Convergent evidence for the pervasive but limited contribution of biomass burning to atmospheric ammonia in peninsular Southeast Asia

Ammonia (NH₃) is an important agent involved in atmospheric chemistry and nitrogen cycling. Current estimates of NH₃ emissions from biomass burning (BB) differ by more than a factor of 2, impeding a reliable assessment of their environmental consequences. Combining high-resolution satellite observations of NH₃ columns with network measurements of the concentration and stable nitrogen isotope composition (δ¹⁵N) of NH₃, we present coherent estimates of the amount of NH₃ derived from BB in the heartland of Southeast Asia, a tropical monsoon environment. Our results reveal a strong variability in atmospheric NH₃ levels in time and space across different landscapes. All of the evidence on hand suggests that anthropogenic activities are the most important modulating control with respect to the observed patterns of NH₃ distribution in the study area. N-isotope balance considerations revealed that during the intensive fire period, the atmospheric input from BB accounts for no more than 21±5 % (1σ) of the ambient NH₃, even at the rural sites and in the proximity of burning areas. Our N-isotope-based assessment of the variation in the relative contribution of BB-derived NH₃ is further validated independently through the measurements of particulate K⁺, a chemical tracer of BB. Our findings underscore that BB-induced NH₃ emissions in tropical monsoon environments can be much lower than previously anticipated, with important implications for future modeling studies to better constrain the climate and air quality effects of wildfires.

1 Introduction

Biomass burning (BB) in tropical vegetation regions due to wildfires has been recognized as a globally important source of trace gases (including CO₂, CO and ozone precursors) and aerosols (mostly black and organic carbon) (Crutzen and Andreae, 1990; Andreae and Merlet, 2001; Shi et al., 2015; Andreae, 2019; Crutzen et al., 1979). Most BB hot spots occur in West Africa and South America (Crutzen and Andreae, 1990; van der Werf et al., 2006; Shi et al., 2015), but recent studies have also highlighted the importance of Southeast (SE) Asia in this regard, mainly because of the much higher population densities near intensive fire burning areas (Huang et al., 2013; Marlier et al., 2013; Lee et al., 2017; Betha et al., 2014). The climate over large parts of SE Asia is governed by a wet (typically May–July) and dry (typically February–April) season caused by seasonal shifts in the monsoon winds. During the dry season, dry plant materials (e.g., forest, peatland, banana leaf) readily ignite, resulting in large wildfires that can markedly modify the atmospheric composition in the tropics, whereas the tropical rain belt causes plentiful rainfall during summer, preventing such fires during the rainy season (Lee et al., 2017; Chu et al., 2018).

Besides carbon soot, BB also emits large amounts of reactive nitrogen compounds (Lobert et al., 1990; Bauters et al., 2018), in particular ammonia (NH₃), which is believed to represent the major source of NH₃ during intensive fire periods (Akagi et al., 2011; Whitburn et al., 2015). However, these emissions are subject to large uncertainties (differences of a factor of 2 or greater) (Bray et al., 2018; Whitburn et al., 2015, 2016b; Van Damme et al., 2015b) For example, BB is probably the second most important NH₃ source after agriculture, contributing 11 %–23 % of the global burden (Paulot et al., 2017; Bouwman et al., 1997). Minor NH₃ sources include fossil fuel burning and biogenic activity (Chang et al., 2012, 2016b, 2019b, 2020). A recent paper also highlighted the underestimated importance of industrial emissions (Van Damme et al., 2018). Once emitted in the atmosphere, NH₃ is rapidly removed by dry or wet deposition (Asman et al., 1998). Excess NH₃ is known to be responsible for several environmental issues: eutrophication of terrestrial and aquatic ecosystem, soil acidification, and loss of plant diversity (Sutton et al., 2008, 2011; Aneja et al., 2008). In the atmosphere, NH₃ can neutralize acid gases (mostly sulfuric acid, nitric acid or hydrochloric acid), resulting in the formation of secondary aerosols that in turn negatively affect climate and human health (Wang et al., 2011, 2013; Paulot and Jacob, 2014; Souri et al., 2017).

To assess the environmental impacts of BB (e.g., air quality and climate change), atmospheric chemistry models incorporating BB-related emissions have widely been used over the past decades (Huang et al., 2013; Aouizerats et al., 2015; Wang et al., 2011, 2013; Souri et al., 2017), but these models are afflicted with a relatively large uncertainty regarding the input parameters used (Hantson et al., 2016; Whitburn et al., 2015; Paulot et al., 2017). The uncertainties, for example, for carbon emissions and for other trace gases (including NH₃), can be over 200 % (Whitburn et al., 2015; Paulot et al., 2017; Zhu et al., 2013; Pan et al., 2020). In recent years, hyperspectral sounders on board satellites have demonstrated their capabilities to directly measure tropospheric column concentrations of NH₃ gas (Van Damme et al., 2014, 2015b, 2018; Clarisse et al., 2009). Therefore, satellite observations offer a “top-down” alternative to the bottom-up estimates. However, the biggest challenge of satellite-based NH₃ assessments is the requirement for the atmosphere to be cloud-free during observations as well as the need for a sizable temperature difference between the land or sea surface and the atmosphere (Van Damme et al., 2015a; Whitburn et al., 2015; Martin, 2008; Streets et al., 2013; Clarisse et al., 2010).

Large uncertainties remain regarding global or regional atmospheric budgets of NH₃, and the attribution of emissions to specific sources, emphasizing the need for independent verification methods. An impressive body of previous work has studied the BB influence on the concentration and composition of aerosols in SE Asia (Betha et al., 2014; Aouizerats et al., 2015; Lee et al., 2017; Bikkina et al., 2019). However, to our knowledge, there are no reports on the detailed spatiotemporal patterns of the atmospheric NH₃ concentration and nitrogen isotopic composition (δ¹⁵N–NH₃) associated with BB in this region. Due to isotopic fractionation associated with NH₃ production, pyrogenic NH₃ displays a distinctly higher δ¹⁵N–NH₃ (δ¹⁵N defined as $(R_{\mathrm{sample}}/R_{\mathrm{standard}}-\mathrm{1})\times \mathrm{1000}$, where R refers to the ${}^{\mathrm{15}}\mathrm{N}{/}^{\mathrm{14}}\mathrm{N}$ ratio in a sample or a standard) than temperature-dependent volatilized sources (Felix et al., 2013; Chang et al., 2016a). The N isotopic analysis of ambient NH₃ has been proven to be a useful tool to constrain sources of NH₃ emissions in the atmosphere, where both natural and anthropogenic activities are relevant (Chang et al., 2019a, b; Elliott et al., 2019). Here, we integrate high-resolution satellite observations with discrete NH₃ concentration measurements and δ¹⁵N–NH₃ data obtained from a regional passive monitoring network during and after the dry season of large-scale forest fires in the mountain areas of northern Thailand, SE Asia.

2 Methods

2.1 Site description

Surrounded by the mountain ranges of the northern Thailand highlands, the Chiang Mai Province covers an area of approximately 20 107 km², with a total population of over 1.7 million. Chiang Mai is characterized by a tropical monsoon climate, tempered by the low latitude and moderate elevation, with warm to hot weather year-round. Some 70 % of the area is covered by forests, and 13.4 % of the area is used for agriculture. A continuing environmental issue in Chiang Mai is smoke pollution from wildfires that primarily occur every year towards the end of the dry season between February and April (Tsai et al., 2013) before the relatively cool and rainy season from May on. During the period from March to July 2018, ambient NH₃ concentrations and δ¹⁵N–NH₃ values were determined at nine monitoring stations across the Chiang Mai Province. Figure 1 illustrates the location of sampling sites (with the different land use regimes indicated), Fig. S1 in the Supplement reports meteorological data for Chiang Mai and Table S1 in the Supplement details the information of each station.

Figure 1 Location of sampling sites: a land cover map (left; revised from Chang et al., 2016b) of the area, and zoomed in sections (right) showing the sampling sites. The chosen sampling sites are representative of a gradient in land use from urban to rural. The images on the right were obtained from © Google Maps.

2.2 Sampling and laboratory analysis

In order to obtain information regarding the spatial and temporal variability in NH₃ concentrations over Chiang Mai, ambient gas-phase NH₃ concentrations at each site were collected weekly using passive sampling devices (PSDs; ALPHA – Adapted Low-cost, Passive High Absorption; Centre for Ecology and Hydrology, Edinburgh, UK) (Chang et al., 2016a). The ALPHA PSD is a circular polyethylene vial (26 mm height, 27 mm diameter) with one open end. The vial holds a 25 mm phosphorous acid-impregnated filter and a polytetrafluoroethylene (PTFE) membrane for gaseous NH₃ diffusion. These PSDs have been widely used in Europe, China and the US, and they are capable of detecting NH₃ concentrations as low as 0.03 µg m⁻³ (Chang et al., 2016a; Puchalski et al., 2011; Liu et al., 2013; Tang et al., 2018). In the laboratory, the ALPHA filter samples were soaked in 10 mL deionized water (18 MΩ cm⁻¹) in a 15 mL vial for 30 min with occasional shaking. Concentrations of NH₃-derived ${\mathrm{NH}}_{\mathrm{4}}^{+}$ in extracts were determined using a Dionex^TM ICS-5000⁺ system (Thermo Fisher Scientific, Sunnyvale, USA). The IC (ion chromatograph) system was equipped with an automated sampler (AS-DV), an IonPac CG12A guard column and a CS12A separation column. Aqueous methanesulfonic acid (MSA, 30 mM L⁻¹) served as eluent at a flow rate of 1 mL min⁻¹. The isotopic analysis of the extracted ${\mathrm{NH}}_{\mathrm{4}}^{+}$ was based on the isotopic analysis of nitrous oxides (N₂O) after chemical conversion (Liu et al., 2014). More precisely, dissolved ${\mathrm{NH}}_{\mathrm{4}}^{+}$ in deionized water (DIW) extracts was oxidized to ${\mathrm{NO}}_{\mathrm{2}}^{-}$ by alkaline hypobromite (BrO⁻) and then reduced to N₂O by hydroxylamine hydrochloride (NH₂OH ⋅ HCl). The produced N₂O was analyzed using a purge and cryogenic trap system (Gilson GX-271, IsoPrime Ltd., Cheadle Hulme, UK) coupled to an isotope ratio mass spectrometer (PT-IRMS; IsoPrime 100, IsoPrime Ltd., Cheadle Hulme, UK) (Liu et al., 2014). In order to correct for any machine drift and procedural blank contribution, international ${\mathrm{NH}}_{\mathrm{4}}^{+}$ (IAEA N1, USGS 25 and USGS 26) standards were processed in the same way as samples (Liu et al., 2014). The analytical precision for N isotope analyses was better than 0.5 ‰ (n=5).

2.3 Isotope-based source apportionment

Isotopic mixing models represent valuable tools to estimate the fractional contributions of multiple sources (emission sources of NH₃ in this study) within a mixture (the ambient NH₃ in this study) (Layman et al., 2012). By explicitly considering the uncertainties associated with the isotopic signatures of single sources and the N isotope fractionation during transformations, the application of Bayesian methods to stable isotope mixing models yields robust probability estimates of source apportionments, and its application to natural systems is more appropriate than the application of simple linear mixing models (Parnell et al., 2010). Here, a novel Bayesian approach using a mixing model, implemented in the SIAR (Stable Isotope Analysis in R) software package, was used to resolve multiple NH₃ source categories by generating potential solutions of source apportionment as true probability distributions of the single source contribution to the total NH₃ pool. The generation of such source contribution probability distributions allows for the estimation of likelihood ranges of source contributions even under under-constrained conditions (i.e., the number of potential sources exceeds the number of different isotope system parameters + 1). The SIAR package is available for download from the packages section of the Comprehensive R Archive Network site (CRAN; http://cran.r-project.org/, last access: 10 May 2021), which has been widely applied in a number of fields (Chang et al., 2019a, b). The model frame and computing methods are detailed in Sect. S1 in the Supplement.

Figure 2 (a) Monthly (March–June) spatial distributions of the NH₃ total columns (molec. cm⁻²) in 2018 obtained from satellite measurements by the Infrared Atmospheric Sounding Interferometer (IASI)/MetOp-A instrument in $\mathrm{0.25}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ cells. (b) Monthly distributions of gridded counts of active fire pixels ($\mathrm{0.25}{}^{\circ }\times \mathrm{0.25}{}^{\circ }$ per cell) derived from the 375 m Visible Infrared Imaging Radiometer Suite (VIIRS). The red and black squares indicate our study area in Chiang Mai. (c) Daily variations in temperature (^∘C) and rainfall (mm) in Chiang Mai city.

2.4 Satellite observations of ammonia and fires

NH₃ total columns (molec. cm⁻²) are retrieved from the Infrared Atmospheric Sounding Interferometer (IASI) observations. The IASI instruments are on board the MetOp satellite series; in this work, we use IASI/MetOp-A (launched in 2006) and IASI/MetOp-B (launched in 2012) data. Both instruments have an overpass time of around 09:30 and 21:30 LST (local solar time when crossing the Equator) and, therefore, provide in total a global coverage four times a day. The retrieval strategy, based on artificial neural networks, is fully detailed in previous work (Whitburn et al., 2016a; Van Damme et al., 2017). Here, we only consider morning observations, as they are more sensitive to the lower layer of the atmosphere. Fire radiative power (FRP) from the Moderate Resolution Imaging Spectroradiometer (MODIS) and fire counts derived from the 375 m Visible Infrared Imaging Radiometer Suite (VIIRS) are also used (Li et al., 2020).

Figure 3 (a) The MODIS FRP (fire radiative power; the size of red dots is proportional to arbitrary FRP values) and (b) the IASI MetOp-A and MetOp-B averaged NH₃ distribution (molec. cm⁻²) for 2018. (c) Number of people per grid cell in the Chiang Mai area in 2018 at a resolution of 3 arcmin.

3 Results and discussion

3.1 Satellite-observed NH₃ distributions

Figure 2a illustrates the monthly spatial distribution of NH₃ columns obtained from IASI in 2018 at a spatial resolution of $\mathrm{0.25}{}^{\circ }\times \mathrm{0.5}{}^{\circ }$ cells. Our study area is set within a large domain of $\mathrm{5.00}{}^{\circ }\times \mathrm{3.25}{}^{\circ }$ (red and black rectangles in Fig. 2a and b, respectively), in which a total of 260 gridded pixels ($\mathrm{0.25}{}^{\circ }\times \mathrm{0.25}{}^{\circ }$ per cell) are used for dividing active fire points (Fig. 2b). Intriguingly, from this plot, one is tempted to conclude that fires do play a very important role in NH₃ emissions, as the NH₃ columns are much higher in March and April (dry season), which is coincident with a high number of monthly fire activities (indicated by the number of fire points). Further, using 11 years (2008–2018) of IASI satellite data, Fig. S2 presents a climatology of monthly NH₃ columns over Chiang Mai at a much finer spatial resolution, which also support the pervasive contribution of BB during dry season (March and April). Based on the average observed temporal distribution of satellite-constrained wildfires, the sampling period in this study can be divided into two contrasting fire-regime periods: the BB season (March and April) and the non-BB season (May and June). Interestingly, however, although the number of fire points in March (43 613 points) is significantly (p<0.01) higher than that in April (27 905 points) (Fig. 2b), the average NH₃ column in March is nearly the same as that in April (Fig. 2a). This implies that there is not a one-to-one relationship between BB and NH₃ emissions and, in turn, that other sources or factors (e.g., soil dryness, agricultural emissions, precipitation and temperature dependence) must also play a significant role.

Given that the average monthly temperature varies only slightly in contrast to the drastic change in rainfall during our study period (Fig. 2c), it is reasonable to assume that temperature-dependent NH₃ volatilization is not the main driver of changes in the NH₃ columns. The amount of rainfall, in contrast, can have a multifaceted impact on NH₃ emissions. Firstly, there is an obvious link between precipitation rates and the number of wildfires, and, if BB is a major NH₃ emission source, we can also expect a relationship between the NH₃ columns and monthly rainfall rates. Secondly, and maybe more importantly, rain will dissolve atmospheric particulate ${\mathrm{NH}}_{\mathrm{4}}^{+}$ and will act to clean the air of NH₃, which may partly explain the low NH₃ levels during May and June. On the other hand, comparison between March and April reveals higher NH₃ levels in April despite higher rain rates, suggesting that processes other than BB and rain-scavenging of BB-derived NH₃ must be relevant factors. In Fig. 3a and b, we superimposed the orography at the scale of the study area (Chiang Mai and surrounding mountains) onto the images of the year-long averaged MODIS FRP (fire radiative power) and IASI-NH₃ for 2018, respectively. Based on visual evaluation alone, it seems obvious that there is no strong correlation between fire intensity/number of fires and the observed IASI-NH₃, suggesting only limited influence of BB on NH₃. However, more strikingly, the IASI-NH₃ distribution matches that of the population density quite well (Fig. 3c). More precisely, hot spots of atmospheric NH₃ (Fig. 3b) appear to be concentrated in urban areas with a dense population. Hence, our satellite remote sensing observations suggest a significant influence of non-BB emissions on NH₃ concentrations, seemingly related to urban anthropogenic activities.

Figure 4 Temporal variations in measured NH₃ concentrations (µg m⁻³) between sites with different land use regimes. The error bar indicates 1 standard deviation.

3.2 Discrete concentration measurements confirm urban areas as hot spots of NH₃ emissions

A total of more than 150 samples were collected in this study for analyzing NH₃ concentrations (Fig. 4). The atmospheric NH₃ concentrations over Chiang Mai ranged from 2.5 to 46.4 µg m⁻³, with mean (±1σ) and median values of 14.5 (±9.2) and 11.4 µg m⁻³, respectively. Consistent with the IASI satellite-based NH₃ assessment, the weighted average NH₃ concentration (mean${}_{\min }^{\max }\pm \mathrm{1}\mathit{\sigma }$) during the dry season (i.e., when wildfires were markedly more abundant) was significantly (p<0.01) higher (20.6${}_{\mathrm{6.8}}^{\mathrm{46.4}}\pm \mathrm{9.8}$ µg m⁻³) than during the rainy season (10.2${}_{\mathrm{2.5}}^{\mathrm{31.9}}\pm \mathrm{5.7}$ µg m⁻³). Again, it is tempting to conclude that there is a direct link between higher atmospheric NH₃ levels and the higher number of BB events. However, there are several aspects that appear to speak against BB as the main, or only, driver of ambient NH₃ concentrations. Firstly, from a global perspective, the ambient NH₃ concentrations that we measured in northern Thailand are generally lower than in tropical regions with a dense population or intensive agricultural production (also see Fig. 2a) (Carmichael et al., 2003; Chang et al., 2016b). Secondly, within the study area, large spatial differences in NH₃ concentrations were found (Fig. 4). Yet, despite their proximity to wildfires at the time, the three rural sites always displayed the lowest NH₃ concentrations (8.3${}_{\mathrm{2.5}}^{\mathrm{26.8}}\pm \mathrm{4.6}$ µg m⁻³; Fig. 4; see detailed discussion in the next section).

Despite relatively large uncertainties, it is well accepted that, globally, atmospheric NH₃ is primarily emitted by agricultural activities and biomass burning (Asman et al., 1998; Bouwman et al., 1997). As a consequence, one would expect the NH₃ concentrations in the atmosphere over rural environments with lush vegetation and agricultural land use to be higher than those in (sub)urban areas, where agricultural activities are mostly absent. In our study, the average NH₃ concentrations at the nine respective sites are 19.5${}_{\mathrm{6.5}}^{\mathrm{39.4}}\pm \mathrm{9.5}$ µg m⁻³ (S1; suburban), 11.9${}_{\mathrm{4.5}}^{\mathrm{19.7}}\pm \mathrm{4.6}$ µg m⁻³ (S2; suburban), 8.8${}_{\mathrm{4.4}}^{\mathrm{16.6}}\pm \mathrm{3.9}$ µg m⁻³ (S3; rural), 9.0${}_{\mathrm{2.8}}^{\mathrm{26.8}}\pm \mathrm{5.8}$ µg m⁻³ (S4; rural), 7.0${}_{\mathrm{2.5}}^{\mathrm{13.7}}\pm \mathrm{3.8}$ µg m⁻³ (S5; rural), 20.2${}_{\mathrm{9.1}}^{\mathrm{40.5}}\pm \mathrm{8.6}$ µg m⁻³ (S6; urban traffic), 18.1${}_{\mathrm{7.2}}^{\mathrm{46.1}}\pm \mathrm{12.1}$ µg m⁻³ (S7; urban traffic), 19.6${}_{\mathrm{4.2}}^{\mathrm{46.4}}\pm \mathrm{10.1}$ µg m⁻³ (S8; urban) and 16.6${}_{\mathrm{6.7}}^{\mathrm{30.6}}\pm \mathrm{8.0}$ µg m⁻³ (S9; urban) (see also compilation in Fig. 4). Thus, against current paradigms, the observed NH₃ concentrations clearly reflect an urban (18.6${}_{\mathrm{4.2}}^{\mathrm{46.4}}\pm \mathrm{9.7}$, n=68) to suburban (${\mathrm{15.6}}_{\mathrm{4.5}}^{\mathrm{39.4}}\pm \mathrm{8.3}$, n=34) to rural (8.3${}_{\mathrm{2.5}}^{\mathrm{26.8}}\pm \mathrm{4.6}$, n=51) gradient. Such a concentration gradient can be taken as evidence that nonagricultural activities (including on-road traffic), at least in some regions, can outweigh agriculture and/or BB as the dominant NH₃ source in urban areas.

Indeed, a growing body of studies confirm that the urban atmosphere can be a hot spot of NH₃ release. Nonagricultural activities, such as wastewater treatment, coal combustion, solid garbage handling, vehicular exhaust and urban green space contribute strongly to urban to NH₃ emissions (Chang et al., 2016a, 2015, 2019b; Teng et al., 2017; Li et al., 2016; Sun et al., 2017). For example, high vehicular NH₃ emissions from three-way catalysts (TWCs) have been demonstrated in chassis dynamometer vehicle experiments, road tunnel tests and through ambient air measurements (Huang et al., 2018; Chang et al., 2016b, 2019b).

Figure 5 The left panel shows weekly variations in the δ¹⁵N values (‰) of ambient NH₃ measured in urban, suburban and rural environments (setting 0 as the breaking point). The error bars indicate 2 standard deviations. The right panels present box plots of the distribution of δ¹⁵N–NH₃ during the BB season and non-BB season for each type of sampling site.

3.3 N isotopic constraints on the sources of natural and anthropogenic NH₃

The correlative analysis of the spatiotemporal concentration patterns with the variations in land use effects provides the first qualitative constraints with respect to the relative importance of natural/BB and anthropogenic NH₃ emissions, but it is insufficient when a more quantitative assessment is required. The N-isotopic composition of NH₃ (i.e., δ¹⁵N–NH₃) can provide help in this regard, as it is sensitive to changes in NH₃ sources with distinct isotopic composition (Elliott et al., 2019; Felix et al., 2013). δ¹⁵N–NH₃ values determined in this study ($-{\mathrm{27.04}}_{-\mathrm{46.28}}^{-\mathrm{12.35}}\pm \mathrm{7.22}$ ‰, n=145) show a relatively large variability in time and space (Fig. 5). NH₃ emitted from the five major NH₃ sources displays distinct isotopic signatures (N-fertilizer application, $-\mathrm{50.0}\pm \mathrm{1.8}$ ‰; urban waste volatilized sources, $-\mathrm{37.8}\pm \mathrm{3.6}$ ‰; livestock breeding, $-\mathrm{29.1}\pm \mathrm{1.7}$ ‰; on-road traffic, $-\mathrm{12.0}\pm \mathrm{1.8}$ ‰; biomass burning, 12 ‰) (see colored bars in Fig. 4) (Chang et al., 2016a; Kawashima and Kurahashi, 2011; Chang and Ma, 2016). Thus, the measurement of δ¹⁵N–NH₃ can be used to distinguish between specific sources and to quantify their contribution to the measured total NH₃ pool. (Note that the isotopic signature determined by Kawashima and Kurahashi (2011) is for particulate ${\mathrm{NH}}_{\mathrm{4}}^{+}$, not NH₃.) As a first step, we examine the spatiotemporal characteristics of the measured δ¹⁵N–NH₃ relative to the N isotopic source signatures to infer seasonal changes in NH₃ sources.

The lowest δ¹⁵N–NH₃ values were observed at the rural sites (S3–S5) during the dry season ($-{\mathrm{32.72}}_{-\mathrm{46.28}}^{-\mathrm{20.38}}\pm \mathrm{6.46}$ ‰, n=21). These δ¹⁵N values are much lower than the δ¹⁵N of BB-related NH₃ and indicate the pervasive influence of agricultural NH₃ emissions in rural environments, rather than BB. During the rainy season, a drastic increase in δ¹⁵N–NH₃ ($-{\mathrm{23.97}}_{-\mathrm{37.99}}^{-\mathrm{12.35}}\pm \mathrm{7.50}$ ‰, n=22) at the rural sites was observed. Again, if BB was the dominating modulator of NH₃ levels, an increased contribution from BB-derived NH₃ during the dry versus the wet season in rural areas should have resulted in higher, not lower, δ¹⁵N–NH₃ values. The increased δ¹⁵N–NH₃ during the non-BB (i.e., rainy) period can probably be explained by the fact that agricultural NH₃ emissions with low δ¹⁵N–NH₃ can be dramatically lowered by continuous and heavy rainfall (Zheng et al., 2018; Chang et al., 2019a) so that at low levels, local sources can become more important (e.g., residential kitchens, nearby burning of biofuels for cooking).

As for the urban sites, the mean δ¹⁵N–NH₃ values at S6–S9 were $-{\mathrm{23.95}}_{-\mathrm{34.00}}^{-\mathrm{16.35}}\pm \mathrm{4.61}$ ‰, $-{\mathrm{25.53}}_{-\mathrm{35.01}}^{-\mathrm{14.10}}\pm \mathrm{6.11}$ ‰, $-{\mathrm{24.47}}_{-\mathrm{33.08}}^{-\mathrm{16.86}}\pm \mathrm{4.20}$ ‰and $-{\mathrm{25.32}}_{-\mathrm{39.44}}^{-\mathrm{17.13}}\pm \mathrm{7.75}$ ‰, respectively. The overall average δ¹⁵N–NH₃ value at the four urban sites ($-{\mathrm{24.82}}_{-\mathrm{39.44}}^{-\mathrm{14.10}}\pm \mathrm{5.74}$ ‰, n=68) was significantly (p<0.01; one-way analysis of variance and paired-sample t test; similarly hereinafter) higher than that at the rural ($-{\mathrm{28.24}}_{-\mathrm{46.28}}^{-\mathrm{12.35}}\pm \mathrm{8.22}$ ‰, n=43) and suburban ($-{\mathrm{29.94}}_{-\mathrm{45.62}}^{-\mathrm{18.78}}\pm \mathrm{7.35}$ ‰, n=34) sites, respectively, indicating a greater contribution of NH₃ emissions from pyrogenic (e.g., on-road traffic) sources. The average value of urban δ¹⁵N–NH₃ during the dry season ($-{\mathrm{24.21}}_{-\mathrm{34.47}}^{-\mathrm{14.10}}\pm \mathrm{4.82}$ ‰, n=28) was very similar to the average value observed during the rainy season ($-{\mathrm{25.25}}_{-\mathrm{39.44}}^{-\mathrm{15.31}}\pm \mathrm{6.33}$ ‰, n=40), after the pronounced decrease in NH₃ concentrations due to wet removal. This rather minor difference can hardly be ascribed to the influence of BB emissions, given the large seasonal fluctuation in wildfire intensity mentioned above. Based on the absolute δ¹⁵N–NH₃ values in the urban settings, and their rather invariant temporal trends, we argue that vehicle/transport is a more important and apparently steady source of pyrogenic NH₃ in the studied urban areas.

The two suburban sites (S1 and S2) are located geographically within the transition zone between the urban and rural environments, and this transitional character seems also indicated by their intermediate δ¹⁵N–NH₃ values ($-{\mathrm{29.94}}_{-\mathrm{45.62}}^{-\mathrm{18.78}}\pm \mathrm{7.35}$ ‰, n=34). However, interestingly, in comparison to the urban and rural sites, the overall δ¹⁵N–NH₃ value for the two suburban sites was significantly (p<0.01) higher during the BB season ($-{\mathrm{24.84}}_{-\mathrm{28.56}}^{-\mathrm{21.49}}\pm \mathrm{2.29}$ ‰, n=14) than that during non-BB season ($-{\mathrm{33.52}}_{-\mathrm{45.62}}^{-\mathrm{18.78}}\pm \mathrm{7.59}$ ‰, n=20) (Fig. 5). In fact, among the three different land use regimes, the average δ¹⁵N–NH₃ was highest for the suburban sites during the BB season, and it was also closer to the NH₃ isotopic signatures of pyrogenic sources during this time, raising questions regarding the importance of the contribution of BB versus road traffic in the suburban areas during the BB season.

3.4 Isotope-based quantification of BB contribution to ambient NH₃

There are several challenges that need to be overcome when trying to more accurately quantify the contribution of BB emissions to ambient NH₃ based on N isotope data. Firstly, given the use of only one isotope parameter (δ¹⁵N; in contrast to NO_x where the δ¹⁸O can also be analyzed), more than three potential NH₃ sources (e.g., urban and suburban sites) will introduce large uncertainties in isotopic end-member mixing models in terms of quantifying their relative contributions to the ambient NH₃ (Chang et al., 2015). Secondly, atmospheric wet scavenging could further compromise or alter the primary NH₃ N isotopic signatures (Elliott et al., 2019; Zheng et al., 2018; Chang et al., 2019a). For these reasons, we focus here on the samples collected at the three rural sites during the dry BB season (lasting 7 weeks) to isotopically examine the contribution of BB emissions to ambient NH₃. We separated these samples into seven groups based on the week of their sampling, and we integrated the measured δ¹⁵N–NH₃ values as well as the N isotopic signatures of potential NH₃ sources (i.e., biomass burning, livestock breeding, fertilizer application) into the Bayesian isotopic mixing model (see Sect. S1 for details). The results of NH₃ source apportionment are reported in Fig. 6. With a certain degree of variability, the contribution of BB to the ambient NH₃ in the rural areas during the 7 weeks of sampling in the dry season was only 21.0 % (±4.7 %). Hence, NH₃ emission from BB is significantly less important than from livestock breeding (37.1±7.1 %) and fertilizer application (41.8±5.9 %). This comes as a surprise, given the fact that the study area belongs to one of the most important BB regions in SE Asia, or even in the world, and the samples used for isotopic source apportionment were collected during the season of intensive BB.

Figure 6 Source apportionment results of ambient NH₃ in rural areas during the dry season based on Bayesian isotopic mixing modeling and the isotopic source signatures. The error bars indicate 2 standard deviations.

Figure 7 (a) Scatterplots of the aerosol K⁺ concentrations versus total NH₃ concentrations, as well as the NH₃ concentrations from BB emissions, at the rural sites during the dry season. (b) Bayesian isotope modeling-based source apportionment results of ambient NH₃ at the rural sites during the dry season, as function of the assumed N isotopic signatures of BB-emitted NH₃.

During the dry season, we also analyzed particulate potassium (K⁺), a chemical tracer of biomass combustion, at two rural sites (S4, S5; in 39 daily fine-particle (PM_2.5) samples). The particulate K⁺ data offer a valuable opportunity to validate our isotope-based source apportionment results. Again, we divided the dry-season data set into seven groups based on the week of NH₃ passive sampling. The correlation between the particulate K⁺ concentration and the total NH₃ concentration at the rural sites was rather poor (r²=0.43; blue symbols in Fig. 7a). Such a weak correlation supports our conclusion regarding the isotope-based source apportionment results (see above), providing additional independent evidence that BB can hardly be the dominant source of NH₃during the sampling period at the studied rural areas. In contrast, the correlation between the particulate K⁺ concentration and the estimated BB-derived NH₃ concentration (instead of total NH₃) is much better (r²=0.76; Fig. 7a) and, thus, further validates our modeling approach. While the independent particulate K⁺ data further increase our confidence in the N-isotope-based assessment, some uncertainty still remains with respect to the robustness of the end-member source δ¹⁵N values, potential source-altering effects, and in turn our estimates of the BB-associated NH₃ contribution. In other words, the latter is probably sensitive to the considered range in the δ¹⁵N of potential NH₃ emission sources, and this range may be quite large or uncertain for at least some of the sources. The δ¹⁵N–NH₃ from BB, in particular, is only poorly constrained, with hardly any reports from the literature (e.g., Kawashima and Kurahashi, 2011). In recent chamber experiments, we found that the δ¹⁵N–NH₃ produced by the combustion of a variety of biomass types (subtropical trees and agricultural residues) ranged between −11.8 ‰ and −4.6 ‰ (Chang and Ma, 2016, Chang et al., 2019a), which is distinctly lower than the N isotopic signature of BB-emitted ${\mathrm{NH}}_{\mathrm{4}}^{+}$ (12 ‰) determined previously (Kawashima and Kurahashi, 2011) and adopted in this study. Assuming that the true N isotopic signatures of BB-emitted NH₃ in the study area falls somewhere within the range of −12 ‰ to 12 ‰ (based on our published data in Chang and Ma, 2016; Chang et al., 2019a) and the value reported in Kawashima and Kurahashi, 2011), we re-calculated the source apportionment estimates as function of the different δ¹⁵N values for BB-emitted NH₃ (Fig. 7b). The estimates are not sensitive to the choice of the N isotopic composition of the BB-associated NH₃ source. Specifically, independent of the chosen δ¹⁵N–NH₃ value, BB is always the least important of the three main NH₃ sources in rural areas, contributing no more than 29.6 %. This is because although the isotopic signatures of BB-emitted NH₃ have a wide range of δ¹⁵N values, their δ¹⁵N–NH₃ values are still significantly (p<0.01) higher (i.e., without overlap as shown in Fig. 4) than the measured δ¹⁵N values of ambient NH₃ at the rural sites.

As illustrated by the pie chart in Fig. 6, the average contribution of BB to ambient NH₃ at the rural sites during the season of intensive fire events is 2.4 µg m⁻³; this value can be regarded as the maximum possible concentration of BB-emitted NH₃ for the urban and rural sites, which are much further away from the fire areas. Based on the total NH₃ concentrations measured at the other sites, we calculate that the contribution of BB to the ambient NH₃ in the urban and suburban areas is of the order of 9.6 % (ranging from 5.2 % to 14.8 %) and 12.3 % (ranging from 6.1 % to 19.9 %), respectively.

4 Conclusion

In this study, we integrated satellite constraints on atmospheric NH₃ levels and fire intensity, discrete NH₃ concentration measurement, and N isotopic analysis of NH₃ in order to assess the regional-scale contribution of BB to ambient NH₃ in the heartland of Southeast Asia. The combined approach provides a cross-validation framework for source apportioning of NH₃ in the lower atmosphere and will thus help to ameliorate predictions of BB emissions beyond the tropics, particularly in areas of high vegetation fire risk. Our results suggest that during the dry wildfire season, BB emissions represent a ubiquitous but comparatively small NH₃ source, which accounts for 9.6 %, 12.3 % and 21.0 % of ambient NH₃ in urban, suburban and rural environments, respectively. While we do not claim that our results necessarily apply at the global scale, and we do not question that BB is one of the most important global NH₃ sources, we find that at least in the heartland of SE Asia, BB related NH₃ emissions to the atmosphere are rather moderate and vary significantly in time and space. Both satellite observations and field and ground-based measurements capture these variations. Our findings underscore that BB-induced NH₃ emissions in tropical monsoon environments can be much lower than previously anticipated. Existing atmospheric transport models may overestimate current, and likely future, NH₃ emissions under changing climate conditions. While the full implications of our results remain to be explored, they promise to provide important guidance for revising NH₃ emissions from BB in atmospheric transport models to assess their impacts on air quality, human health and climate change.