Organic pollutants from tropical peatland fires: regional influences and its impact on lower stratospheric ozone

The particularly strong dry season in Indonesia in 2015, caused by an exceptional strong El Niño, led to severe peatland fires resulting in high volatile organic compound (VOC) biomass burning emissions. At the same time, the developing Asian monsoon anticyclone (ASMA) and the general upward transport in the intertropical convergence zone (ITCZ) efficiently transported the resulting primary and secondary pollutants to the upper troposphere/lower stratosphere (UTLS). In this study, we assess the importance of these VOC emissions for the composition of the lower troposphere and the UTLS, and 5 we investigate the effect of in-cloud oxygenated VOC (OVOC) oxidation during such a strong pollution event. This is achieved by performing multiple chemistry simulations using the global atmospheric model ECHAM/MESSy (EMAC). By comparing modelled columns of the biomass burning marker hydrogen cyanide (HCN) to spaceborne measurements from the Infrared Atmospheric Sounding Interferometer (IASI), we find that EMAC properly captures the exceptional strength of the Indonesian fires. 10 In the lower troposphere, the increase in VOC levels is higher in Indonesia compared to other biomass burning regions. This has a direct impact on the oxidation capacity, resulting in the largest regional reduction in hydroxyl radicals (OH) and nitrogen oxides (NOx). Even though an increase in ozone (O3) is predicted close to the peatland fires, particular high concentrations of phenols lead to an O3 depletion in eastern Indonesia. By employing the detailed in-cloud OVOC oxidation scheme Jülich Aqueous-phase Mechanism of Organic Chemistry (JAMOC), we find that the predicted changes are dampened and that by 15 ignoring these processes, global models tend to overestimate the impact of such extreme pollution events. In the ASMA and the ITCZ, the upward transport leads to elevated VOC concentrations in the UTLS region, which results in a depletion of lower stratospheric O3. We find that this is caused by a high destruction of O3 by phenoxy radicals and by the increased formation of NOx reservoir species, which dampen the chemical production of O3. The Indonesian peatland fires regularly occur during El Niño years and contribute to the depletion of O3. In the time period from 2001 to 2016, we find that 20 the lower stratospheric O3 is reduced by about 0.38 DU and contributes to about 25 % to the lower stratospheric O3 reduction observed by remote sensing. By not considering these processes, global models might not be able to reproduce this variability in lower stratospheric O3. 1 https://doi.org/10.5194/acp-2020-1130 Preprint. Discussion started: 17 November 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Particularly strong Indonesian wildfires during the El Niño in 2015 led to severe air pollution and reduced visibility (Kim 25 et al., 2015;Lee et al., 2017) resulting in increased morbidity and mortality (Marlier et al., 2013;Reddington et al., 2014;Crippa et al., 2016) in South-East Asia (SEA). In general, El Niño is a large-scale climate anomaly, which is characterised by significantly warmer eastern equatorial Pacific Ocean sea surface temperatures (Trenberth, 1997), resulting in a dry season in SEA (Weng et al., 2007). The very strong El Niño phase in 2015-2016, which is the third strongest on record (after 1997(after -1998(after and 1982(after -1983(after , NOAA, 2020, led to a particularly strong dry season in Indonesia (Jiménez-Muñoz et al., 2016). In the 30 past, much of the originally forested and moist peatland in Kalimantan and Sumatra has been drained and cleared during agricultural land management. In order to clear these forests, landscape fires are commonly used. Even small local fires in these regions during non El Niño years may induce particular strong biomass burning emissions. Gaveau et al. (2014) estimate that a local one-week Indonesian biomass burning event in 2013 contributed to about 5-10 % of Indonesian's total greenhouse gas emissions in that year. The additional drying during El Niño years favours fires that burn deep down into the peat and (RO 2 ) react with HO 2 , NO x , and NO 3 , and undergo self-and cross-reactions (Sander et al., 2019). Isocyanic acid (HNCO) is a chemical constituent that is heavily emitted by biomass burning and potentially harmful to humans Roberts et al., 2011;Leslie et al., 2019). In order to properly represent this toxic constituent within EMAC, MOM has been extended to represent the atmospheric chemistry of HNCO. For this, the mechanism proposed by Rosanka et al. (2020d) is implemented into MOM. Their mechanism includes formamide as an additional chemical source of HNCO and chemical mechanisms for 130 nitromethane, methylamine, dimethylamine, and trimethylamine.
The atmospheric aqueous-phase chemistry is modelled using the SCAVenging submodel (SCAV, Tost et al., 2006). It simulates the removal of trace gases and aerosol particles by clouds and precipitation. SCAV calculates the transfer of species into and out of rain and cloud droplets using the Henry's law equilibrium, acid dissociation equilibria, oxidation-reduction reactions, heterogeneous reactions on droplet surfaces, and aqueous-phase photolysis reactions (Tost et al., 2006). As mentioned 135 earlier and as demonstrated by Rosanka et al. (2020b), in-cloud OVOC oxidation significantly influences the atmospheric composition. However, the ordinary differential equations (ODE) systems resulting from the combination of gas-phase and in-cloud aqueous-phase suffer from (1) a higher stiffness due to fast acid-base equilibria and phase-transfer reactions, and (2) load imbalance on High-Performance Computing (HPC) systems due to the sparsity of clouds. This leads to a significant increase in computational costs when using larger chemical mechanisms like the Jülich Aqueous-phase Mechanism of Organic Chemistry 140 (JAMOC), i.e. larger ODE systems (Rosanka et al., 2020c). Using JAMOC in each simulation performed in this study is thus not feasible. As a trade-off, JAMOC is used in a simulation subset in order to address and estimate its implications on the other simulations. Thus, two different aqueous-phase mechanisms are used within this study: (1) the standard aqueous-phase mechanism of EMAC (in the following called ScSta), which includes a detailed oxidation scheme and represents more than 150 reactions (Jöckel et al., 2016), and (2) JAMOC (Rosanka et al., 2020c), which includes a complex in-cloud OVOC oxidation 145 scheme. In JAMOC, the phase transfer of species containing up to ten carbon atoms and the oxidation of species containing up to four carbon atoms are represented. Similar to MOM, both aqueous-phase mechanisms are modified to include the changes proposed by Rosanka et al. (2020d) to properly represent HNCO.

Biogenic and Biomass Burning VOC Emissions
In the atmosphere, biogenic and biomass burning emissions are the dominant sources of VOCs. The largest biogenic emis-150 sions take place in the equatorial region (e.g. Amazon Basin, Central Africa) with additional emissions in the Northern (NH) and Southern Hemisphere (SH) extratropics. The MESSy submodel Model of Emissions of Gases and Aerosols from Nature (MEGAN, Guenther et al., 2006) is used to calculate biogenic VOC emissions. The global emissions of isoprene, the most abundant biogenic VOC, are scaled to 595 Tg a −1 , the best estimate of Sindelarova et al. (2014). based on biomass burning emission factors and dry matter combustion rates. For the latter, data from the GFAS are used that are based on satellite observations of the fire radiative power obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite instruments (Kaiser et al., 2012). In BIOBURN, the emission strength depends on the dominant fire type in the respective area. From the GFAS dataset used in EMAC, in 2015, the dominant fire type over Indonesia is tropical forest fire. However, as discussed earlier, peatland fires contribute substantially to the Indonesian fires. The GFAS dataset 160 of EMAC is changed such that the dominant fire type over Indonesia is a combination of peat and tropical forest fires with equal contributions (following van der Werf et al., 2017). In general, biomass burning emission factors for VOCs are based on Akagi et al. (2011). Biomass burning emissions for HNCO, formamide, nitromethane, methylamine, dimethylamine, and trimethylamine are implemented following Rosanka et al. (2020d) using emission factors from Koss et al. (2018) for HNCO and formamide.

Observational comparison
The evaluation of model simulation results against global observational datasets of VOC abundance can be performed for only a few species, mainly because of the limited availability in spaceborne measurements of such compounds. Among them, several VOCs are retrieved globally from the observations made by the nadir-viewing hyperspectral Infrared Atmospheric Sounding Interferometer (IASI, Clerbaux et al., 2009). Embarked on the Metop platforms on sun-synchronous polar orbits, IASI crosses 170 the equator at 9:30 and 21:30 local solar time and achieves a global coverage twice daily with a fairly dense spatial sampling.
Although significant enhancements of carbon monoxide (CO) and ammonia (NH 3 ) have already been captured by the IASI measurements in the 2015 Indonesian fires (Whitburn et al., 2016b;Nechita-Banda et al., 2018), here we make use of the HCN abundance retrieved from the IASI/Metop-A and -B observations to assess the ability of EMAC to represent such an important biomass burning event. This choice is explained in Sect. 3. In addition, IASI methanol (CH 3 OH) data are used to assess the 175 impact of in-cloud OVOC oxidation in the model simulations (Sect. 5).
The retrieval method used to obtain the HCN measurements from the IASI observations follows closely the version 3 of the Artificial Neural Network for IASI (ANNI), which already allowed the retrieval of a suite of VOCs, including CH 3 OH (Franco et al., 2018). ANNI is a general retrieval framework that consists in quantifying, for each IASI observation, the spectral signature of the target gas with a sensitive hyperspectral metric, and in converting this metric into gas total column 180 via an artificial feedforward neural network (NN). Details on the ANNI retrieval approach, the HCN retrieval specificities, and the HCN product itself are provided in Appendix A. We refer to Franco et al. (2018) for a description of the IASI methanol retrievals. The satellite datasets exploited in this study consist of daily global distributions of HCN and CH 3 OH total columns derived from the daytime observations (approximately 9:30 a.m., local time) of the IASI/Metop-A and -B overpasses. These offer a better measurement sensitivity than the evening overpasses (Franco et al., 2018). Scenes affected by clouds or poor 185 retrieval performance are removed from the final dataset by specific filters. Examples of daily regional distributions of HCN columns in the 2015 Indonesian fires as well as the seasonal global distributions of HCN as retrieved from IASI are presented in Appendix A. Those highlight the ability of IASI to capture the enhancements of HCN during biomass burning events as well as its downwind transport over long distances. 190 Within this study, seven simulations are performed, which can be summarised in three simulation sets. Each simulation differs either in the biomass burning emissions, the aqueous-phase mechanism used, or the modelled time period. T106L90MA and using JAMOC is computationally not feasible. Therefore, the EMAC's standard aqueous-phase mechanism is used and the resolution is reduced to T42L90MA.

The representation of biomass burning events in EMAC
HCN mainly originates from combustion processes and is therefore largely emitted by biomass burning (Shim et al., 2007).
Other emission sources including industrial activities, automobile exhaust, and domestic biofuel are assumed to be very weak 210 (Lobert et al., 1990;Li et al., 2009). Reactions involving acetonitrile (CH 3 CN) are the only gas-phase source of HCN, but those are estimated to be a minor contribution to the atmospheric HCN burden (Li et al., 2009). The slow oxidation of HCN by OH and O( 1 D) is considered to be the most important atmospheric gas-phase sink, leading to long chemical lifetimes (Cicerone and Zellner, 1983). However, due to a strong ocean uptake, the atmospheric lifetime is reduced to a few months (Li et al., 2000(Li et al., , 2009. The almost exclusive biomass burning source, combined with a long atmospheric residence time that 215 allows for long-range transport, makes HCN a widely used primary tracer of biomass burning emissions and fire plumes (Li et al., 2009). Other typical fire tracers such as CO and NH 3 either have several other sources or are too short-lived to track fire plumes over long distance. Therefore, HCN satellite data from IASI are used here to evaluate the performance of EMAC in representing the 2015 Indonesian peatland fires. Figure 4 shows the comparison of modelled HCN total columns to IASI satellite retrievals for the three months with strong peatland emissions in Indonesia. At the beginning of the Indonesian fires, the emitted HCN is transported westwards leading to high HCN column values over the Indian Ocean. While the fires are ongoing throughout October, the strong westward transport of HCN results in the complete covering of the Indian ocean. Some HCN is also transported eastwards over Australia and the Pacific Ocean. In November, the air masses from Indonesia mix with emissions from Africa and the eastward transported air masses reach South America. In general, EMAC strongly underestimates HCN when its biomass burning source 225 is not taken into account (simulation REF). Once the HCN biomass burning emissions are taken into account, the overall underprediction in EMAC is mostly resolved. However in the FIR simulation, HCN is partially overpredicted in SEA during the main fire period (September and October). Additionally, HCN columns are slightly overestimated in CSA (September and October) and CAF (September). Interestingly, the relative model bias (not shown) is similar in all three regions. Due to the particular strength of the Indonesian fires, the absolute bias is more pronounced in SEA. EMAC's representation of HCN is 230 associated with some uncertainties. The occurrence of biomass burning events is very low in the Amazon Basin, suggesting that the high HCN column values are not caused by fires or transport from CSA. Especially in NH autumn (SON), the Amazon basin is known to be highly influenced by biogenic emissions. Shim et al. (2007) already suggested that biogenic HCN emissions may contribute to atmospheric concentrations by up to 18 %. In the submodule MEGAN, biogenic HCN emissions are taken into account and contribute about 15 % to the total HCN emissions, suggesting that EMAC's overprediction is not 235 caused by a misrepresentation of other sources. Additionally, it is expected that the atmospheric lifetime of HCN is reasonably well represented, since globally HCN columns are well reproduced. Moreover, the ocean uptake accounts for 1.2 Tg(N) a −1 , which is well in the range of 1.1 to 2.6 Tg(N) a −1 proposed by Li et al. (2000) and very close to the Singh et al. (2003) estimate of 1.0 Tg(N) a −1 . The representation of biomass burning within EMAC depends on satellite observations (Sect. 2.1.2), which retrieve the fire radiative power and are thus sensitive to clouds. This introduces some uncertainties in regions that are 240 characterised by the frequent occurrence of clouds, like equatorial Asia. Focusing on Indonesia, Liu et al. (2020) compared five different global fire inventories and found that GFAS, the inventory used in this study, represents the strength of these fires best. Still, GFAS even tends to slightly underestimate the strength, when compared to regional observations in Singapore, Malaysia, and Indonesia. This suggests that the magnitude of the Indonesian fires is well represented in EMAC. In this study, we use the emission factors optimised for atmospheric models by Akagi et al. (2011), which suggest 5.0 g kg −1 for HCN 245 from peatland fires. However, from the literature a high uncertainty in the emission factors for HCN are reported. From recent field measurements in Indonesia and Malaysia, Stockwell et al. (2016) and Smith et al. (2018) report values ranging from 0.34 g kg −1 to 8.21 g kg −1 , whereas lab measurements for Indonesian peatland by Stockwell et al. (2015) suggest values between 3.30 g kg −1 and 3.83 g kg −1 . Overall, this results in a mean emission factor of 4.40 g kg −1 across all studies (Andreae, 2019), suggesting that some of EMAC's overestimation is caused by a slightly too high HCN emission factor. Due to its long 250 lifetime, HCN is transported over long distances. West of Indonesia, EMAC also predicts higher HCN columns than observed by IASI, suggesting that some of the overprediction is caused by the deviation of horizontal transport (further discussed in Sect. 7). Figure 5 gives the frequency of the global HCN EMAC total column bias in relation to the IASI retrievals during the Indonesian peatland fires, once including biomass burning emissions in the simulations and once not. This comparison confirms that HCN is strongly underestimated when its main source is not represented in EMAC. With HCN from biomass burning, the mean column bias reduces from −5.32 ×10 −15 molecules cm −2 to −1.06 ×10 −15 molecules cm −2 and its variance reduces from 1.75 ×10 −31 molecules 2 cm −4 to 2.57 ×10 −30 molecules 2 cm −4 , significantly improving the representation of HCN in EMAC. From this analysis we conclude that even though EMAC does not reproduce HCN columns perfectly, the Indonesian fires are reasonably well represented, especially when considering the exceptional strength of the 2015 Indonesian fires (for further discussion see Appendix A and Fig. A3). This also holds true considering all global biomass burning emission events.  Table 3 provides an overview on the global and regional changes (between simulation REF and FIR) in the tropospheric burden of each species discussed in the following subsections. The regional changes reported in 265 Table 3 are calculated for the respective main biomass burning season defined in Table 1.

Impact on VOCs
Many VOCs are characterised by short lifetimes resulting in highly location-dependent changes within the troposphere. Direct emissions are the only source of atmospheric hydrocabons. Globally, biomass burning emissions of VOCs significantly increase the atmospheric concentration of many hydrocarbons, with acetylene (C 2 H 2 ) and ethane (C 2 H 6 ) being the two hydrocarbons 270 impacted the most. Their tropospheric burden increases by 20.5 % and 32.6 %, respectively, with the highest regional change is in SEA.
The two most abundant aromatics, benzene (C 6 H 6 ) and toluene (C 7 H 8 ), are strongly emitted by biomass burning events. In the FIR simulation, the tropospheric burden of C 6 H 6 increases by 27.3 %. Toluene has a slightly lower increase of only 15.3 %.
The tropospheric concentration of oxygenated aromatics also increases strongly. The most dominant change is predicted for 275 phenol (C 6 H 5 OH), whose tropospheric burden is more than doubled and increases to 2.3 Gg. Even though phenol is directly emitted by biomass burning, the overall high aromatic emissions lead to an enhanced chemical production of phenol from benzene oxidation. The highest absolute change is observed in SEA. However, due to low aromatic background concentrations, the relative increase is higher in ALA, CSA, and NAU. Methanol (CH 3 OH) is directly emitted by biomass burning emissions but also chemically produced from VOC oxidation.

280
Globally, methanol increases by 7.9 % when biomass burning VOC emissions are taken into account. The high VOC emissions during the Indonesian peatland fires result in the highest changes in SEA. The two α-dicarbonyls glyoxal (OCHCHO) and methyl glyoxal (CH 3 C(O)CHO) are primarily produced from VOC oxidation. The global burden of glyoxal and methyl glyoxal increases by 9.3 % and 1.3 %, respectively. Again, the highest absolute changes are predicted in SEA. However, the highest relative change occurs in ALA due to generally low background VOC concentrations.
In the atmosphere, organic acids are mainly produced from the photo-oxidation of biogenic and anthropogenic VOCs but may also be emitted from biomass burning. Formic acid (HCOOH) is slightly impacted by biomass burning VOC emissions and globally increases by 4.9 % with the highest changes in SEA and Africa (CAF and SAF). The acid impacted the most by biomass burning is acetic acid (CH 3 CO 2 H), which globally gains 23.3 % with the highest changes in SEA, CAF, CSA, and SAF. Interestingly, the high increase predicted in CSA only leads to a low relative rise. This is due to generally high background 290 concentrations in this region from high biogenic VOC emissions.

Impact on hydroxyl radicals (OH)
In general, organic molecules react with OH by either H-abstraction or addition to double bonds, making OH the most important daytime VOC oxidant. Figure 6a gives the mean tropospheric surface OH concentration in 2015 and Fig. 6b presents the changes due to biomass burning VOC emissions. OH concentrations are significantly reduced in most regions with frequent 295 biomass burning events. This reduction is caused by the direct reaction of OH with VOCs, and the enhanced formation of CO from VOC degradation. The reduction in OH is not uniformly distributed and depends on the local chemical regime. In Indonesia, the high VOC emissions lead to the highest absolute and relative OH reduction. The enhanced oxidation of VOCs by OH leads to an overall increase in HO 2 . In ALA and NAS, the most northern areas of interest, the absolute change in OH is low. Within the biomass burning plume, the enhanced HO 2 concentrations react with NO producing OH and compensating 300 the OH reduction by VOC degradation, resulting in a regional surface OH increase. Still, outside the biomass burning plume, an overall decrease in OH is predicted in ALA and NAS. Here, VOCs from biomass burning become the highest OH sink resulting in strong relative changes in OH reactivity. In general, OH reactivity is the highest in the Amazon Basin (100 s −1 ) and the lowest in Antarctica (0.5 s −1 ). The additional VOC emissions in Indonesia result in a significant increase of about 50 % in the OH reactivity, which is similar to the increases predicted in ALA and NAS. respectively. The additional VOC emissions significantly reduce the regional concentrations in tropospheric NO x . In SEA, the absolute changes are large but small in relative (about 8 %), whereas the highest absolute and relative NO x changes are predicted in ALA. These reductions are caused by enhanced reactions of RO 2 with NO x resulting in an increased formation 310 of NO x reservoir species (i.e. alkyl and acyl peroxy nitrates) and nitrogen-containing aromatics (e.g. nitrophenols).
NO 3 is the most important nighttime oxidant, which is globally increased by about 5 %. On the one hand, the formation of NO 3 is enhanced from aromatic RO 2 reacting with NO 2 , but on the other hand the loss of NO 3 by reactions with RO 2 and aldehydes is increased. In the two northern regions (ALA and NAS), the elevated O 3 and regionally increased NO 2 concentrations induce an enhanced formation from inorganic reactions, resulting in an additional rise of NO 3 . The absolute 315 increase in NO 3 is high in SEA, especially in Indonesia. Here, the particularly large increase in phenols results in enhanced concentrations of phenyl peroxy radicals (C 6 H 5 O 2 ), which form NO 3 when reacting with NO 2 following Jagiella and Zabel

Impact on ozone (O 3 )
The perturbed NO x -HO x relation consequently leads to changes in tropospheric O 3 . Figure 8a shows the mean tropospheric

Pollution and toxic conditions
The direct emission and degradation of primarily emitted VOCs lead to the formation of toxic compounds that are of special 345 interest in highly populated areas. One prominent example is nitrophenols, which are known to have a high phytotoxic activity that is enhanced by a high photochemical stability (Grosjean, 1991). Rippen et al. (1987) and Natangelo et al. (1999) suggested that nitrophenols could have contributed to the forest decline in Northern and Central Europe in the 80's but also in other parts of the world. In the atmosphere, nitrophenols are mainly formed from the oxidation of the aromatic compounds benzene, toluene, phenols, and cresols (Nojima et al., 1975;Atkinson et al., 1980;Grosjean, 1984), of which the first three are emitted nitrophenols, which have been measured in rain droplets (Leuenberger et al., 1985;Schummer et al., 2009). JAMOC represents the phase transfer of some nitrophenols. However, their reduction due to this additional sink is calculated to be below 1 %. This insignificant reduction results from the missing OH sink of these nitrophenols (Hems and Abbatt, 2018) in JAMOC. Therefore, the predicted nitrophenol concentrations are expected to be slightly overestimated. Still, the overall increase of nitrophenols in biomass burning areas is a potential danger for plants in these regions where plants are already under stressed conditions due 360 to the biomass burning itself. At the same time, nitrophenols are known to absorb solar radiation (Hems and Abbatt, 2018) and therefore enhance hazy conditions in those areas (Lee et al., 2017), contributing to increased morbidity and mortality (Crippa et al., 2016).
Isocyanic acid (HNCO) is also known to be a toxic constituent of biomass burning emissions. It is linked to protein carbamylation, which causes adverse health effects such as rheumatoid arthritis, cardiovascular diseases, and cataracts (Wang 365 et al., 2007;Roberts et al., 2011;Leslie et al., 2019). It is expected that the protein carbamylation potentially starts if humans are exposed to ambient concentrations above 1 ppb (Roberts et al., 2011). Rosanka et al. (2020d) already reported that HNCO concentrations are high in regions characterised by strong biomass burning events. Globally, similar high concentrations are predicted in this study. However, we predict higher concentrations in Indonesia than Rosanka et al. (2020d), who reported that ambient HNCO conditions of 1 ppb are exceeded for less than 30 days in Indonesia in 2011. The year 2011 is known to have 370 low biomass burning emissions in this region (van der Werf et al., 2017). Figure 10 shows the number of days, in which this threshold is exceeded during the 2015 Indonesian peatland fires. Here, 1 ppb of HNCO is regularly exceeded and some regions are affected during the complete fire period. This causes potentially severe health effects for the population of Indonesia.

The influence of in-cloud OVOC oxidation during the Indonesian peatland fires
The influence of in-cloud OVOC oxidation is addressed by applying JAMOC during the Indonesian fire period (simulations induced by JAMOC, the changes from the Indonesian fires due to the in-cloud OVOC oxidation are calculated following: (1) Figure 11 shows the changes in the zonal mean concentration over Indonesia and the Indian Ocean of all OVOCs explicitly reacting in JAMOC for the simulations without JAMOC (Fig. 11a) and the predicted changes due to JAMOC ( Fig. 11b; 380 calculated using Reaction 1), focusing on the Indonesian fire period (SON). Due to the high solubility of many OVOCs and their in-cloud oxidation, their concentration is strongly reduced at altitudes that are characterised by frequent cloud events. The additional in-cloud sink for methanol, glyoxal, and methyl glyoxal leads to a lower increase in their burden ranging between 23 and 32 %. Figure 12 shows the Probability Density Function (PDF) for EMAC's methanol column bias when compared to IASI satellite retrievals (Franco et al., 2018)  FIR JAMOC ). A high fraction of SEA is covered by oceans. Millet et al. (2008) suggested that some regions of the Pacific and Indian Ocean are a net source of methanol. As discussed by Rosanka et al. (2020b), EMAC represents the ocean as a net methanol sink. Therefore, when comparing the predictions of methanol from EMAC to satellite observations, a certain underestimation is expected. Thus, simulation FIR JAMOC compares the best with IASI retrievals, since it has overall the lowest relative biases.

395
Changes in hydrocarbons are minimal due to their low solubility, whereas strong changes are predicted for the relative insoluble O 3 . Due to in-cloud OVOC oxidation, the initially predicted increase in O 3 in western Indonesia and over the Indian Ocean (Sect. 4.4) is dampened by more than 60 % once JAMOC is implemented. This reduced increase is caused by the increasing importance of clouds as O 3 sinks. This process is globally analysed by Rosanka et al. (2020b) and is based on the enhanced HO 2 formation in cloud droplets by OVOC oxidation. Within clouds, HO 2 is in acid equilibrium with the superoxide 400 anion (O − 2 ), which actively destroys O 3 . To conclude, in-cloud OVOC oxidation is important to properly represent the resulting impacts from strong pollution events especially during the monsoon season. Overall, the predicted impact on VOCs, radicals, and O 3 is dampened by the in-cloud oxidation and models neglecting this process probably tend to overestimate the impact of such an event. It is widely recognised that clouds may act as a source of secondary organic aerosols (SOA), which even enhances by in-cloud oxidation processes 405 (Blando and Turpin, 2000;Ervens et al., 2011;Ervens, 2015). Ervens et al. (2011) suggested that cloud processes might contribute in the same order to SOA formation as gas-phase processes. Within this study, SOA formation from cloud processes are not explicitly represented. However, it is expected that the enhanced VOC concentrations from biomass burning will lead to an increased SOA formation from aqueous-phase processes due to the enhanced formation of oligomers (e.g. from glyoxal and methyl glyoxal) within clouds.

410
Some of the biomass burning VOC emissions from SEA are quickly transported by the ASMA and the general tropical updraft into the UTLS (see Sect. 1 and Vogel et al., 2015), which significantly increases lower stratospheric VOC concentrations. The effect of this transport process can be observed in Fig. 11. For example, the highest increase in glyoxal, methanol, and phenol concentrations in the lower stratosphere (48.5, 33.1, and 149.4 %, respectively) is predicted in November. In the following 415 months, these VOCs actively react with O 3 and change the lower stratospheric radical chemistry. Overall, this results in a reduction in lower stratospheric O 3 peaking in April 2016, as shown in Fig. 13. Especially in the tropics, lower stratospheric O 3 diminishes by more than 12 ppb. We find that the substantial increase in phenols, caused by the high aromatic emissions from Indonesia, favours the formation of phenoxy radicals that contribute the most to this O 3 depletion via Reaction R2. Under high-NO x conditions, many VOCs form NO x reservoirs (i.e. alkyl and acyl peroxy nitrates), which flattens the peak of the 420 NO x burden in the lower stratosphere by 7.5 % and increases the HO 2 burden by 3.3 %.  In the second half of the 20 th -century, stratospheric O 3 declined mainly due to halogen-containing substances from anthropogenic activities (Molina and Rowland, 1974). After the Montreal Protocol has been implemented in 1989, a slowdown of the anthropogenic stratospheric depletion was observed (Strahan and Douglass, 2018). Even though O 3 recovers in the upper-and mid-stratosphere, a decline in lower stratospheric O 3 is observed by remote sensing measurements (Kyrölä et al., 2013;Nair 450 et al., 2015;Vigouroux et al., 2015). Recently, a lower stratospheric O 3 decline of about 1.5 DU has been reported between 2001 and 2016 by Ball et al. (2018, their Fig. 3). This reduction is mainly attributed to meteorological variability's in dynamical processes (Chipperfield et al., 2018;Ball et al., 2019). Between 2001 and 2016, we predict a lower stratospheric O 3 decrease of about 0.38 DU, when using the lower stratosphere definition of Ball et al. (2018). They define the lower stratosphere between 147-32 hPa (about 13-24 km) above 30°in latitudinal direction and between 100-32 hPa (about 17-24 km) below 30°lati-455 tude. Based on our results, we therefore appoint that biomass burning VOC emissions from SEA (in particular the Indonesian peatland fires) contribute to this observed decline by about 25 %, which is robust against the influence of in-cloud OVOC oxidation (about 20 %). To our knowledge, most global stratospheric models do not consider this kind of VOC emissions and their chemistry, and are thus not able to capture this variability. However, it is important to keep in mind that our simulations are to some degree idealised in order to be able to isolate the impact of these emissions. For example, meteorological variations 460 induced by changes in the chemical composition are neglected.
Another interesting aspect is the reduction in lower stratospheric NO x . In the UTLS, aviation is the only direct anthropogenic activity and contributes about 3-5 % to the total anthropogenic climate change (Lee et al., 2010). Here, aviation NO x emissions lead to a formation of O 3 and a depletion of methane (CH 4 ). Recently, Rosanka et al. (2020a) showed that the enhancement in O 3 is limited by the background concentrations of NO x and HO x . If enough HO x is available, a lower background NO x 465 concentration results in a higher O 3 gain. In general, low background HO x concentrations limit the O 3 gain in winter. In our study, we find that in the North Atlantic flight sector (between 400-100 hPa), the NO x burden is reduced due to SEA fires by about 6 % with regional changes of more than 20 % in 2015. At the same time, HO x increases regionally by 10 %.
Therefore, VOC emissions from frequently occurring Indonesian peatland fires potentially favour the formation of O 3 from aviation activities.

Model uncertainties
The most important aspects that influence our results are the representation of the transport processes, using different model resolutions, and the chemical kinetics. Each aspect is associated with some uncertainties of which all are shortly discussed in this section.
The magnitude of the depletion in lower stratospheric O 3 depends closely on the representation of the vertical transport that 475 conveys the emitted VOCs into the UTLS. In order to evaluate the vertical transport processes of global models, 222 Radon ( 222 Rn, radioactive decay half-lifetime of 3.8 days) is typically used (Mahowald et al., 1997;Zhang et al., 2008;Jöckel et al., 15 https://doi.org/10.5194/acp-2020-1130 November 2020 c Author(s) 2020. CC BY 4.0 License. 2010). Jöckel et al. (2010) and more recently Brinkop and Jöckel (2019) analysed the ability of EMAC to capture the 222 Rn surface concentrations and vertical profiles. Their findings indicate that the vertical transport is well represented in EMAC (using the T42L90MA resolution) and that they are comparable to the earlier analysis with ECHAM5 (the base model of 480 EMAC) by Zhang et al. (2008). Figure 4 shows that the horizontal transport is also an important aspect that influences the distribution of the emitted VOCs from Indonesian peatland fires. Evaluating the horizontal transport using observations (like 222 Rn) is currently not possible. Recently however, Orbe et al. (2018) compared transport time scales of various global models, including EMAC. They found that the horizontal transport from NH mid-latitudes to the tropics differs by 30 %. Based on this comparison, it can be assumed that the horizontal transport is reasonably well represented in EMAC.

485
In this study, we transfer our process understanding from the fine-(T106L90MA) to the coarse-resolution (T42L90MA) simulations. It is therefore important to understand how well transport processes agree between both resolutions. Currently, no direct analysis has been performed that focuses on the impact of different resolutions on transport processes in EMAC.
However, Aghedo et al. (2010) analysed the influence of different horizontal and vertical resolutions in ECHAM5. Since EMAC uses the same horizontal and vertical transport scheme as ECHAM5, we assume that their findings also apply to EMAC. They 490 find that the vertical transport mainly depends on the number of levels used. By increasing the number of layers from 19 to 31 levels, the mass transported into the stratosphere reduces globally by about 36 %, whereas increasing the resolution from T42 to T106 only decreases the vertically transported mass globally by about 10 %. Here, the influence is the lowest (about 7 %) at high latitudes and highest in the tropics (about 17 %). Aghedo et al. (2010) suggested that the higher impact in the tropics is probably related to tropical convection processes. Increasing the resolution changes the meridional transport in most regions 495 by less than 2 % and is thus negligible. For our purposes, differences in the inter-hemispheric transport are also negligible.
The mean transport time from the NH to SH decreases from 11.9 to 11.8 months and for the SH to NH transport from 11.4 to 11.5 months when increasing the horizontal resolution from T42 to T106. By using the same vertical resolution (90 levels), the highest uncertainty introduced by using different resolutions is eliminated. It is therefore expected that the important transport processes are comparable and properly represented in both resolutions. 500 We find that the reaction of phenoxy radicals with O 3 (Reaction R2) has a significant influence at the surface, in the troposphere, and the lower stratosphere. As discussed by Taraborrelli et al. (2020), the chemical kinetics used in MOM to represent this O 3 loss is associated with some uncertainties. Currently, only the measured reaction rate constant for C 6 H 5 O is available and this is used for all phenoxy radicals. Yet, no experimental evidence has been found for the formation of phenyl peroxy radical (C 6 H 5 O 2 ), which might influence the cycling nature of this O 3 loss by Reactions R1 and R2. However, this product 505 is still to be expected. Even with different products, a significant deletion of O 3 is anticipated by Reaction R2. At the same time, the reaction rate from Tao and Li (1999) is reported to be at the lower end, whereas a higher reaction rate would increase the depleted O 3 . Additionally, Taraborrelli et al. (2020) report that MOM neglects the non-HONO formation channel from nitrophenol photolysis, which does not destroy the aromatic ring and reforms phenoxy radicals (Cheng et al., 2009;Vereecken et al., 2016). It is therefore expected that, due to increasing nitrophenol concentrations in the lower troposphere (Sect. 4.5) as 510 well as in the UTLS, the importance of Reaction R2 as an O 3 sink is potentially underestimated.
In this study, the influence of VOC emissions from reoccurring Indonesian pealtand fires is analysed with the main focus on 2015, a particularly strong year. This is achieved by performing multiple global simulations using EMAC. By comparing EMAC's prediction of HCN columns to IASI satellite retrievals, we show that EMAC properly represents the emissions 515 from the Indonesian peatland fires and global biomass burning events. Our results indicate that VOC emissions from biomass burning are important to reproduce hydrocarbons and secondary OVOCs in the atmosphere. Compared to other biomass burning regions, a particularly strong increase is modelled in SEA region, due to the unique emission footprint from the Indonesian peatland fires. Regionally, significant changes in radical concentrations (HO x and NO x ) are predicted. In general, O 3 increases in the lower troposphere with the highest changes in the NH high latitudes due to strong fires in Boreal Asia. However, on a 520 global scale, tropospheric changes in O 3 are negligible. High aromatic emissions from peatland fires lead to a depletion of O 3 in eastern Indonesia. The enhanced formation of nitrophenols and strong HNCO emissions create toxic conditions in most parts of Indonesia, directly influencing its population. The overall impact in the lower troposphere is reduced when in-cloud OVOC oxidation is taken into account. Especially, the O 3 increase initially predicted is reduced due to its enhanced destruction within clouds. However, the increased formation of oligomers in cloud droplets potentially leads to enhanced SOA concentrations.

525
The ongoing ASMA and the general tropical upward transport during the Indonesian fires, lift the emitted VOCs and their oxidation products quickly to the UTLS. Here, the enhanced VOC concentrations contribute to the depletion of lower stratospheric O 3 . In particular, the destruction by phenoxy radicals plays a key role. The predicted O 3 depletion is in line with remote sensing measurements and supported by our results that these VOC emissions contribute by about 25 %. Although high VOC emissions from biomass burning events in ALA and NAS have a large regional impact, their impact on the UTLS is negligible 530 due to missing fast upward transport at higher latitudes.

Appendix A: HCN retrievals from IASI observations
The spaceborne data of HCN columns used in this study are obtained from the IASI radiance spectra by applying the version 3 of the Artificial Neural Network for IASI (ANNI) retrieval framework. Initially developed for the retrieval of NH 3 and dust from the IASI observations (Whitburn et al., 2016a;Clarisse et al., 2019), ANNI v3 incorporates updates and modifications to 535 allow the retrieval of a suite of VOCs. Until now, it has been used to retrieve methanol, formic acid, and PAN (Franco et al., 2018), then acetone (Franco et al., 2019) and acetic acid (Franco et al., 2020). Here, we perform the HCN retrieval by applying the full ANNI v3 procedure. As this approach has already been described in detail (see Franco et al., 2018, and references therein), we limit ourselves here to a summary of the main retrieval steps, and to the elements specific to the retrieval of HCN.
Examples of HCN columns from IASI single overpasses in the 2015 Indonesian fire plumes and averaged distributions are also 540 presented.
As mentioned in Sect. 2.1.3, the ANNI retrieval method proceeds in two major steps. First, in each individual IASI radiance spectrum, the target species is detected and the strength of its absorption is quantified by a metric called the Hyperspectral Range Index (HRI). Then, the HRI is converted into a gas total column by means of an artificial feedforward neural network (NN), which also provides an uncertainty on the retrieved column.

545
The HRI is a dimensionless metric of the magnitude of the spectral signature of a target species in a given IASI spectrum, relative to the spectral variability of a "background" atmosphere in absence of the target gas, i.e. a variability resulting from all other parameters that contribute to the spectral radiance, such as other atmospheric gases (see Walker et al., 2011). The HRI is calculated over the main spectral range, in which the target species absorbs. The HCN absorption band (ν 2 branch) included in the IASI spectrum is situated close to a strong Q branch of CO 2 near 720 cm −1 . Therefore, the whole 700-800 cm −1 550 spectral range covering many HCN features is used to calculate the HRI. The CO 2 line mixing in that range is accounted for as described by Duflot et al. (2013). A first HRI of HCN was already set up for the IASI observations by Duflot et al. (2015), but here we set up a new more sensitive one following the iterative procedure presented by Franco et al. (2018).
In contrast to Duflot et al. (2015), who used pre-calculated coefficients to link the HRI to the HCN total column, the ANNI v3 procedure implements an artificial feedforward NN for this purpose. Such a NN is set up to mimic in a comprehensive way 555 the complex connections that exist between the HRI, the state of the atmosphere and Earth's surface, and the gas abundance.
Setting up a NN requires a training phase, in which the NN learns from the presentation of an extensive dataset including all the necessary input and output variables. In ANNI v3, the NN inputs are the HRI, a spectral baseline temperature, the H 2 O columns, the temperature profile, the surface pressure and emissivity, and the IASI viewing angle, whereas the output is the HCN column. Here, we built this training set from over 250,000 synthetic IASI spectra simulated by a line-by-line 560 radiative transfer model. The advantage of such a synthetic training set is that it is free of the noise and/or scarcity of real measurements and that the spectra can be generated in large amounts in order to make the training set -and hence the NNrepresentative of all possible conditions. For example, the NN set up for HCN is trained to retrieve gas column from 1 ×10 14 to 15 ×10 16 molecules cm −2 . Actually, two separate synthetic datasets are assembled per target species, one being representative of conditions close to emission sources, the other of mixing/transport conditions (see Whitburn et al., 2016a;Franco et al., 565 2018, for the rationale). Each training set leads to the setup of a specific NN that is used to globally retrieve the target species in emission or transport regimes, successively. The training performances are similar to those of the other VOCs retrieved with ANNI v3 and are reached with a NN made of two computational layers, each layer deploying eight nodes.
In addition to the total column, the NN returns an associated error that is calculated via a perturbation method of the input variables (see Whitburn et al., 2016a). A pre-filter prevents the retrieval on cloudy scenes (cloud coverage > 10 %) or for 570 observations with missing ancillary data. Consistent with the other ANNI VOCs products, a post-filter discards the individual retrievals affected by too large uncertainties or poor measurement sensitivity to HCN, specifically when | column (HCN) / HRI (HCN) | > 8 ×10 15 molecules cm −2 (A1) or spectral baseline temperatures < 268 K. This post-filter is not (directly) driven by the gas abundance, but rather by the thermal contrast (Franco et al., 2020). Finally, the constant climatological background of target gas abundance that is not 575 accounted for by the HRI has been estimated as 1.85 ×10 15 molecules cm −2 for HCN (see Franco et al., 2018); this offset is thus added to the individual retrieved columns. Once set up, the NN is fed for each individual IASI observation with the appropriate input data. Here, we chose to use the ERA-5 reanalysis dataset (Hersbach et al., 2020) Table 3. Absolute changes in the tropospheric burden for each region and each species discussed. Regional differences are calculated for the main biomass burning season (see Table 1) and the global changes are calculated for the complete year of 2015. If not stated otherwise, absolute differences are given in Gg and relative changes are provided in parenthesis in %. Most radical burdens are presented in mol.