Increasing potential of biomass burning over Sumatra, Indonesia induced by anthropogenic tropical warming

Uncontrolled biomass burning in Indonesia during drought periods damages the landscape, degrades regional air quality, and acts as a disproportionately large source of greenhouse gas emissions. The expansion of forest fires is mostly observed in October in Sumatra favored by persistent droughts during the dry season from June to November. The contribution of anthropogenic warming to the probability of severe droughts is not yet clear. Here, we show evidence that past events in Sumatra were exacerbated by anthropogenic warming and that they will become more frequent under a future emissions scenario. By conducting two sets of atmospheric general circulation model ensemble experiments driven by observed sea surface temperature for 1960–2011, one with and one without an anthropogenic warming component, we found that a recent weakening of the Walker circulation associated with tropical ocean warming increased the probability of severe droughts in Sumatra, despite increasing tropical-mean precipitation. A future increase in the frequency of droughts is then suggested from our analyses of the Coupled Model Intercomparison Project Phase 5 model ensembles. Increasing precipitation to the north of the equator accompanies drier conditions over Indonesia, amplified by enhanced ocean surface warming in the central equatorial Pacific. The resultant precipitation decrease leads to a ∼25% increase in severe drought events from 1951–2000 to 2001–2050. Our results therefore indicate the global warming impact to a potential of wide-spreading forest fires over Indonesia, which requires mitigation policy for disaster prevention.


Introduction
Wide-spreading biomass burning damage landscape and air quality on regional scale, by changing forest to peatland and by inducing haze (Fearnside 1997, Emmanuel 2000, Chan Environmental Research Letters Environ. Res. Lett. 9 (2014 104010 (7pp) doi:10.1088/1748-9326/9/10/104010 Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. et al 2001, Aiken 2004). Releasing huge amounts of carbon dioxide (CO 2 ), carbon monoxide (CO), methane (CH 4 ), nitric oxide and nitrogen dioxide (NO x ) and particulates by combustion and resultant deforestation (Levine 2001) act as a source of greenhouse warming (Page et al 2002, van der Werf et al 2008, which gives rise to a global-scale socio-economic problem (Emmanuel 2000, Aiken 2004, so that monitoring of fire hot spots is operated by satellites for a recent few years (Tansey et al 2008). Biomass burning in Indonesia is mainly associated with deforestation, agricultural expansion and transmigration from heavily populated areas (Siegert et al 2001, Fearnside 1997, Wooster et al 2012. Burning expands uncontrollably and into underground peat deposits when extreme droughts occur during the main dry season from June to November (Field and Shen 2008, Field et al 2009, Wooster et al 2012. Therefore, attribution of past drought and fires to anthropogenic climate change is crucial for quantifying carbon-climate feedback through change in fire frequency. Present work examines the extent to which severe fire events in the recent decades have been driven by anthropogenic climate change and how their occurrence is likely to change in the future. For this purpose, we analyze observed data, two sets of atmospheric general circulation model (AGCM) ensemble experiments driven by observed sea surface temperature (SST) for 1960-2011, one with and one without an anthropogenic warming component, and the results of multimodel ensembles (MMEs) obtained from the Coupled Model Intercomparison Project Phase 5 (CMIP5) (Taylor et al 2012).

Data, model and methodology
We focus on the island of Sumatra. Severe fires have occurred in Sumatra during drought years since at least the 1960 s (Field et al 2009), and it was identified recently analysis of high-resolution satellite data as having uniquely high rates of deforestation during 2000-2012 (Hansen et al 2013). We examined the southern half of the island (0.5°N-4.5°S, 100°E -106°E) where the fire problem is most acute, and which is part of Indonesia's primary, coherent rainfall region (Aldrian and Susanto 2003). To determine the drought conditions under which severe haze events occur, we used visual extinction coefficient (B ext ) calculated from airport visibility reports since 1960 (source: www.ncdc.noaa.gov/data-access/ land-based-station-data/land-based-datasets), which provides the longest quantitative biomass burning record available in Indonesia (Field et al 2004, Field et al 2009. The records where the visibility could have been affected by fog or precipitation, and not smoke haze, were excluded (Husar et al 2000, Field et al 2009. We performed segment-piecewise regression between monthly B ext and accumulated precipitation averaged over the southern part of Sumatra for 1960-2011 and defined α at the breakpoint value of the piecewise regression at which the determination of regression coefficients is maximized (Field et al 2009). The observed precipitation was derived from the Global Historical Climatology Network (GHCN).
To examine the contribution of anthropogenic warming to the past drought events, we performed the control (ALL) and no-anthropogenic warming (NAT) runs for 63 years from 1949 to 2011 and used monthly outputs after 1960. They are Atmospheric Model Intercomparison Project (AMIP)-type experiments using the Model of Interdisciplinary Research and Climate version 5 (MIROC5) (Watanabe et al 2010). ALL is driven by SST and sea ice data derived from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) version 3 (Rayner et al 2003), and historical anthropogenic (such as greenhouse gasses) and natural forcing factors (the 2006-2011 period runs were under Representative Concentration Pathways (RCP) 4.5), for 1949-2011. For the NAT run, we fixed anthropogenic forcing at 1850 conditions, and removed time-varying anthropogenic signals in SST and sea ice, which were estimated in advance using two ten-member ensembles of the coupled general circulation model (CGCM) attribution experiments with or without anthropogenic forcing. Further details of the procedure are described in Shiogama et al (2013).
The future increase in the frequency of droughts were analysed from the monthly outputs of the Coupled Model Intercomparison Project Phase 5 (CMIP5) historical and RCP 8.5 experiments until 2100 from nine CGCMs; CanESM2, CCSM4, CNRM-CM5-LR, HadGEM2-ES, IPSL-CM5A-LR, MIROC5, MPI-ESM-LR, MPI-ESM-Mr and MRI-CGCM3 (see http://cmip-pcmdi.llnl.gov/cmip5/availability.html for more detail), which were chosen based on the availability of the associated AMIP experiments. The seasonal cycles of precipitation are highly correlated between the historical run and AMIP sharing the same atmosphere model (Table S1), so we used AMIP as references to the CGCM simulations.

Past occurrence of severe droughts and forest fires over Sumatra
Severe fire events occur conditionally with strong droughts, as presented by a scatter diagram between B ext and accumulated precipitation (figure 1(a)). Over southern Sumatra, fivemonth accumulated precipitation (P accum ) from observations has a clear nonlinear relationship with B ext . B ext increases rapidly when P accum is less than 694 mm, which defines a threshold for strong droughts, α (red dashed line in figure 1(a)). The time series of P accum exhibits a distinct seasonal cycle as well as interannual variability, but drought events defined by α coincide with the occurrence of fires (red curve in figure 1(b)). Physically, this threshold corresponds to peat fuels reaching a moisture content low enough to support combustion. Because peat fires cannot realistically be extinguished with the available fire-fighting resources, burning stops only with the return of the monsoon and the rise of the water table Shen 2008, van der Werf et al 2008).
We first determine the relationship between observed B ext and P accum for a simulated climate, ALL run. ALL run simulates the nonlinear dependence of B ext on P accum , albeit with a climatological mean bias in precipitation (blue dots in figure 1(a)). Time evolution of the simulated P accum is highly correlated with observations (r = 0.79) and captures the severe drought events during 1978-2008 (blue curve in figure 1(b)). However, the model fails to reproduce droughts before 1978 and after 2009, probably due to uncertainty in the forcing data before 1970 s (Kent et al 2007) and after 2009, with the latter taken from the Representative Concentration Pathways (RCP) 4.5 (Shiogama et al 2013).
The model's ability to simulate the observed P accum is reliable only to the extent that it is for the right physical mechanisms. It has been shown that large-scale climate variability caused rainfall anomalies and hence influenced the occurrence of fire episodes over Indonesia (Field et al 2009, Wooster et al 2012. The two dominant large-scale influences over Indonesian precipitation are El Niño (EN) events (Field et al 2009, Wooster et al 2012, which shift the terminal end of the Walker circulation eastward, and a positive phase of the Indian Ocean Dipole (pIOD) (Field et al 2009), which weakens the low-level equatorial westerlies over the tropical Indian Ocean, leading to a reduction in precipitation over Indonesia (Saji et al 1999). A plot of the SST-based indices for the El Niño-Southern Oscillation and IOD from the model confirms that most severe drought events are observed when EN and pIOD co-occur (figure 2(a)). Furthermore, composite anomalies in SST, P accum and low-level winds associated with the drought events show that positive SST anomalies in the central-eastern equatorial Pacific and in the western Indian Ocean, indicative of EN and pIOD, respectively, work to pull moist air out of the equatorial western Pacific and favour dry conditions over Indonesia (figure 2(b)). While EN and pIOD are primarily independent phenomena, as represented by a weak correlation in figure 2(a) (r = 0.26), it is their cooccurrence which leads to the most severe and extended droughts (figure S1).
The ALL AGCM ensemble accurately captures the influence of EN and pIOD on P accum over the western Pacific and hence the underlying mechanism behind anomalously dry years ( figure 2(d)). The scatterplot using ten ensemble members indicates that most of the severe drought events occur when EN and pIOD coincide (figure 2(c)). More than half of the members (red dots in figure 2(c)) consistently simulate that severe droughts mostly happened when EN and pIOD occurred. 20% of the ensemble members showed severe droughts non-EN and non-pIOD periods (yellow dots in figure 2(c)), illustrating that the occurrence of drought in Sumatra is inherently probabilistic but systematically conditioned by EN and pIOD.

The contribution of past anthropogenic warming in severe drought in Sumatra
To elucidate the possible greenhouse warming contribution to drought frequency in Sumatra, we constructed another set of the AGCM runs, called NAT (no-anthropogenic warming) run. The anthropogenic warming component in SST has a pattern similar to future changes (Stocker et al 2013), but with smaller magnitude, having maxima in the eastern Pacific and western Indian Oceans ( figure S2).
The NAT run shows far fewer severe droughts over Sumatra, even during EN-pIOD periods (figure 2(e)), suggesting that Sumatra would have been wetter in recent decades without anthropogenic warming. The probability of severe droughts, as determined by P accum less than α, was 3.11% during the 52 years in the ALL run, but it reduced to 0.23% in 2 1 0 1 9 6 0 1 9 6 3 1 9 6 6 1 9 6 9 1 9 7 2 1 9 7 5 1 9 7 8 1 9 8 1 1 9 8 4 1 9 8 7 1 9 9 0 1 9 9 3 1 9 9 6 1 9 9 9 2 0 0 2 2 0 0 , which measures forest fires and five-month accumulated precipitation (P accum ) derived from Global Historical Climatology Network (GHCN) station data (red) and ten-member ALL runs (blue). The forest fires with large B ext conditionally occur when P accum less than a threshold, α, which is determined using the segment-piecewise regression indicated by solid lines. The shading denotes the range of α in the atmospheric general circulation model (AGCM). (b) Time series of observed (red) and simulated (blue) P accum , together with B ext (black). The ensemble-mean P accum is shown for the AGCM, with the spread shown by shading. Severe drought events with P accum < α are indicated by dots.
the NAT run ( figure S3). The drying of Sumatra is apparently inconsistent with the tropical warming and moistening due to higher SST in the ALL run, but can be explained by referring to the circulation change. The difference in P accum and lowertropospheric winds between the two ensembles in October (ALL minus NAT) shows westerly and easterly anomalies over the western Pacific and the Indian Ocean, respectively, which are responsible for the P accum decrease over Sumatra ( figure 2(f)). This indicates a weakening of the mean east-west circulation for 1960-2011, which is consistent with the recent slowdown of the Walker circulation (Tokinaga et al 2012). The reduced precipitation over Indonesia seems to be more associated with the enhanced precipitation around the Philippines.

Future projection of drying and increasing drought frequency over Sumatra
Future fire activity in Sumatra will depend on future land use and drought occurrence. We quantified the latter by analysing the output of the Coupled Model Intercomparison Project Phase 5 (CMIP5) (Taylor et al 2012) RCP8.5 experiments conducted using nine coupled general circulation models (CGCMs). Since α cannot be obtained from coupled models that have their own interannual variability, it was instead calculated from the concomitant AMIP experiments for 1979-2008. In each AMIP, α was calculated in a similar way to figure 1(a) and was used to determine the threshold for drought in the companion CGCM experiments. Figure 3(a) shows the change in P accum in Sumatra, summed for every 50 years from 1901 to 2100. The base period is 1951-2000. The multi-model ensemble (MME) mean (indicated by the thick curve) shows a decrease in P accum toward the end of this century. The long-term decrease in P accum is consistent with the increase in the frequency of severe drought events in Sumatra ( figure 3(b)). The spread across models is large during the latter half of the 21st century, which may arise from uncertainty in the future changes in EN and pIOD (Collins et al 2010, Cai et al 2013 and from different changes in the SST gradient in the east and west of the tropical Pacific Ocean. However, majority of multi CGCMs (six out of nine) indicate increasing drought frequency during the first half of the century, a total increase of about 25%. The drying tendency as represented by decrease in P accum can be more robust if we assume no change in the property of EN and pIOD (figure S4, further described later). The increasing number of drought events is also seen over most of the Indonesian islands (figure S5).
Fire episodes typically reach their peak in October in Sumatra (Field et al 2009), following lower-than normal rainfall beginning in June. We therefore compare MME mean P accum in October between 1951-2000 and 2001-2050 when the model spread is narrower compared to 2051-2100 ( figure 4(a)). Since the degree of tropical ocean warming depends on the model used, change in P accum has been scaled by the tropical-mean SST change in each model before taking the MME mean. The result shows drier conditions over southwestern Indonesia in 2001-2050, and wetter conditions over the western Pacific and northwestern Indian Oceans. P accum is reduced by 48.5 mm K −1 over Sumatra and by 6.1 mm K −1 over southern Kalimantan (Field et al 2009), an adjacent fire-prone island. The lesser P accum reduction over Kalimantan is due partly to wetter condition over the western Pacific (figure 4(a)), giving no robust increase in the drought frequency over this region.
There are two mechanisms that affect the tropical precipitation change due to global warming: wet-get-wetter (precipitation increases in regions that are already rainy) (Held andSoden 2006, Huang et al 2013) and warmer-getwetter (precipitation increases where the rise in SST exceeds the tropical-mean surface warming) (Huang et al 2013). To isolate these two effects, we analysed nine AGCM experiments in CMIP5, with configurations similar to Atmospheric Model Intercomparison Project (AMIP), but were additionally forced either by a spatially uniform SST increase of 4 K (AMIP + 4 K) or by a SST change derived from CMIP3 MME quadruple CO 2 simulation (AMIP + Pattern). Figures 4(b), (c) show the change in 30-year mean P accum in AMIP + 4 K and AMIP + Pattern from AMIP, respectively. When SST is increased uniformly, the pattern of P accum change can be interpreted in terms of the wet-get-wetter mechanism. Namely, precipitation increases over the Asian monsoon region to the north of Indonesia where mean precipitation is rich in the current climate (contours in figure 4(b)), resulting in a slight decrease in P accum over Indonesia (P accum is reduced by 15.4 mm K −1 over Sumatra). When the pattern change of SST is added, the P accum decrease is amplified over Indonesia, accompanied by the P accum increase over the central equatorial Pacific anchored by the maximal SST rise there (Huang et al 2013) (figure 4(c)). This is consistent with the warmer-get-wetter mechanism and decreases P accum by 75.0 mm K −1 over Sumatra. Our result thus indicates that both mechanisms are responsible for the north-south contrast of precipitation change over the maritime continent, leading to a robust decrease of P accum over Indonesia in the CMIP5 MME. When we compare the P accum decrease over Sumatra for 1951-2000 between RCP8.5 and AMIP + Pattern, the latter shows higher confidence of drying (figure S4), indicating that the increasing drought frequency would be more robust than we see in figure 3(b) if we eliminate uncertainty for the future change in EN (Cai et al 2014) and IOD behaviour. (a) Change in mean P accum in October calculated for every 50-year window, relative to 1951-2000, using nine historical and Representative Concentration Pathways (RCP) 8.5 runs. (b) Change in the frequency of drought events defined by the normalized number of months that show P accum less than α for every 50-year window. The thick curves represent the multimodel ensemble (MME) mean, whilst the spread across models is indicated by shading. Estimates of 50% and 75% ranges for three periods of 1901-1950, 2001-2050 and 2051-2100 are shown by bars.

Summary
Biomass burning in Indonesia occurring mostly due to human shifting cultivation and cause haze disasters influencing the country and surroundings (Fearnside 1997, Aiken 2004. The expansion of forest fires is mostly observed in October in Sumatra favored by persistent droughts during dry season from June to November (Field et al 2009). Contribution of anthropogenic warming to the past and future severe droughts is of vital importance, but it is yet unclear.
We present that a recent weakening of the Walker circulation associated with tropical ocean warming increased the probability of severe droughts, despite increasing tropicalmean precipitation. Present work shows that anthropogenic warming has contributed to the increasing number of drought events over Sumatra for the past 50 years and that it will likely be amplified by 2050. Precipitation increases to the north of the equator coinciding with a drier condition over Indonesia. Enhanced warming in the central equatorial Pacific amplifies this north-south contrast of the precipitation change resulting in the robust increase in severe drought events by about 25% from 1951-2000 to 2001-2050. Our results therefore indicate that Indonesia will experience drier condition during the boreal summer in the future as the effect of the global warming and impact to a potential of wide-spreading forest fires over Indonesia. Without dramatic reforms in land use and large caveats, this suggests that there will be more severe fires in Sumatra, which themselves serve as an additional source of CO 2 , NO x and CO emissions to the atmosphere (Levine 2001, Page et al 2002.