El Niño and health risks from landscape fire emissions in Southeast Asia.

Emissions from landscape fires affect both climate and air quality1. In this study, we combine satellite-derived fire estimates and atmospheric modeling to quantify health effects from fire emissions in Southeast Asia from 1997 to 2006. This region has large interannual variability in fire activity due to coupling between El Niño-induced droughts and anthropogenic land use change2,3. We show that during strong El Niño years, fires contribute up to 200 μg/m3 and 50 ppb in annual average fine particulate matter (PM2.5) and ozone (O3) surface concentrations near fire sources, respectively. This corresponds to a fire contribution of 200 additional days per year that exceed the World Health Organization (WHO) 50 μg/m3 24-hour PM2.5 interim target (IT-2)4 and an estimated 10,800 (6,800-14,300) person (~2%) annual increase in regional adult cardiovascular mortality. Our results indicate that reducing regional deforestation and degradation fires would improve public health along with widely established benefits from reducing carbon emissions, preserving biodiversity, and maintaining ecosystem services.

Fires are pervasive instruments of land management in the tropics for clearing debris in the process of deforestation and agricultural management. These fires enable an economical and effective method for expanding and maintaining agricultural production, but release gases (including O3 precursors) and aerosols (mostly black and organic carbon) that interact with the climate system[5], degrade surface air quality[1], and jeopardize public health[6]. Fires associated with deforestation emitted ~1.0 Pg C/yr from 2000–107, with considerable interannual variability from droughts in tropical forests[8]. Fires also contribute to PM2.5 and O3 increases, which are both detrimental to public health[1,4]. Projections of greater fire activity in a warming climate[9] suggest increasing contributions to atmospheric concentrations and population exposure.

Globally, most fires occur in Africa and South America[8], but recent studies have highlighted the importance of Southeast Asia because of high population densities near high fire activity[10]. Regional emissions may differ by a factor of 50 between opposite phases of the El Niño-Southern Oscillation (ENSO). In the Global Fire Emissions Database version 3 (GFED3), regional fire emissions were 1069 Tg C during the 1997 El Niño but only 21 Tg C during the 2000 La Niña[8], illustrating the nonlinearity between fires and drought[11]. During the warm phase of ENSO and the Indian Ocean Dipole, cool sea surface temperature anomalies near Indonesia decrease regional rainfall[2,12]. Landowners ignite fires to clear land and manage agricultural areas[3], and although typically too wet to combust, deforestation and degradation have enhanced the susceptibility of peatland forests (with carbon-rich peat deposits) to human-ignited fire during droughts[13].

PM2.5 and O3 exposure increases hospital admissions and mortality from respiratory and cardiovascular diseases, even at low concentrations[4]. During the 1997–98 fires in Southeast Asia, daily ground-level PM concentrations reached hazardous levels[6], with concomitant negative impacts on respiratory and general health[14]. Increases in respiratory illnesses were also reported in Singapore from transported emissions[15]. While these studies offer some information on the health effects of fires, they have been confined to specific locations or time periods by limited data availability.

We expand on these local studies by using satellite data and two atmospheric models, NASA GISS-E2-PUCCINI general circulation model (GCM) and Harvard University’s GEOS-Chem chemical transport model (CTM) to estimate pollutant concentrations and corresponding regional mortality from 1997 to 2006, applying existing concentration-response functions from the epidemiological literature (See Methods). Atmospheric models simulate the transport of fire emissions and formation of pollutants, offer a continuous spatiotemporal dataset in a region with limited ground monitoring but large rural populations, and allow us to examine how climate and emissions influence aerosol and trace gas concentrations interannually.

Our study region is a 50°×30° area (92.5°E-142.5°E, 20°N-10°S) encompassing the Association of Southeast Asian Nations (ASEAN). In 2005, the population was approximately 540 million (Fig. 1a). Fire activity, predominately in the Indonesia islands of Sumatra and Borneo (Fig. 1b), peaks during the dry southern monsoon of September and October, along with potential spring burning[2,6].

Figure 1

Study area population and locations of fire activity

a, 2005 population density, in persons per km2, for countries belonging to the Association of Southeast Asian Nations (ASEAN). Data from CIESIN GPWv3[29] at 0.25° resolution. b, 1997–2006 mean fire emissions, in g C/m2/month at 0.5° resolution, from the Global Fire Emissions Database version 3[8].

The additional contribution of fires to annual surface PM2.5 and O3 concentrations in 1997, a strong El Niño year, greatly increases the number of days that exceeded the WHO interim targets of 50 μg/m3 24-hour PM2.5 (IT-2) and 80 ppb 8-hour maximum O3 (IT-1), which are both twice the WHO’s air quality guidelines (Fig. 2; Supplementary Table S1). In 1997, both models show two distinct areas of fire-derived PM2.5 over Sumatra and Borneo with concentrations elevated by 50–200 μg/m3 and with increases of 50–150 days over the WHO interim targets. O3, in contrast, had widely distributed increases of 25–50 ppb and up to 150 exceedance days. Corresponding results with all sources in 1997 are in Supplementary Fig. S1, this simulation captured the general temporal evolution seen in ground observations (Supplementary Figs. S2, S3, S4; Supplementary Table S2).

Figure 2

Modeled annual mean 1997 surface concentrations and corresponding additional daily exceedances in 1997 due to fires only

a, PM2.5 b, O3 annual concentrations and daily exceedances over World Health Organization (WHO) interim targets (50 μg/m3 daily PM2.5 (IT-2) and 80 ppb 8-hour maximum O3 (IT-1)). Annual concentrations are from 24-hour PM2.5 and 8-hour maximum O3. GISS refers to GISS-E2-PUCCINI and G-C refers to GEOS-Chem.

We explored how modeled concentrations with and without fire emissions affect population exposure to WHO interim targets (Supplementary Table S1)[4]. Decadal exposure over these interim targets, along with the fraction of exposure due to fire, shows how the major influence of fires was not confined to the 1997–98 El Niño (Fig. 3). Interannual variability in exposure for both short- and long-term guidelines is dominated by the fire contribution of PM2.5 and O3; the WHO’s 25 μg/m3 annual PM2.5 interim target (IT-2) is never exceeded without including fire emissions.

Figure 3

Population exposure above World Health Organization (WHO) interim targets

a, Exposure over 50 μg/m3 24-hour PM2.5 interim target (IT-2). b, Exposure over 25 μg/m3 annual PM2.5 interim target (IT-2). c, Exposure over 80 ppb 8-hour maximum O3 interim target (IT-1). d, Fraction of population exposure above each WHO interim target that is attributable to fires. Each case is calculated with and without GFED3 fire emissions using GISS-E2-PUCCINI results, which was close to the average concentration estimate. Refer to Supplementary Table S1 for estimated health effects. Note the logarithmic scale for (a) and (c).

We also tested the sensitivity of regional health impacts, including exceedances and cardiovascular disease mortality, to using the original model or satellite-scaled model PM2.5 estimates (Supplementary Figure S5). The mortality estimates combine modeled pollutant concentration changes from fires with published epidemiological relationships between exposure to O3 or PM2.5 total mass and cause-specific mortality (See Methods). In Table 1, 1997 and 2000 highlight the considerable differences in health effects between years with high and low fire contributions. For example, PM2.5 annual exposure in 2000 hardly exceeds the WHO interim target and O3 exposure is 100 times lower than in 1997. During high fire years, fire emissions increase the adult cardiovascular disease mortality burden by approximately 10,800 (6,800–14,300) annual deaths from PM2.5 exposure and an additional 4,100 (2,300–5,900) annual deaths from O3.

Table 1

Fires-only concentration, exposure, and mortality using different models for an El Niño (1997) and La Niña (2000)

Average ASEAN annual concentration due to fires only (from 24-hour PM2.5 and 8-hour maximum O3); additional exposure due to fires above the annual 25 μg/m3 PM2.5 interim target (IT-2; ×106 person-years) and above the 80 ppb daily 8-hour maximum O3 interim target (IT-1; ×107 person-days); cardiovascular mortality due to fires only (×103 people), with the range from 95% confidence intervals from epidemiological studies. GISS refers to GISS-E2-PUCCINI and G-C refers to GEOS-Chem, also with satellite scaling factors.

a) PM2.5
	Concentration (μg/m3)		Exposure above IT-2 (×106 person-years)		Mortality (×103 people)
	1997	2000	1997	2000	1997	2000
GISS	7.8	0.3	55.6	0.0	9.9 (8.0–11.4)	1.0 (0.8–1.2)
G-C	3.7	0.2	25.8	0.0	8.7 (6.8–10.7)	1.5 (1.1–1.9)
GISS MISR	10.7	0.4	57.0	0.0	11.2 (9.6–13.5)	1.3 (1.0–1.6)
G-C MISR	7.4	0.5	59.1	4.7	10.1 (8.1–11.8)	1.7 (1.4–2.1)
GISS MODIS	12.0	0.4	66.6	0.0	12.5 (10.1–14.2)	1.5 (1.1–1.8)
G-C MODIS	8.3	0.5	50.3	0.0	12.1 (9.7–14.3)	2.3 (1.8–2.8)
AVERAGE	8.3	0.4	52.4	0.8	10.8 (6.8–14.3)*	1.6 (0.8–2.8)*

Maximum error range.

Modeled annual adult cardiovascular disease mortality shows a strong correlation with the multivariate El Niño Index (MEI)[16], which was averaged over the July to October dry season (Fig. 4). We present the most conservative mortality estimates, but this relationship holds with varying relative risk (RR) relationships or durations of exposure (Supplementary Fig. S6). Reduced convection during El Niño years likely increases exposure by elevating emissions[5,11] and increasing aerosol lifetimes by reducing wet deposition..

Figure 4

Additional annual cardiovascular disease (CVD) mortality from exposure to fire-contributed annual PM2.5 and 24-hour O3, along with the Multivariate El Niño Index (MEI)[16]

Results for 1997–2006 are from the baseline GISS-E2-PUCCINI and GEOS-Chem concentrations, with the power-law RR relationship for CVD mortality. R2=0.87–0.91 for PM2.5 and R2=0.82–0.89 for O3. AOD-scaled results and sensitivity analysis are in Table 1, Supplementary Table S4, and Supplementary Fig. S6.

Uncertainty in our health effect estimates comes primarily from: 1) the fire emissions dataset, 2) atmospheric modeling, and 3) concentration-response equations. First, van der Werf et al. (2010) estimated fire carbon emissions uncertainty at 20% globally, though higher in equatorial Asia due to peat carbon stock uncertainties and in years before MODIS data[8]. Second, although the lack of ground stations precludes an in-depth evaluation, available ground data and satellite AOD indicate that both models are likely conservative (Supplementary Figs. S3, S4, S5). The range between the two model scenarios (PM2.5 and O3) and two satellite AOD optimized results (PM2.5 only) provide some insight about uncertainty related to transport and deposition processes. There is up to a factor of two difference between models (less among satellite-optimized estimates), but this range is expected given previous findings that model physics and parameterizations drive more variation in aerosols than emissions[17]. Differences in our PM2.5 concentrations are primarily driven by lower precipitation and wet deposition in the GISS model, which increase aerosol lifetime relative to GEOS-Chem (data not shown). However, for the purposes of health impacts the results are much closer (Table 1). This is due to the nonlinear relationship between the RR and exposure, which reduces differences between mortality estimates at high concentrations. Finally, we address mortality equation uncertainties through 95% confidence intervals around the concentration-response estimates (Table 1) and various estimates of the RR and PM2.5 exposure relationship (Supplementary Tables S3 and S4; Supplementary Fig. S6). Additional epidemiological factors that we did not address are extrapolation of RR equations to high concentrations and applying equations developed in the U.S. to non-U.S. populations. In addition, evidence for potential differences in PM2.5 toxicity between urban pollution in U.S. cities and Southeast Asian fire emissions is too limited to warrant using separate epidemiological equations[18], so we assume that total PM2.5 mass is the most appropriate metric.

These uncertainties and additional factors contribute to our substantially lower regional PM2.5 mortality estimates relative to the global analysis of Johnston et al. (2012)[10]. The two estimates are not directly comparable. Our conservative estimates (Supplementary Table S5) are based on a tailored regional analysis for ASEAN countries and use updated fire emissions, multiple atmospheric models, epidemiological equations developed over a wide concentration range, and cause-specific disease estimates (Supplementary Table S6). We did not include children since the epidemiological equations were developed for adults over 30 years; this cuts out more than half of the population and ignores risks to infants and children.

While previous work in Borneo has emphasized the value of avoided deforestation in terms of carbon emissions[19], it is important to also account for health. By demonstrating the direct link between climate variability and health impacts from fire emissions throughout Southeast Asia, we offer additional support for policies that use regional climate forecasts to restrict burning during high fire risk seasons. Fire emissions during 1997 to 2006 repeatedly exposed 1–11% of the population in Southeast Asia to PM2.5 and O3 above WHO interim targets during El Niño years. Although the regional influence of climate change is uncertain[20], these observed trends would be exacerbated by the potential for more frequent droughts related to El Niño and increased baseline cardiovascular disease caused by demographic shifts towards sedentary lifestyles and increased animal product consumption. Reducing fires from deforestation and land management would benefit public health in addition to global-scale benefits for carbon storage and biodiversity.

Methods

Fire emissions estimates are from GFED3, a global gridded monthly emissions dataset that combines surface reflectance and active fire detection data from several satellites to detect the spatiotemporal variability of burned area[21]. This drives a biogeochemical model that estimates fuel loads, combustion completeness, and emissions[8]. GFED3 is available since 1997 at 0.5°×0.5°. We define landscape fires to include all burning sources; in Southeast Asia this includes peat, forest, agricultural waste burning, deforestation and degradation.

We use two models: the NASA GISS-E2-PUCCINI GCM from 1997–2007 and Harvard University’s GEOS-Chem CTM[22] from 1997–2006. See Supplementary Information for descriptions of the models, spin-up, and boundary conditions. Both were run at 2°×2.5°, including a control run without fire emissions and a perturbed run with GFED3 emissions. We define years from July 1st to June 30th to avoid splitting a burning season into two years. Since meteorological fields for GEOS-Chem are available through December 2006, we only have a complete 2006 “fire-year” from GISS. For PM2.5, we analyzed 24-hour and annual average concentrations. For O3, we used 1–2pm concentrations as a proxy for the 8-hour maximum (Supplementary Fig. S4) and 24-hour concentrations for mortality calculations.

Aerosol optical depth (AOD) data from MISR and MODIS satellite instruments, available from 2001–2006, were used to scale modeled AOD; these scaling factors were then applied to modeled PM2.5 (Supplementary Fig. S5). While AOD represents total column aerosol loading, it is often closely related with surface abundance[23], and hence provides some measure of large-scale biases in the models. Scaling factors were applied to surface PM2.5 for all grid boxes, maintaining the modeled spatial and temporal distribution of aerosols.

We evaluate health effects by estimating: 1) exposure above WHO short-term and annual air quality targets, and 2) cause-specific adult mortality. Mortality attributable to fires combines the relative risk (RR) from changes in pollutant exposure with baseline observed mortality rates. We focus on cardiovascular disease because it is a proximal outcome from exposure that will be experienced annually. However, this underestimates total mortality due to other long-term effects and short-term exposure. The equations that we use were developed for adults (less than half of the regional population[24]).

We applied a power-law relationship between RR and PM2.5. Due to the lack of data on differential health effects of biomass smoke particles[18], we use an equation developed for total PM2.5 mass: which describes the relationship between PM2.5 exposure and cardiovascular disease mortality risk over a large concentration range[25]. Pope et al. (2011) published values for α and β by reanalyzing previous estimates of RR and dose of PM2.5 (in mg) from ambient air pollution, second-hand smoke, and cigarette smoke. For cardiovascular disease, α=0.2685 and β=0.2730. Although ambient PM2.5 concentrations from fires will not reach the cigarette smoke doses included in Pope et al. (2011), this equation was essential due to our high ambient concentrations above the range of other studies. Since 95% confidence intervals were given for each individual study but not the overall relationship, we refit a power-law relationship to approximate the uncertainty based on the individual studies’ upper or lower limits, respectively. The annual average of 24-hour total mass PM2.5 concentrations were used for (C), assuming a constant average inhalation rate (I) of 18 m3/day to convert to PM2.5 dose (in mg)[25]. We separately calculated the RR using concentrations with and without fires due to the equation’s nonlinearity. We then followed the approach of Ostro et al. (2004)[26] to calculate the attributable fraction (AF) and annual mortality (ΔM): where the average annual baseline mortality rate (Mb) was calculated from adult deaths due to cardiovascular disease, averaged over the countries in ASEAN[27]. Population with ages greater than 30 years was from the UN Population Division[24] and CIESIN’s Gridded Population of the World version 3 and Future Estimates, aggregated to the model resolution[28,29]; both were interpolated from 5 yearly data to annual estimates.

For O3, the linear RR is given by: where δ=1.11 (0.68–1.53) is the percent increase in cardiovascular disease morality per 10 ppb increase in 24-hour O3 concentrations, based on a meta-analysis of U.S. and non-U.S. studies[30]. Daily mortality due to fire pollution is then estimated with: using the population characteristics described above. We assume that mortality is evenly spread throughout the year (Mb is not year-specific so we divide consistently by 365), and sum by days per year to obtain annual estimates. GEOS-Chem includes leap years, but GISS uses a fixed 365 day calendar. Bell et al. (2005) concluded that the O3 mortality burden was insensitive to PM[30], indicating that this is separate from PM2.5 mortality.