomillenniaNat. Geosci. 1 697–702
McGregor S, TimmermannA, EnglandMH, ElisonTimmOandWittenberg AT 2013 Inferred changes in ElNiño-southernoscillation variance over the past six centuriesClim. Past 92269–84
Meehl GA 2015Decadal climate variability and the early-2000shiatusUSCLIVARVariations edDMenemenlis and JSprintall 13 1–6
Meehl GA,HuA and Santer BD2009TheMid-1970s climate shiftin the pacific and the relative roles of forced versus inherentdecadal variability J. Clim. 22 780–92
Mieville A, Granier C, Liousse C,GuillaumeB,Mouillot F,Lamarque J-F, Gregoire J-M and PetronG 2010 Emissions ofgases and particles frombiomass burning using satellite dataand an historical reconstructionAtmos. Environ. 44 1469–77
MoKC, Schemm J-K E andYoo S-H 2009 Influence of ENSO andtheAtlanticmultidecadal oscillation on drought over theUnited States J. Clim. 22 5962–82
Monks S A, Arnold S R andChipperfieldMP2012 Evidence for ElNiño-southern oscillation (ENSO) influence onArctic COinterannual variability through biomass burning emissionsGeophys. Res. Lett. 39 L14804
Mouillot F and Field CB 2005 Fire history and the global carbonbudget: a 1ox 1ofire history reconstruction for the 20thcenturyGlob. Change Biol. 11 398–420
NaikV,Mauzerall D,Horowitz L, SchwarzkopfMD,RamaswamyV andOppenheimerM2005Net radiativeforcing due to changes in regional emissions of troposphericozone precursors J. Geophys. Res. 110D24306
NepstadD et al 2006 Inhibition of Amazon deforestation and fire byparks and indigenous landsConservation Biol. 20 65–73
NevisonCD,MahowaldNM,Doney SC, Lima ID,van derWerf GR, Randerson J T, BakerD F, Kasibhatla P and
McKinleyGA 2008Contribution of ocean, fossil fuel, landbiosphere, and biomass burning carbon fluxes to seasonaland interannual variability in atmospheric CO2 J. Geophys.Res. 113G01010
PechonyO and Shindell DT 2010Driving forces of global wildfiresover the pastmillennium and the forthcoming century Proc.Natl Acad. Sci. 107 19167–70
PfeifferM, Spessa A andKaplan JO 2013Amodel for globalbiomass burning in preindustrial time: LPJ-LMfire (v1.0)Geosci.Model Dev. 6 643–85
Phillips A S,Deser C and Fasullo J 2014 Evaluatingmodes ofvariability in climatemodels Eos 95 453–5
PowerM J et al 2008Changes infire regimes since the Last GlacialMaximum: an assessment based on a global synthesis andanalysis of charcoal dataClim. Dyn. 30 887–907
Prentice I C, KelleyD I, Foster PN, Friedlingstein P,Harrison S P andBartlein P J 2011Modelingfire and theterrestrial carbon balanceGlob. Biogeochem. Cycles 25GB3005
Randerson J T et al 2006The impact of boreal forestfire on climatewarming Science 314 1130–2
RaynerNA, ParkerDE,Horton EB, FollandCK,Alexander LV,Rowell D P, Kent EC andKaplan A 2003Global analyses ofsea surface temperature, sea ice, and nightmarine airtemperature since the late nineteenth century J. Geophys. Res.108 4407
Rogers BM, Soja A J, GouldenML andRanderson J T 2015Influence of tree species on continental differences in borealfires and climate feedbacksNat. Geosci. 8 228–34
Sátori G,Williams E and Lemperger I 2009Variability of globallightning activity on the ENSO time scaleAtmos. Res. 91500–7
Shevliakova E, Pacala SW,Malyshev S,Hurtt GC,Milly PCD,Caspersen J P, Sentman LT, Fisk J P,WirthC andCrevoisier C 2009Carbon cycling under 300 years of land usechange: importance of the secondary vegetation sinkGlob.Biogeochem. Cycles 23 GB2022
SkinnerWR, ShabbarA, FlanniganMDand LoganK 2006 Largeforestfires in Canada and the relationship to global seasurface temperatures J. Geophys. Res. 111D14106
Taylor AH, Trouet V and Skinner CN2008Climatic influences onfire regimes inmontane forests of the southernCascades,California, USA Int. J.Wildland Fire 17 60–71
ThonickeK, Venesky S, Sitch S andCramerW2001The role offiredisturbance for global vegetation dynamics: coupling fire intoa dynamic global vegetationmodelGlob. Ecol. Biogeogr. 10661–7
ToscaMG,DinerD J, GarayM J andKalashnikovaOV2015Human-caused fires limit convection in tropical Africa: firsttemporal observations and attributionGeophys. Res. Lett. 426492–501
TrenberthKE and SheaD J 2006Atlantic hurricanes and naturalvariability in 2005Geophys. Res. Lett. 33 L12704
Trouet V, Taylor AH,Wahl ER, Skinner CN and Stephens S L 2010Fire-climate interactions in theAmericanWest since 1400CEGeophys. Res. Lett. 37 L04702
van derWerf GR, Randerson J T, CollatzG J, Giglio L,Kasibhatla P S, ArellanoA F,Olsen SC andKasischke E S2004Continental-scale partitioning offire emissions duringthe 1997–2001 ElNiño/LaNiña period Science 303 73–6
van derWerf GR, Randerson J T,Giglio L, Collatz G J,Kasibhatla P S andArellanoA F 2006 Interannual variabilityin global biomass burning emissions from1997 to 2004Atmos. Chem. Phys. 6 3423–41
van derWerf GR, Randerson J T,Giglio L, GobronN andDolmanA J 2008Climate controls on the variability of fires inthe tropics and subtropicsGlob. Biogeochem. Cycles 22GB3028
Venevsky S 2001 Broad-scale vegetation dynamics in north-easternEurasia—observations and simulations PhDThesisUniversität für Bodenkultur, Vienna 150 pp
Voulgarakis A,MarlierME, Faluvegi G, Shindell DT,Tsigaridis K andMangeon S 2015 Interannual variability oftropospheric trace gases and aerosols: the role of biomassburning emissions J. Geophys. Res. 120 7157–73
WardDS, Kloster S,MahowaldNM,Rogers BM,Randerson J T andHess PG 2012The changing radiativeforcing offires: globalmodel estimates for past, present andfutureAtmos. Chem. Phys. 12 10857–86
WardDS andMahowaldNM2015 Local sources of global climateforcing fromdifferent categories of land use activities EarthSyst. Dyn. 6 175–94
Wiedinmyer C, Akagi SK, YokelsonR J, Emmons LK, Al-Saadi J A,Orlando J J and Soja A J 2011The fire inventory fromNCAR(FINN): a high resolution globalmodel to estimate theemissions fromopen burningGeosci.Model Dev. 4 625–41
Wittenberg AT 2009Are historical records sufficient to constrainENSO simulations?Geophys. Res. Lett. 36 L12702
Wittenberg AT 2015 Low-frequency variations of ENSOUSCLIVARVariations edAKumar andYXue 13 26–31
Wittenberg AT, Rosati A, Delworth T L, Vecchi GA andZeng F2014 ENSOmodulation: is it decadally predictable? J. Clim.27 2667–81
Yang J, TianH, TaoB, RenW,Kush J, Liu Y andWangY 2014Spatial and temporal patterns of global burned area inresponse to anthropogenic and environmental factors:reconstructing globalfire history for the 20th and early 21stcenturies J. Geophys. Res. Biogeosci. 119 249–63
YueC et al 2014Modelling the role offires in the terrestrial carbonbalance by incorporating SPITFIRE into the global vegetationmodelORCHIDEE: I. Simulating historical global burnedarea andfire regimesGeosci.Model Dev. 7 2747–67
Zennaro P et al 2014 Fire in ice: twomillennia of boreal forestfirehistory from theGreenlandNEEM ice coreClim. Past 101905–24
