Peat is an accumulation of partially decayed vegetation or organic matter that can be further classified as fibric, hemic, or sapric (by increasing degree of decomposition and density, Wüst et al., 2003). Different amounts of roots; sound or rotten logs; charred logs, char, and ash from previous burns; and mineral soil are frequently mixed in with the peat along with varying amounts of water. On undisturbed sites deeper peat is normally more decomposed and denser, but on disturbed sites the upper layer is sometimes already removed by previous fires, while dredging for canals can place “older peat” on top of younger peat and road building can compact the peat. Traditional peat classification schemes can be less straightforward for disturbed areas. For instance, ferns and grasses can contribute fibrous roots to a layer of older, even sapric, material. We note that the Kalimantan peat burned in the FLAME-4 lab study that we will compare to was sampled in both undisturbed forest (one sample) and previously logged/burned forest (one sample), whereas the peat fires sampled in this field work were all on moderately to heavily disturbed sites, which is generally where fire activity is the highest.

Peat deposits can burn at > 100 % fuel moisture (defined as 100 × (wet - dry)/dry)), where “wet” refers to the weight of a fresh fuel sample and “dry” refers to the fuel weight after oven drying until mass loss ceases. This is because the glowing front pre-dries the fuel as it advances. Peat combustion can occur as a glowing front in an expanding pit or undercut, but with direct access to surface air (Huang et al., 2016), which we term “lateral spreading.” The glowing front can be covered by ash or initially propagate downward on inclusions or in cracks in initially, mostly unburned peat, which we refer to as “downward” spreading, but this is much less common. Figure S1 in the Supplement shows photographs of these spread modes. The glowing front is the site of gasification reactions (O2 oxidation of char) that produce mostly CO2, CO, CH4, NH3, and little visible aerosol. The heat from glowing combustion pyrolyzes the adjacent peat, producing relatively more organic gases and copious amounts of white smoke (with high OA content) (e.g., Fig. 3 in Yokelson et al., 1997). Wind increases the glowing front temperature. Oxygen availability is likely higher for lateral spreading than downward spreading fire and the overburden in downward spreading fires may scavenge some emissions. Occasionally peat can support brief, small flames if the surface peat is not too dense, or has high flammable inclusion content or at high wind speeds (Yokelson et al., 1996, 1997).

During 8 days from 31 October through 7 November, we sampled 35 separate plumes at six different peatland areas with two areas being revisited (Table S1 in the Supplement). All smoke sampling was conducted directly in the visible plumes (Fig. S1) and all background sampling was conducted just outside (usually upwind) of the plumes in paired fashion. The surface fuels at all sites were nonexistent or limited to ferns, charred logs, or patchy second growth forest, but they were neither present in heavy loading nor burning in most cases. This facilitated sampling “pure emissions” from the smoldering peat. On each day from 1 to 7 November, about four plumes originating from various peat types or depths were grab sampled about 10 times each by FTIR, at least once by WAS, and usually by filters. This provided data for 27 plumes, each assigned a letter identifier in our tables from A–Z to AA. Eight additional plumes were quickly, opportunistically, sampled by just WAS, which was the fastest sampling method to complete. On 5 and 6 November, seven of the plumes with letter identifiers were also sampled continuously between 10 and 30 min apiece with both PAXs (coincident with FTIR, WAS, and filter sampling). Twenty-two filter samples were collected from 19 different “lettered” plumes from 1 to 7 November. The full set of filter-based analyses will be reported separately (Jayarathne et al., 2016). The sites and fires sampled included a variety of peat types, disturbance levels, spread modes, burn depths, etc. A brief chronological narrative of the sampling follows and most of the site characteristics that we were able to document are shown in Table S1. A site map is given in Fig. S2.

31 October (site 1). Two WAS samples were collected while scouting this site known locally as “South Bridge West” late in the afternoon. The site (site 1 in Table S1) had hemic and fibric peat burning at 30–60 cm depth and was the most disturbed of all the sites sampled.

1 November (site 1), plumes A–D. The “South Bridge West” site 1 was revisited and sampled by WAS, FTIR, and filters, which began the series of intensively sampled plumes designated by letters. Plume C included emissions from surface peat that were partially impacted by flames during wind gusts.

2 November (site 2), plumes E–H. This site was the least disturbed of the sites we sampled but had been logged and was known to have burned once before the fire we sampled. In addition, site 2 was close enough to a canal that its hydrology would have been impacted. The site is known locally as “South Bridge East.” The peat was hemic and fibric and burn depth ranged from 18 to 28 cm.

3 November (site 3), plumes I–L. The “White Shark (Hiu Putih)” site comprised hemic and fibric peat burning at depths of 33–52 cm.

4 November: (site 4) plumes M–N; (site 5) Plume P. Site 4 was known locally as the “Mahir Mahar” site and plume M provided our best measurements of the emissions from burning sapric peat. The other plumes sampled were burning in hemic and fibric peat types. The burn depths sampled on this day varied over a narrow range near 21–22 cm.

5 November (site 1), plumes Q–T. The South Bridge West site was revisited. Burn depths were 25–50 cm and the peat was hemic and fibric.

6 November (site 2), plumes U–W. The South Bridge East site was revisited. The peat was hemic and fibric and burn depths were 20–30 cm.

7 November (site 6), plumes X–Z–AA. Some shallow peat combustion was sampled at this site, known locally as Tangkiling Road.

The excess mixing ratios above the background level (denoted ΔX for each gas-phase species “X”) were calculated for all the gas-phase species in the grab samples and CO2, CO, and CH4 in the real-time data. The grab samples were collected in a way that avoided possible artifacts for some gases due to adsorption on filters or in flow meters and they were used to produce a self-consistent complete set of data on trace gas emissions, as described next. The molar emission ratio (ER; e.g., ΔX / ΔCO) for each gaseous species X relative to CO or CO2 was calculated for all the FTIR and WAS species. The plume-average ER for each FTIR or WAS species measured in multiple grab samples was estimated from the slope of the linear least-squares line (with the intercept forced to zero) when plotting ΔX vs. ΔCO (or ΔCO2) for all samples of the source (Yokelson et al., 2009; Christian et al., 2010; Simpson et al., 2011). Forcing the intercept decreases the weight of the lower points relative to those obtained at higher concentrations that reflect more emissions and have greater signal to noise. Alternate data reduction methods usually have little effect on the results, as discussed elsewhere (Yokelson et al., 1999). For a handful of species measured by both FTIR and WAS it is possible to average the ERs from each instrument for a source together as in Yokelson et al. (2009). However, in this study, we either worked up the independently sampled WAS data as a separate set of ER or used the more extensive FTIR ERs when there were a few “overlap species” (primarily CH3OH, C2H4, C2H2, and CH4).

From the ERs, EFs were derived in units of grams of species X emitted per kilogram of dry biomass burned by the carbon mass balance method, which assumes all of major carbon-containing emissions have been measured (Ward and Radke, 1993; Yokelson et al., 1996, 1999): EFXgkg-1=FC×1000×MMxAMC×ΔXΔCO∑j=1nNCj×ΔCjΔCO, where FC is the measured carbon mass fraction of the fuel, MMx is the molar mass of species X, AMC is the atomic mass of carbon (12 g mol-1), and NCj is the number of carbon atoms in species j; ΔCj or ΔX referenced to ΔCO are the fire-average molar ERs for the respective species. The carbon fraction was measured (ALS Analytics, Tucson) for seven samples of Kalimantan peat from sites ranging from heavy to no disturbance and averaged 0.579 ± 0.025 (Stockwell et al., 2014). EFs are proportional to assumed carbon content, making future adjustments to evolving literature-average EFs trivial if warranted based on additional carbon content measurements. The denominator of the last term in Eq. (1) estimates total carbon. For nearly all the plumes, the mass ratio of EC and OC to the simultaneous co-located CO, measured by the FTIR (see below), was added to the estimate of total carbon. Thus, our total carbon estimate for the grab samples includes all the gases measured by the FTIR or WAS in grab samples of a source and the carbon in the aerosol measured on the filters. Ignoring the carbon emissions not included or not measurable by our suite of instrumentation (typically higher molecular weight oxygenated organic gases) likely inflates the EF estimates by less than ∼ 1–2 % (Yokelson et al., 2013; Stockwell et al., 2015), which is small compared to the 4 % uncertainty due to natural variability in peat carbon content.

Biomass fire emissions vary naturally as the mix of combustion processes varies. The relative amount of smoldering and flaming combustion during a fire can be roughly estimated from the modified combustion efficiency (MCE). MCE is defined as the ratio ΔCO2 / (ΔCO2+ΔCO) and is mathematically equivalent to 1/(1+ΔCO / ΔCO2) (Yokelson et al., 1996). Flaming and smoldering combustion often occur simultaneously during biomass fires, but a very high MCE (∼ 0.99) designates nearly pure flaming (more complete oxidation) while a lower MCE (∼ 0.75–0.84 for biomass fuels) designates pure smoldering. Plume-average MCE was computed for all plumes using the plume-average ΔCO / ΔCO2 ratio as above. In the context of biomass or other solid fuels, smoldering refers to a mix of solid-fuel pyrolysis (producing NMOG and OA) and gasification (producing mainly NH3, CH4, and inorganic gases with little visible aerosol) (Yokelson et al., 1997).

The time-integrated excess Babs and Bscat from the PAXs were used to directly calculate the plume-average SSA (defined as Bscat/(Bscat+Babs)) at both 870 and 405 nm for each source. The PAX time-integrated excess Babs at 870 and 405 was used directly to calculate each plume-average AAE (Eq. 2). AAE=-log⁡Babs,1Babs,2log⁡λ1λ2 Aerosol absorption is a key parameter in climate models; however, inferring absorption from total attenuation of light by particles trapped on a filter or from the assumed optical properties of a mass measured by thermal/optical processing, incandescence, etc. can sometimes suffer from artifacts (Andreae and Gelencsér, 2006; Subramanian et al., 2007). In the PAX, the 870 nm laser is absorbed in situ by black carbon containing particles only, without filter or filter-loading effects that can be difficult to correct. We directly measured aerosol absorption (Babs, Mm-1) and used the literature-recommended MAC (4.74 m2 g-1 at 870 nm) to estimate the BC concentration (µg m-3) (Bond and Bergstrom, 2006). The PAXs (and filters) were co-sampled with the FTIR measuring CO2, CO, and CH4 in real time. The mass ratio of the integrated excess BC in the plume measured on the PAX to the integrated excess CO measured by the FTIR was multiplied by the EF CO based on the real-time FTIR data to determine EFs for BC (g kg-1). Note the total C for the carbon mass balance for the EFs calculated for real-time data is based on the integrated excess amounts of just the three main gases and aerosol carbon, which will inflate the EFs by a small amount (typically 1–3 %) compared to the larger suite of gases used for the grab sample calculations.

To a good approximation, sp2-hybridized carbon (i.e., BC) has an AAE of 1.0 ± 0.2 and absorbs light proportional to frequency. Thus, Babs due only to BC at 405 nm would be expected to equal 2.148 × Babs at 870 nm and we assumed that excess absorption at 405 nm, above the projected amount, is associated with BrC absorption. This method of attributing BrC absorption is based on several assumptions discussed in detail elsewhere that are likely most valid in cases where the BrC absorption is dominant such as in these peat fire smoke plumes (Lack and Langridge, 2013). In theory, a BrC concentration (µg m-3) could be calculated using a literature-recommended BrC MAC of 0.98 m2 g-1 at 404 nm (Lack and Langridge, 2013). The BrC mass calculated this way would be intended to be roughly equivalent to the total OA mass, which as a whole weakly absorbs UV light, and not the mass of the actual chromophores. However, the MAC of Lack and Langridge (2013) is appropriate for more typical biomass burning with a mix of flaming and smoldering, whereas the peat aerosol is overwhelmingly organic and at low BC / OA ratios the MAC is much smaller (Saleh et al., 2014; Olson et al., 2015). Thus, instead we divided the Babs at 405 nm assigned to BrC by the co-measured OC mass to estimate the peat smoke MAC referenced to bulk OC. The EFs for scattering and absorption at 870 and 405 nm (EF Babs, EF Bscat) are reported directly in units of m2 per kg of dry fuel burned by multiplying the ratios of Babs and Bscat to co-measured real-time CO by the real-time EF CO. We note that most of the related measurements of elemental and organic carbon on the filters will be discussed separately by Jayarathne et al. (2016).

In general, we found very high correlation in the ER plots indicating the plumes were well-mixed and implying low uncertainty in the individual plume EFs. Figure 1 shows a selection of such plots for plume N and it is also seen that the smoke mixing ratios were far above background. This experiment was not well-designed for comparison, but we have noted excellent WAS/FTIR agreement previously under more rigorous, but drier, conditions (e.g., Christian et al., 2003; Hatch et al., 2016) and we found that these 2015 field WAS results compared well with online measurements during FLAME-4 peat fire sampling for many major species, as discussed later in the paper.

Table S2 presents all the trace gas EFs for all 35 plumes sampled while Table 1 shows all our study-average EFs and 1 standard deviation of the means for all the gases that were significantly elevated in the smoke plumes. In the pure peat combustion that we were able to sample, the major trace gas emissions by mass (EF ≳ 0.5 g kg-1) were carbon dioxide (1564 ± 77), carbon monoxide (291 ± 49), methane (9.51 ± 4.74), hydrogen cyanide (5.75 ± 1.60), acetic acid (3.89 ± 1.65), ammonia (2.86 ± 1.00), methanol (2.14 ± 1.22), ethane (1.52 ± 0.66), dihydrogen (1.22 ± 1.01), propylene (1.07 ± 0.53), propane (0.989 ± 0.644), ethylene (0.961 ± 0.528), benzene (0.954 ± 0.394), formaldehyde (0.867 ± 0.479), hydroxyacetone (0.860 ± 0.433), furan (0.772 ± 0.035), acetaldehyde (0.697 ± 0.460), and acetone (0.691 ± 0.356). These results are shown in a bar chart in Fig. 2. C6–C10 alkanes summed to 0.87 ± 0.57 g kg-1, roughly consistent with the 0.59 g kg-1 of C6–C10 alkanes emitted by a peat fire sampled by two-dimensional GC in the FLAME-4 lab study (Hatch et al., 2015). Hatch et al. (2015) also measured 0.43 g kg-1 of C11–C15 alkanes, which is probably a reasonable estimate for our field fires. The larger alkanes (> C10) are efficient OA precursors (Presto et al., 2010). BTEX (benzene, toluene, ethylbenzene, xylenes) compounds are also high-yield OA precursors (Wang et al., 2014) and important air toxics; they were emitted in total at 1.49 ± 0.64 g kg-1. Air toxics are discussed further in Sect. 3.5.2 with the FLAME-4 lab data included. Additional discussion of NMOG emissions and detailed comparison with previous (e.g., FLAME-4) trace gas measurements on lab peat fires is presented in Sect. 3.5.1.

The MCE of the smoke sources ranged from 0.693 to 0.835 with an average of 0.772 ± 0.035 (n = 35), indicating essentially pure smoldering combustion. For most biomass fires there is both flaming and smoldering combustion and EFs for flaming compounds are observed to correlate with MCE while EFs for smoldering compounds (most NMOGs) tend to be anticorrelated with MCE (Burling et al., 2011). However, these fires burned by smoldering only with no high MCE values (e.g., > 0.9) and little or no dependence of EFs on MCE was observed.

Figure 3 shows an example of the PAX real-time Babs at 870 and 405 nm collected on 5 November along with the co-located CO data. Note the scaling of the axes and the dominance of Babs at 405 nm, though the ratio of 870/405 is seen to increase towards the end of the sampling period (the traces are slightly offset so that the background trace is visible). The excess values above background that were used to calculate all the quantities described above had a similar excellent signal to noise ratio in all cases. Babs at 405 and CO remain correlated, but the ratio of Babs at 405 to CO decreases towards the end of the 5 November data, which is consistent with an increase in the glowing / pyrolysis ratio (Yokelson et al., 1997). Variation in the mix of these smoldering processes likely causes some of the variation in EFs.

Table 2 shows all PAX-measured quantities, the MCE from the co-sampled real-time FTIR data, and the small subset of filter EC, OC, and PM2.5 data that were co-sampled with the PAXs for all seven plumes along with the study averages and standard deviations. Consistent with the lack of flaming, the emissions of BC were negligible (0.0055 ± 0.0016 g kg-1) (Christian et al., 2003; Liu et al., 2014). Aerosol absorption at 405 nm was 52 times larger than at 870 nm and BrC contributed an estimated 96 % of the absorption at 405 nm. Average AAE was 4.97 ± 0.65 (range 4.29–6.23). The SSA at 405 nm (0.974 ± 0.016, range 0.941–0.989) was marginally lower than SSA at 870 nm (0.998 ± 0.001, range 0.997–0.999). Clearly, estimating aerosol absorption from BC measurements alone would be inadequate for this source.

Pure pyrolysis has lower MCE than glowing and, thus, pyrolysis is implicated as the source of BrC via the correlation of AAE with lower MCE (r2= 0.65) (Fig. 4a). We note the data cover a small MCE range and thus the relationship shown is not well constrained for extrapolation much beyond the range shown. We also find that AAE correlates strongly with SSA at 405 nm (Fig. 4b). In this case, the trend line shown is likely illustrative of peat fire aerosol but, again, not suitable for extrapolation to other fuels or beyond the range shown.

By plotting EF Bscat vs. EF PM2.5 for all seven plumes sampled by PAX and filters (Fig. 5) we get a rough estimate of the mass-scattering efficiency (MSE) of the peat fire aerosol at 405 nm based on the slope of 2.96 ± 0.67 m2 g-1 (r2= 0.80). The plot compared EFs measured in the same plumes, but in some cases at slightly different times due to a PAX auto-zero or a filter clogging. If we restrict the plot to the four plumes where the timing of the sampling was identical, the slope is 3.05 m2 g-1 (r2= 0.81). Either value of the MSE is close to MSEs obtained at illumination wavelengths in the range 532–550 nm (3–5 m2 g-1) in other studies of BB aerosol with lower values characteristic of fresher smoke (Tangren, 1982; Patterson and McMahon, 1984; Nance et al., 1993; Burling et al., 2011). However, based on average BB aerosol size distributions (Reid et al., 2005), our MSE may be underestimated on the order of 5–10 % due to the difference in sampling cutoffs (2.5 microns for filters and 1.0 μ for PAX). By comparing the EF Babs at 405 nm assigned to BrC with EF OC from the filters on the same plumes (Fig. 6) we can estimate the MAC of the bulk OC. As above, two MAC estimates are possible. Using the mean value for all seven plumes we get 0.09 ± 0.08 m2 g-1 where the large coefficient of variation is due to one larger MAC value near 0.27 m2 g-1. Keeping the dynamic nature of the emissions chemistry shown in Fig. 3 in mind, if we restrict our analysis to the same four plumes where sample timing was identical (but different size cutoffs; blue points in Fig. 6) and plot EF Babs-405 vs. EF OC we get a slope of 0.071 ± 0.03 m2 g-1. The MACs obtained either way are similar but again underestimated by a few percent due to cutoff differences and much smaller than MACs for average biomass burning OA (0.98; Lack and Langridge, 2013). However, we confirm the expected MAC near 0.1 m2 g-1 for the extremely low BC (or EC) to OA ratio in the aerosol (Saleh et al., 2014; Olson et al., 2015).

While EC and BC are considered approximately equivalent for some combustion sources (e.g., diesel fuel combustion), our EF EC for peat fires is noticeably larger than the EF BC although both the EC and BC values are very small (Table 2) compared to typical values for combustion aerosol. This is the expected result in this case for several reasons. The peat smoke plumes sampled outdoors likely contain very small amounts of soot from rare instances of flaming and also a small amount of entrained small char particles produced by pyrolysis of the peat on site by the glowing combustion front (Santín et al., 2016). Both soot and char are detected to some extent as EC (Andreae and Gelencsér, 2006; Han et al., 2007, 2010, 2016) and our EC subfractions evolving at lower temperatures confirm some char was present (NIOSH, 1999). The char particles tend to be larger (1–100 microns; Han et al., 2010) and would be more efficiently sampled by the filters, which had a 2.5 micron cutoff as opposed to the PAX with a 1.0 micron cutoff. Char tends to absorb long wavelengths less efficiently than soot (Han et al., 2010) and the PAX would therefore be relatively insensitive to any sampled char for this reason also. The accuracy of both the PAX BC and the thermal optical EC detection is challenged by the low EC or BC to OC ratio (Andreae and Gelencsér, 2006). However, both measurements are useful and point to the same key results: the aerosol is overwhelmingly organic and the organic fraction contributes most of the light absorption.

In a previous study of aerosol emissions from burning Sumatran peat in a lab setting, Christian et al. (2003) measured an EF for OC + EC by the thermal optical technique of ∼ 6 g kg-1 that had OC / EC of 151. More extensive comparison of our field PM2.5, EC, and OC data with lab measurements, including the FLAME-4 EC / OC data, will be presented in Jayarathne et al. (2016).

Turning to optical properties, Liu et al. (2014) reported some SSA values and the AAE for smoldering Kalimantan peat (Fire 114) from FLAME-4: MCE (0.74), AAE (6.06), SSA-405 (0.94), and SSA-781 (1.00). These are very consistent with our data (Table 2) and especially with our lowest MCE field sample: MCE (0.726), AAE (6.23), SSA-405 (0.941), and SSA-870 (0.997). They also report data for a FLAME-4 peat fire with some brief flaming (Fire 154) and obtain for example an AAE of 3.02, which is below our lowest AAE of 4.28. Their average AAE 4.45 ± 2.19 for Indonesian peat is not significantly smaller than ours (4.97 ± 0.65) and it should be kept in mind that the determination and comparisons of AAE can be affected by the use of different wavelength pairs (Lewis et al., 2008; Chakbarty et al., 2016). BrC absorption is very small at both 781 and 870 nm so the high SSA at the long wavelengths in both studies and similar AAEs are consistent with minimal BC absorption and dominant absorption by BrC. In summary, when comparing to published laboratory studies of tropical peat burning, especially for smoldering combustion in the lab, we get good agreement in the sense of extremely low EC or BC to OC ratios and for the aerosol optical properties.

The biomass of the surface layer in logged/disturbed peatlands is small compared to the peat, and even the biomass of intact peat-swamp forest is small compared to peat loading as noted by Page et al. (2002). However, peat is only one component of the total peatland fuel and potentially a diminishing component as exploitation and repeated fires are continued over many years (Konecny et al., 2016). As the peat fuels are consumed on a site, the loading of surface fuels likely also decreases. We did not see much evidence of active surface fuel combustion, but our sampling was just after the peak regional PM10 levels, which may have had a larger contribution from surface fuels. Numerous “hotspots” were detected in the region and both flaming and smoldering were evident in the news media coverage (https://worldview.earthdata.nasa.gov/). The fraction of total annual regional emissions due to emissions generated during the peak regional impacts is difficult to estimate since a long period of moderately elevated emissions could produce as much or more emissions as a shorter, higher level of emissions. The overall mix of fuels burning in the region during the peak regional pollution would have been hard to assess in any case since visibility dropped to ∼ 10 m, making driving dangerous and even a regional fire survey with an aircraft problematic. Further, surface fuel emissions would likely be associated with some amount of flaming combustion that would be hard to sample properly with most ground-based instruments. Finally, under extremely polluted conditions it is hard to acquire background samples or isolate and measure individual fuel contributions/EFs so that the variable relative contributions of peat and surface fuels (primary and secondary forest, cropland, grassland, etc.) can be explicitly modeled on a regional scale. Our sampling, somewhat fortuitously, unambiguously probed the emissions from the major fuel component, peat, of special concern in Southeast Asia.

Our sampling was also near the end of the fire season when the relative amount of total annual deep burning vs. total annual surface burning could potentially be measured (an earlier assessment would underestimate the deep peat burning). We sampled and observed areas with peat burning at depths from 18 to 60 cm. However, we also accessed our sites at times across areas that had recently burned with consumption of some surface fuels, but with only shallow consumption of the organic soil layer. Thus, applying an average peat burn depth for all burned area from our sampled burn depths would be biased high and a better estimate of the average burn depth will likely result from the lidar data collected. However, burned area is likely underestimated in inventories since they rely on remote sensing data that miss some of the hotspots and burned area used in bottom-up estimates, as well as some of the fire products (e.g., CO, aerosol) used in top-down approaches. The information gap is caused by high regional cloud cover; orbital gaps; rapid growth of new vegetation, which is strongly associated with shallow burn depth (Cypert, 1961; Kotze, 2013); and other factors (Lu and Sokolik, 2013; Reddington et al., 2016; Reid et al., 2013). Thus, overestimating burn depth and underestimating burned area tends to cancel when coupling these terms to estimate fuel consumption. A 2015 airborne campaign surveying regional smoke could have theoretically assessed the overall regional smoke characteristics, but this did not occur. With the caveat that fire use has evolved in Kalimantan over the years, we can compare to airborne atmospheric chemistry measurements conducted during the 1997 El Niño haze event, as detailed next.

We now compare our ground-based measurements of “pure” peat smoke to the only available airborne regional smoke measurements, which were part of the Pacific Atmospheric Chemistry Experiment 5 (PACE-5) campaign in Kalimantan during the peak of another El Niño event (Sawa et al., 1999). During late October 1997, airborne sampling was conducted west of Banjarmasin along a flight leg several hundred kilometers long at four flight levels between 1.3 and 4.4 km altitude. The flight was ∼ 100 km south of Palangkaraya and encountered 3–9 ppm of CO and ∼ 500 m visibility at lower altitudes (Sawa et al., 1999). Gras et al. (1999) noted that no visible flame fronts were observed from the aircraft and estimated one SSA for a Kalimantan smoke plume as 0.98. We can estimate an SSA at 530 nm by linear interpolation between 870 and 405 nm and obtain a similar value (0.981). They measured large hygroscopic growth factors of 1.65 which agreed well with tests of peat combustion they cite by Golitsyn et al. (1988). From the same flight Sawa et al. (1999) reported NOx / CO ERs of 0.00019 to 0.00045, which they attributed to a lack of flaming combustion, but also possibly faster losses of NOx than CO. We observed several individual values in their range (our minimum was 0.00028), but our average NOx / CO ER is higher (0.0012 ± 0.0007). The comparison is good in that the ranges overlap and are consistent with smoldering combustion, but some fast NOx losses probably also impacted the airborne ERs. The PACE-5 team speculated that high SO2 emissions could contribute to the hygroscopicity and cited unpublished lab tests that confirmed high SO2 from burning peat. We did not see evidence of elevated SO2, but our measurements were conducted further inland, possibly away from Holocene coastal sulfidic sediments invoked by Gras et al. (1999) as a possible source of SO2. During FLAME-4, no SO2 was detected from burning peat in the lab except for the one sample of coastal peat which was collected in North Carolina (Table S2 in Stockwell et al., 2015). This suggests that the emissions from burning coastal peat deposits are impacted by their known chemical differences (Cohen and Stack, 1996).

Hamada et al. (2013) measured CO2, CO, and CH4 emissions from a peat fire near Palangkaraya during the 2009 El Niño. Based on 23 samples, they report CO / CO2 and CH4 / CO2 ERs of 0.382 and 0.0261, which are 31 and 56 % higher than our study averages, respectively, but within our range for individual plume averages. Their data are consistent with a smoldering-dominated burn and an MCE of 0.724, which is within our range for individual fires; one of ours was lower (0.693), though our study average was higher (0.772 ± 0.035).

Two very recent studies probed peat fire emissions during the 2015 El Niño. Huijnen et al. (2016) measured three EFs for peat fires also near Palangkaraya. Their “peat-only” EFs are 255 ± 39, 1594 ± 61, and 7.4 ± 2.3 g kg-1 for CO, CO2 and CH4, respectively. Their means are all within 1 standard deviation of our means and their EFs are within +1.9, -13, and -22 % of ours, respectively. Not many details of the measurements are given, but the agreement is good. Parker et al. (2016) report three space-based measurements of the ER for Kalimantan fires in September–October 2015 for CH4 / CO2 ranging from 0.0062 to 0.0136. This is lower on average than the CH4 / CO2 ERs reported for peat combustion in the in situ studies cited above (range ∼ 0.011–0.035). The difference is consistent with our expectation noted above that some flaming-dominated consumption of surface fuels likely contributed to regional emissions in 2015. However, a glance at Fig. 6 in Parker et al. (2016) shows that some of highest retrieved levels of these gases, which they attribute to fires, are far offshore and/or upwind of the fires. Thus, more evaluation is clearly needed to determine whether space-based approaches can accurately measure CH4 / CO2 ERs (e.g., Agustí-Panareda et al., 2016).

The basic application of EFs is to multiply them by a total fuel consumption to generate total emissions for a desired region (Seiler and Crutzen, 1980). Our EFs in this work are intended for use with peat consumption estimates to calculate total emissions from the peat component. Major uncertainties would include natural variation of the EFs (e.g., the standard deviations of the EFs given in Table 2) and variation in %C, density, and burn depth of the peat. Konecny et al. (2016) list some other %C and burn depth measurements, which are generally close to our values. We plan to present further data on these issues in a separate paper. We note that in a previous review of BB EFs, Akagi et al. (2011) estimated literature average values for EFs for pure peat. Following Page et al. (2002) they also computed “peatland” EFs by combining the peat EFs and fuel consumption with EFs and fuel consumption for tropical peat-swamp forest, which was considered as the only surface fuel type. This was potentially appropriate for 1997. However, given ongoing land-use trajectories, it is now clear that many different types of surface fuels and a variety of fuel combinations are important (Miettinen et al., 2016). The work here presents EFs specific for the major peat component that can be coupled with peat fuel consumption estimates and that ideally contribute to emissions estimates after combining with fuel consumption estimates and EFs for the relevant surface fuel types. Many of the EFs and fuel consumption values for other surface fuel types are tabulated in Akagi et al. (2011). Another earlier set of trace gas EF previously available for tropical peat burning was from a laboratory study (Christian et al., 2003) and was also adopted in IPCC guidelines (Table 2.7 in IPCC, 2014). We suggest our new and more extensive field-measured values are more appropriate and that this involves significant adjustments for the EFs for most gases compared to the 2003 study, notably CO2 (-8 %), CH4 (-55 %), NH3 (-86 %), and CO (+39 %). Improved EFs, at least for Kalimantan, for numerous other gases are found in Table 2. Finally, this work also provides previously unavailable field measurements of aerosol optical properties. Both the aerosol and trace gas data in this study should be used with the understanding that many quantities will be affected by smoke evolution (e.g., Hobbs et al., 2003; Abel et al., 2003; Yokelson et al., 2009; Akagi et al., 2012; Alvarado et al., 2015).

In this section we explore combining our new field data with the FLAME-4 lab data to develop an even more comprehensive set of EFs for the peat component of peatland fires.