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

Smoke haze, caused by vegetation and peat fires in Southeast Asia, is of major concern because of its adverse impact on regional air quality. We apply two different methods (a chemical transport model and a Lagrangian atmospheric transport model) to identify the locations of fires contributing to the increased mass concentration of particulate matter with diameters less than 2.5 μm (PM2.5) in Singapore over the period 2004–09. We find that fires in southern Sumatra account for the greatest percentage of the total fire enhancement to PM2.5 concentrations in Singapore (42–62%), with fires in central Sumatra and Kalimantan contributing 21–35% and 14–15%, respectively. Furthermore, we find that fires in these regions also increase PM2.5 concentrations in other major cities across Southeast Asia. Our results suggest that acting to reduce fires in southern and central Sumatra (specifically in the eastern parts of the provinces of Jambi, South Sumatra, Lampung and Riau) and southwest Kalimantan (the southern extent of the provinces of West, Central and South Kalimantan) would have the greatest benefit to particulate air quality in Singapore and more widely across Southeast Asia.


Introduction
Vegetation and peat fires occur across Southeast Asia due to a combination of El Niño-induced droughts and anthropogenic land-use change (van der Werf et al 2008). These fires result in deforestation and forest degradation (Langner et al 2007, Carlson et al 2012, and release large quantities of CO 2 (Page et al 2002) and other atmospheric pollutants, degrading regional air quality and negatively impacting human health (Emmanuel 2000, Kunii et al 2002, Johnston et al 2012. The mass concentration of particulate matter (PM) with diameters less than 2.5 μm (PM 2.5 ) is associated with adverse health effects and increased mortality (e.g., Ostro et al 1999).
Most previous attention has been on health impacts of PM 2.5 from fossil fuel combustion (e.g., Lim et al 2012), but particulates from vegetation and peat fires are also thought to be harmful to human health (Frankenberg et al 2005, Emmanuel 2000, Johnston et al 2012, Pavagadhi et al 2013. During large fire years, particulate emissions from fires across Southeast Asia have been estimated to cause 10 800 premature deaths in the region each year . Particulate emissions from vegetation and peat fires result in substantial concentrations of PM 2.5 across large regions of Southeast Asia, including the Indochina peninsula (See et al 2006, Gautam et al 2013, Huang et al 2013 and southern China (Fu et al 2012). This study focuses on Singapore, which regularly experiences substantial PM 2.5 concentrations resulting from a combination of local emissions from industry and transportation sectors as well as long-range transport of smoke from fires (Balasubramanian et al 2003, See et al 2006. Observed 24 hr average PM 2.5 concentrations in Singapore vary from ∼10 μg m −3 on clean days to between ∼40 and ∼300 μg m −3 on polluted days (Balasubramanian et al 2003, See et al 2006. Atmospheric back trajectory methods have been used to show that both increased PM 2.5 concentrations (Balasubramanian et al 2003 and poor visibility (See et al 2006) in Singapore occur when air masses pass over regions with fires. Elevated PM 2.5 concentrations in Singapore are coincident with elevated concentrations of levoglucosan, a tracer for biomass burning , Engling et al 2014, providing further evidence of a large contribution from fires.
Vegetation and peat fires in Southeast Asia rarely occurred before human influence (Goldammer 2006, Schultz et al 2008 and the majority of present day fires in the region are intentionally lit. Fires are used to clear forests to make way for agriculture (Langner et al 2007, Carlson et al 2012 as well as an agricultural management tool. Extensive drainage has also made peatlands susceptible to fires (van der Werf et al 2008). In Sumatra, large fires have occurred since at least 1960 whereas the first large fires in Kalimantan occurred in the 1980s, reflecting the different historical pattern of land-use change across the two regions (Field et al 2009). Future patterns of land-use change will play an important role in determining the extent of future fires across the region (Carlson et al 2012).
Regional scale air pollution and the transport of smoke from vegetation fires across national boundaries have resulted in substantial regulatory interest and attempts to manage and control these fires. For example, the Association of Southeast Asian Nations Agreement on Transboundary Haze Pollution commits the member countries to tackle transboundary haze pollution caused by vegetation and peat fires. Understanding the varying contribution of fires across different regions to particulate pollution is critical to future efforts to monitor and reduce fire.
In this study, we explore the contribution of vegetation and peat fires in different regions across Southeast Asia to PM 2.5 concentrations in Singapore over a six-year period (2004-09). Our aim is to provide quantitative information on the locations of fires that contribute most to PM 2.5 concentrations in Singapore and other large cities in Southeast Asia. This is the first study to quantify the regional contribution of fires to PM 2.5 in Southeast Asia over a multi-year period. Whilst the location of fires is well known, the impact of fires on air quality depends both on fire location combined with subsequent atmospheric transport of the smoke. Identifying the locations of fires causing the greatest air quality degradation will help prioritize action to combat fires and formulate sustainable land-use policy.

Methods
We applied two different methods to identify the regions that make the largest contribution to PM 2.5 concentrations in Singapore.
Firstly, we applied a 3D global aerosol model, GLOMAP (Mann et al 2010) (described in section S1 of the supplementary material), which simulates black carbon (BC), particulate organic matter, sulfate, sea spray and dust with a horizontal resolution of 2.8°by 2.8°. GLOMAP is driven by analyses from the European Centre for Medium-Range Weather Forecasts, meaning our simulations are consistent with the large-scale atmospheric flow from the assimilated observations. The model includes detailed aerosol processes, simulating the evolution of aerosol particles from emission/ production to removal from the atmosphere. Our previous work has demonstrated the ability of GLOMAP to capture observed aerosol concentrations in diverse atmospheric environments (e.g. Spracklen et al 2007, 2010, 2011a, 2011b, Mann et al 2010. For the analysis, we performed a series of simulations over the period January 2004-December 2009 in which we switched off emissions from fires occurring in specific regions (defined in figure 1). To evaluate the model against observations (section 3.1), we performed two simulations over the period January 2000-December 2011 (one control simulation including all global fire emissions and one simulation where no fire emissions were included).
Secondly, we applied a Lagrangian atmospheric transport model, ROTRAJ (Methven et al 2003) (described in section S2), to calculate the history of air masses arriving in Singapore (figure S3). We calculated the cumulative exposure of air masses to fire emissions of carbon monoxide (CO) over five days of atmospheric transport to Singapore. These five-day atmospheric back trajectories were calculated every six hours (0, 6, 12, 18 h UTC) for the period of January 2004-December 2007. The results from the back-trajectory analysis were aggregated for monthly or annual means and averaged over the eight regions shown in figure 1. The resolution of the ROTRAJ model is determined by the resolution of wind fields (1.0125°) used to drive the transport of the air masses (section S2) and the emission database (0.5°× 0.5°), both of which have a higher spatial resolution than GLOMAP. Thus, results from the ROTRAJ model will complement those from GLOMAP, which contains more detailed aerosol processes but coarser spatial resolution.
In both methods we used fire emissions from the Global Fire Emissions Database (GFED3; van der Werf et al 2010) (described in section S3). For use in the global model, the emissions are split by region (figure 1) on a 0.5°by 0.5°grid (the resolution of the GFED3 emissions) and then degraded to the resolution of the global model (2.8°by 2.8°). Figure 2(a) shows the GFED3 fire emissions flux of BC over Southeast Asia. The largest BC emission fluxes over this region are from fires located in central and southern Sumatra and southwest Kalimantan. Figure 2(b) shows the dominant fire type in each grid cell, indicating that these large emission fluxes mainly arise from deforestation and peat fires, which dominate emissions from fires over Southeast Asia (van der Werf et al 2010). Strong emission fluxes of BC also arise from deforestation fires in western Malaysian Borneo, deforestation and agricultural fires in Cambodia, Laos and Myanmar, and savannah-type fires in northern Australia. Figure 3 shows a Taylor Diagram (Taylor 2001) summarizing the evaluation of monthly-mean PM and aerosol optical depth (AOD) simulated by GLOMAP against observations. The locations of in-situ observations of PM 2.5 (Singapore) and PM 10 (Petaling Jaya, Danum Valley and Bukit Kototabang) and the 13 AERONET stations are shown in figure 1. The length of observation period varies by site (for PM, mean = 3 years; AOD, mean = 4.6 years) and is detailed in table S1. In a Taylor diagram, the agreement between model and observations is quantified in terms of their correlation (R), their centred root mean square error (RMSE) and the ratio of their standard deviations, called the normalized standard deviation (NSD). The observations are represented by a point on the horizontal axis at unit distance from the origin; points closer to this represent better agreement between modelled and observed values. The positive correlation between the model (with fires) and observations shows that the model framework is able to capture the seasonal and inter-annual variability in AOD and PM that is driven by variability in fire emissions combined with the vertical and horizontal transport of emissions from their source and the microphysical processing of the aerosol particles.

Evaluation of modelled aerosol over Southeast Asia
Particulate emissions from fires are uncertain, so we compared simulated aerosol with GFED3 emissions against a simulation using emissions from the National Centre for Atmospheric Research Fire Inventory (FINNv1; Wiedinmyer et al 2011) (described in section S3). At the eight AERONET stations in northern Thailand and Vietnam, where agricultural fires dominate emissions (figure 2(b)), the model with FINNv1 emissions better captures AOD (R = 0.80-0.92, NSD = 0.62-0.98) compared to GFED3 (R = 0.29-0.85, NSD = 0.44-0.74). Improved agreement with FINNv1 emissions at these sites is likely because emissions from small fires, which dominate emissions in agriculture fire regions, are not well represented in GFED3 . At the five other AERONET stations, located in regions mainly influenced by deforestation and peat fire emissions (figure 2(b)), model performance with GFED3 (R = 0.33-0.59, NSD = 0.54-1.15) is more comparable to that  table S1). The black dots show 15 major cities (with populations greater than 1 million) selected for analysis in section 3.4.
with FINNv1 (R = 0.45-0.59, NSD = 0.51-1.28). Additionally, when evaluated against PM observations located in regions dominated by deforestation and peat fire emissions, the model with GFED3 emissions (R = 0.77-0.89, NSD = 0.23-0.58) is comparable to the model with FINNv1 emissions (R = 0.71-0.89; NSD = 0.39-0.57). We use GFED3 fire emissions for the rest of this analysis as this dataset is available for 2000-11 (FINNv1 is only available for 2002-12) and we have shown that the two datasets are comparable in regions dominated by deforestation and peat fires which are of most interest to this study.
GLOMAP (with GFED3 emissions) underpredicts observed annual mean PM concentrations in both urban (Singapore, normalized mean bias (NMB) = −64%; Petaling Jaya, NMB = −77%) and rural locations (mean NMB = −56%). GLOMAP is also negatively biased against observed AOD (mean NMB = −29%). In urban locations, underprediction of PM and AOD with a coarse resolution (∼300 km) global model is expected because urban scale pollution is not resolved. Underprediction in rural areas is consistent with previous studies that have found a persistent underestimation of AOD in biomass burning influenced regions, requiring a scaling factor of ∼2-5 to be applied to fire emissions (e.g. Kaiser et al 2012, Tosca et al 2013. We do not scale fire emissions in this analysis since there is considerable uncertainty associated with this scaling factor, which is likely to have large spatial variability. Since we underpredict PM concentrations, particularly in urban locations, we do not focus on the fractional contribution of fires to total PM 2.5 . Instead, the focus of our study is to compare the relative contributions of fires from different regions to PM 2.5 concentrations and explore how these contributions change over the seasonal cycle and between different years. 3.2. Regional contribution of fires to PM 2.5 over Singapore  largest contribution in October. There are also substantial contributions from fires in August (mean 5.1 μg m −3 ) and September (mean 9.5 μg m −3 ). During these months, the Southwest Monsoon predominates over Southeast Asia, with much of the southern part of the region experiencing dry season conditions conducive to fires. Fires contribute little to PM 2.5 during November-January when the Northeast Monsoon brings heavy rain to the region, reducing the susceptibility of vegetation to burning. Towards the end of the Northeast Monsoon season and during the onset of the Southwest Monsoon season, fires contribute moderately to monthly-mean PM 2.5 over Singapore, particularly in February (mean 3.6 μg m −3 ), March (mean 2.8 μg m −3 ) and June (mean 3.3 μg m −3 ).
There is considerable inter-annual variability in the contribution of fires to simulated monthly-mean PM 2.5 in Singapore: in October fires contribute as little as 0.3 μg m −3 to as much as 55.6 μg m −3 depending on the year (figures 4(a) and S1). Although anthropogenic land-use change is the underlying cause of the majority of vegetation and peat fires occurring over Southeast Asia (Langner et al 2007, Carlson et al 2012, the frequency and magnitude of these fires are linked to inter-annual variability in sea surface temperature and atmospheric conditions, namely El Niño Southern Oscillation (ENSO) (van der Werf et al 2004, Field and Shen 2008, Logan et al 2008, Shi et al 2014. During El Niño years, rainfall over Southeast Asia decreases, which can lead to drought conditions increasing the susceptibility of . Satellite observations show that ENSO has a substantial effect on the number of fire plumes and the aerial extent of smoke clouds observed over Borneo (Tosca et al 2011). The westward transport of smoke from Sumatra is also thought to be greater during El Niño years, particularly during October-November, due to the combination of higher smoke-plume injection heights and anomalous easterlies over tropical Southeast Asia (Xian et al 2013). The plume injection heights are fixed in our model, but a strong increase in PM 2.5 over Singapore is simulated in October 2006 due to the combination of an increase in GFED3 fire emissions and the prevailing wind direction (specified by analysed meteorological fields). Figure S1 shows that contributions from fires to simulated PM 2.5 in Singapore are substantially higher in dry season months with a positive Oceanic Niño Index (ONI; indicating El Niño conditions) relative to those with negative ONI values (indicating La Niña conditions). In the weak to moderate El  Figure 4(b) shows the contribution from fires occurring in different regions (defined in figure 1) to simulated PM 2.5 concentrations in Singapore. Fires in southern Sumatra make the largest contribution to monthly-mean PM 2.5 in Singapore with long-term mean contributions of 7.2 μg m −3 (max. 20.7 μg m −3 ) in September and 7.0 μg m −3 (max. 37.7 μg m −3 ) in October. As a fractional contribution, fires in southern Sumatra account for 76% of total enhancement to PM 2.5 from fires in September and 66% in October. In these months, fires occurring in Indonesian Borneo make the second largest contribution to PM 2.5 (mean 3.0 μg m −3 , max. 17.3 μg m −3 in October), accounting for 12% of the total fire enhancement in September and 28% in October. From February to August, fires in central Sumatra make the largest contribution to PM 2.5 , peaking in February (mean 3.1 μg m −3 ). During this period, the fractional contributions from fires in central Sumatra range from 87% in February to 47% in August.
There is a small contribution to PM 2.5 from fires in Indochina in January, February and March (max. 0.9 μg m −3 ), during which the Indochina region experiences relatively dry conditions. Fires occurring outside the regions discussed above have minor impacts on air quality in Singapore.
Over the whole study period , fires in southern Sumatra account for the largest fraction of the total fire enhancement to PM 2.5 in Singapore (42%), with fires in central Sumatra and Indonesian Borneo contributing 35% and 14%, respectively. During El Niño years (2004,2006,2009) there are greater fractional contributions from fires in southern Sumatra (54%) and Indonesian Borneo (17%) relative to the La Niña years (29% and 9%, respectively). The fractional contribution from fires in central Sumatra remains similar between the El Niño (22%) and La Niña (26%) years, but is particularly large during 2005 (84%) due to the persistence of El Niño conditions throughout the 2004/2005 winter wet season, leading to drought conditions in this region (Yulianti et al 2013). Fires occurring in Peninsular Malaysia and Indochina account for a relatively small fraction (∼2-3%) of the total fire enhancement to PM 2.5 between 2004 and 2009, with slightly larger fractional contributions in La Niña years (4 and 12%, respectively).

Atmospheric back trajectory results
We used atmospheric back trajectories to further explore the relative contributions from fires in different regions to smoke transported to Singapore. Figure 5 shows the relative contributions from each region to fire-emitted CO transported to Singapore for the period January 2004 and December 2007. The largest amount of fire emission transported to Singapore occurs in September and October with the greatest contribution from fires in southern Sumatra and a secondary contribution from fires in Indonesian Borneo. Fires in central Sumatra result in transport of fire emissions to Singapore during March-October, with the greatest contributions in March, June and August. This analysis demonstrates substantial seasonal (figure S3) and inter-annual variability in the transport of fire emissions to Singapore with maximum contributions simulated in September and October 2006, matching the variability simulated by the global model ( figure S1).
Over the whole study period , fires in southern Sumatra make the largest relative contribution to the total fireemitted CO transported to Singapore (62%). This is consistent with previous work that demonstrated that more than 90% of hazy days in Singapore (visibility < 8000 m) occurred when back trajectories passed over southern Sumatra on days with fire hot spots (See et al 2006). Fires in central Sumatra and Indonesian Borneo contribute 21% and 15%, respectively of the total fire emissions transported to Singapore. We note that whilst this analysis accounts for locations of fires and the atmospheric transport of emissions, it does not account for variations in atmospheric residence time driven by changes in deposition of particulates. Assuming an infinite residence time of the fire-emitted CO, this method predicts a larger relative contribution from fires in southern Sumatra (thus a smaller relative contribution from fires in central Sumatra) compared to the global model where changes in residence times are explicitly simulated. Figure 6 shows the spatial distribution of fire emissions transported to Singapore during 2004-07. Fires in southern Sumatra, central Sumatra and southwest Kalimantan are the largest contributors of fire-emitted CO transported to Singapore, matching the results from the global model. However, the higher spatial resolution of this analysis compared to the global model simulations allows finer scale features to be resolved. The fires in southern Sumatra are located in the eastern part of the province of South Sumatra and the southeastern parts of the provinces of Jambi and Lampung. The fires in central Sumatra are located in the eastern part of the province of Riau, and the fires in southern Borneo are located in the southern extent of the provinces of West, Central and South Kalimantan.  Relative contributions are calculated as the total CO from each emission grid cell emitted into air parcels arriving at Singapore as a fraction of the total CO emitted into all air parcels over the entire period and domain. Emission from grid cells into each air mass trajectory is calculated as the emission flux at the grid cell over which the trajectory passes multiplied by the residence time of the trajectory over the grid cell.

Contribution of fires to PM 2.5 in other major cities in Southeast Asia
To explore the impact of fire emissions over the wider region of Southeast Asia, we used the GLOMAP model to simulate the contribution of fire emissions to PM 2.5 concentrations in 15 cities across the region with a population greater than one million (locations shown in figure 1). Figure 7 shows the simulated long-term (2004-09) mean contribution of fires to monthly-mean PM 2.5 concentrations in six of these cities. It is important to highlight that the simulated enhancement to PM 2.5 concentrations from fires is greater in four of these cities (Palembang, Pekanbaru, Phnom Penh and Mandalay) than in Singapore; due to their closer proximity to biomass burning regions.
Cities in southern Sumatra (Palembang) and western Java (Jakarta) are affected by fires in a similar way to Singapore, with peak contributions from fires to PM 2.5 in September (mean Palembang: 11.8 μg m −3 ; Jakarta: 5.4 μg m −3 ) and October (Palembang: 14.7 μg m −3 ; Jakarta: 6.2 μg m −3 ). , fires in southern Sumatra account for 60-74% of the total fire enhancement to PM 2.5 in Jakarta and Palembang, with 5-7% from fires in central Sumatra. Fires in regions located outside those defined in figure 1 (e.g. northern Australia) account for 8-27% of the total fire enhancement to PM 2.5 .
In Kuala Lumpur (Peninsula Malaysia) and Pekanbaru (Riau, central Sumatra) there is a considerable contribution from fires to PM 2.5 throughout the year with the largest contribution from fires in central Sumatra (69-74% of the total fire enhancement). Both cities are also impacted by fire emissions from southern Sumatra during August-October (17-20% of the total fire enhancement).
The contribution of fires to PM 2.5 in Batam, Riau Islands (not shown in figure 7) is similar to Singapore, with the largest contribution from fires in southern Sumatra (51% of the total fire enhancement to PM 2.5 ) and secondary contributions from fires in central Sumatra (23%) and Indonesian Borneo (15%). The cities of Singapore and Batam are close together and are affected by similar wind patterns and transport pathways of smoke emissions. Simulated long-term mean contributions of fires to PM 2.5 over the other analysed cities in Indonesia (Surabaya, Medan and Makassar) and Manila in the Philippines are relatively small (mean ⩽ 2 μg m −3 ).
Simulated PM 2.5 concentrations over cities in Indochina (Hanoi, Ho Chi Minh, Yangon, Mandalay, Phnom Penh, and Bangkok) are affected by local fire emissions from January to April, with little contribution from fires during May-November. The two cities with the largest contributions from fires, Phnom Penh and Mandalay, are shown in figure 7. Contributions to PM 2.5 concentrations from fires in this region (dominated by deforestation and agricultural fires) show less inter-annual variability than the contributions from fires in Sumatra and Borneo (dominated by peat and deforestation fires).

Conclusions and implications
We have used two complementary methods to explore the contribution of smoke from vegetation and peat fires across Southeast Asia to PM 2.5 concentrations in Singapore. Firstly, we used a global aerosol model to simulate PM 2.5 concentrations and investigate the contributions from different defined regions by switching off fire emissions from one region at a time. Secondly, we used a Lagrangian atmospheric transport model to simulate the transport of fire emissions to Singapore. Both methods used fire emissions derived from satellite remote sensing. Our global model simulations are restricted to a coarse spatial resolution meaning that the model is unable to resolve urban scale pollution and absolute PM 2.5 concentrations in Singapore are underpredicted. We are therefore not able to quantify the fractional contribution of fires to total PM 2.5 concentrations in Singapore. Despite the coarse spatial resolution of our model, it is able to capture observed seasonal and inter-annual variability in observed PM and AOD. Future work should repeat our analysis with regional atmospheric models with higher spatial resolution. Nevertheless, our two methods produced consistent results providing additional confidence in our conclusions.
Over the studied period (2004-09), we found that Singapore is impacted by smoke from vegetation and peat fires predominantly in June-October and February-March, with the largest contribution from fires occurring in September and October. We found that fires in southern Sumatra account for the largest percentage of the total fire enhancement to PM 2.5 in Singapore (42-62%), with fires in central Sumatra and Indonesian Borneo (Kalimantan) contributing 21-35% and 14-15%, respectively. Therefore, fires in central Sumatra that are closest to Singapore were not the dominant source of fireinduced PM 2.5 pollution in Singapore between 2004 and 2009. Instead, fires in South Sumatra (located ∼550 km from Singapore) contributed more to PM 2.5 pollution in Singapore. Establishing this source-receptor relationship in a quantitative manner provides useful information for policy makers in the region to develop appropriate air quality management strategies to protect public health. Our work suggests that preventing fires occurring in southern Sumatra (in the eastern part of the provinces of Jambi, South Sumatra and Lampung), central Sumatra (in the eastern part of the province of Riau) and southwest Kalimantan (in the southern extent of the provinces of West, Central and South Kalimantan) would result in the largest benefit to particulate air quality over Singapore.
We also explored the impact of vegetation and peat fires on PM 2.5 concentrations across other major cities in the region. We found the fires that contributed most to PM 2.5 concentrations in Singapore also contributed substantially to PM 2.5 concentrations across the rest of the region. Jakarta, Palembang and Batam are mostly impacted by fires in southern Sumatra (accounting for 51-74% of the total fire enhancement to PM 2.5 ), whereas Kuala Lumpur and Pekanbaru are impacted most by fires in central Sumatra (accounting for 69-74% of the total fire enhancement to PM 2.5 ). Therefore, targeting fire reduction efforts to improve air quality in Singapore will also improve air quality in other major cities in Indonesia and Malaysia.
The large fire emissions from central and southern Sumatra and Kalimantan that are causing air quality issues in Singapore are coincident with ongoing deforestation and expansion of oil palm plantations (Koh et al 2011, Ramdani and Hino 2013. Forest cover in Southeast Asia declined by 1.45 million hectares per year over the period 2000-10 (Stibig et al 2014), with natural forest largely being replaced by oil palm plantations (Koh et al 2011). The extent of future fires in the Southeast Asia region will be linked to patterns of land-use change (Carlson et al 2012), potentially exacerbated by changing climate conditions (Field et al 2009). In particular, the expansion of large-scale oil palm plantations, which have been linked to the use of fire for land clearing activities (Varkkey 2013), is likely to be a key driver of future fires. Future land use change, under business as usual scenarios, includes continued expansion of oil palm plantation in regions that we have identified to be crucial for air quality in Singapore: southern Kalimantan (Carlson et al 2012) and central to southern Sumatra (e.g. Wicke et al 2011, Miettinen et al 2012. In Kalimantan, 79% of allocated oil palm leases remain undeveloped (Carlson et al 2013), creating the potential for large future expansion of this land use type. In this region, rates of deforestation are significantly greater in oil palm concessions compared to natural forest timber concessions or protected areas (Gaveau et al 2013), suggesting that maintenance and expansion of protected areas and natural forest timber concessions would help reduce deforestation and prevent future fires. Strict enforcement of regulations prohibiting the use of fire to prepare lands for plantation agriculture may mitigate smoke emissions and improve regional air quality. However, plantation agriculture on peatlands requires extensive drainage (Carlson et al 2012) creating conditions conducive to the spread of accidental fire during El Niño droughts (Field et al 2009). Furthermore, oil palm plantations have been linked to enhanced regional ozone concentrations due to altered emissions of biogenic volatile organic compounds from vegetation and emissions of oxides of nitrogen from agro-industrial activity (Hewitt et al 2009, Warwick et al 2013. To ensure improved air quality in Singapore and surrounding regions, fire management needs to be combined with sustainable agricultural expansion that does not lead to deforestation, forest degradation or peatland drainage (Smit et al 2013).