IASI is a hyperspectral sounder onboard the three polar-orbiting MetOp satellites, launched in 2006 (IASI-A, now decommissioned), 2012 (IASI-B), and 2018 (IASI-C) jointly by CNES (Centre National d’Études Spatiales) and EUMETSAT (European Organisation for the Exploitation of Meteorological Satellites). The main objective of IASI is to provide a continuous and long-term collection of measurements to support meteorology, namely temperature and humidity tropospheric profiles with high vertical resolution and precision, but it has also been used to monitor both the environment and climate on a global scale through observations of atmospheric composition .

IASI measures the spectrally resolved thermal infrared radiation emitted by the Earth and the atmosphere system at the top of the atmosphere (TOA) across the spectral range 645 to 2760 cm⁻¹ at a spectral resolution of 0.5 cm⁻¹ and spectral sampling of 0.25 cm⁻¹. Due to its high spectral resolution and spectral sampling, IASI is able to measure a large number of trace gases, including some species observed only sporadically in the measured spectra . These species include 10 gases with clear spectral signatures that are always present in the satellite measurements, CO₂, N₂O, CFC‐11, CFC‐12, OCS, H₂O, CH₄, O₃, CO and HNO₃ and 14 reactive trace gases. Some of these reactive gases are the typical products of biomass burning, such as HCN, which has been assessed in previous IASI studies, and is retrieved using the absorption band centred at ∼ 712 cm⁻¹ (ν2 band) .

The sensitivity of IASI retrievals near the surface is limited and exhibits a strong dependence on thermal contrast (with large differences between daytime and nighttime), which constrains the amount of vertical information that can be obtained. For CO, the averaging kernels indicate two distinct vertical contributions: one from the lower troposphere, peaking on average at about 5 km, and another from the upper troposphere, near 10 km. The top panels of Fig.  show the CO averaging kernels (solid lines) and total column averaging kernel (dashed lines) for IASI measurements on 2 November within the wildfire plume (110° E, 14.5° S) and outside the plume (95.4° E, 35° S). In contrast, HCN retrievals are primarily sensitive to the mid to upper troposphere, with peak sensitivity between 9 and 12 km, as shown in the bottom panels of Fig.  reporting HCN averaging kernels (solid lines) and total column averaging kernel (dashed lines) for in-plume and out-of-plume scenarios.

The concentrations of HCN and CO during the 2015, 2019 and 2023 fire seasons were retrieved using the University of Leicester IASI Retrieval Scheme (ULIRS). The ULIRS scheme is a retrieval algorithm originally developed at the University of Leicester and subsequently by the UK National Centre for Earth Observation to study tropospheric trace gas concentrations from IASI satellite measurements. Here we examine results for carbon monoxide and hydrogen cyanide under clear sky and partially cloudy conditions from the radiances measured by the IASI instrument. A full description of the ULIRS and a summary of the main features are given in , with relevant updates outlined in this work. In essence, the ULIRS retrieves atmospheric trace gas partial and total column information from IASI top-of-atmosphere radiances. To do this, the scheme incorporates an optimal estimation method to constrain the inversion with a priori information about the variables to be retrieved, in this case CO and HCN, alongside H₂O, CO₂ and temperature (profile and surface).

The CO retrieval setup largely follows the setup of but with a single global average a priori profile constructed with ACE v3.6 data covering 2004 to 2018 for the stratospheric profile at one-sigma uncertainty (above 15 km). This was merged with the troposphere (below 15 km) using TOMCAT model data over the 2006 to 2015 period. The number of retrieval levels compared to was also reduced to 17 equidistant pressure levels from the surface up to 50 hPa. The ULIRS CO utilises optimal estimation in common with other IASI schemes such as FORLI-CO (Fast Optimal Retrievals on Layers for IASI) which produces the operational CO product for the EUMETSAT AC-SAF. The main difference being in their vertical grids, a priori assumptions, and forward modeling methods. Both schemes use look-up-tables of absorbances to speed up the radiative transfer calculations.

The HCN retrieval employs a fixed altitude grid with 13 levels, 3 km spacing from 0 to 18 km with levels then at 7 km spacing up to and including 60 km. Due to the weak HCN absorption, ULIRS retrievals are calculated in a limited spectral region, 710–715 cm⁻¹, containing the ν2 band of HCN in order to avoid interferences from the absorption of other species in the spectrum, in particular avoiding line mixing effects of a CO₂ Q branch centred near 720 cm⁻¹. Only 11 selected spectral channels are used in the process: 710.25, 710.50, 710.75, 712.50, 712.75, 713.50, 713.75, 714.00, 714.25, 714.75, 715.00 cm⁻¹. The a priori profile was constructed from INTEX-B aircraft information in the troposphere and ACE v4 data in the stratosphere over polluted scenes. A loose constraint on the HCN a priori uncertainty of 100 % was assumed. Here we limit both the CO and HCN retrievals to a maximum of 10 iterations. Other schemes that produce HCN products from IASI include the ANNI (Artificial Neural Network for IASI) , which provides rapid, global, total column mapping. While ULIRS retrievals often have a Degrees of Freedom for Signal (DOFS) near 1 for HCN, suggesting limited strict vertical profiling, the scheme still provides sensitivity to broad vertical distributions rather than a fixed total column. This allows for analysis of vertical partitioning and high-altitude transport (e.g., biomass burning plumes) in the troposphere.

ULIRS makes use of the Reference Forward Model (RFM), a line-by-line (LBL) radiative transfer model developed at the University of Oxford , which can be used to simulate the TOA spectrum as measured by a spaceborne sensor. Line-by-line methods for numerically solving RT equations, such as the RFM, can be time-consuming and require large computational power. Retrieval schemes generally require much faster output, so to improve RFM speeds, a set of absorption coefficients for each absorbing gas is calculated beforehand and interpolated onto the IASI grid. RFM calculations in the ULIRS CO and HCN retrievals make use of spectroscopic data in the form of derived look-up tables (LUTs) including H₂O and O₃. For CO, LUTs were used from . For HCN these were derived using a set of Curtis-Godson equivalent pressures and nine temperature levels spanning ±60 K from the mid-latitude reference atmosphere of . These climatological reference atmospheres cover tropical, mid-latitude day/night and polar summer/winter atmospheric conditions. Testing the RFM radiance output via the LUTs, using the climatology files of , yielded results within 1 % of the full LBL calculations. A summary of the main information of the CO and HCN retrieval is present at Table S1 of the Supplement.

The Maido Observatory (21.08° S, 55.38° E; ∼ 2160 m a.s.l.), located on Réunion Island in the Indian Ocean, is a high-altitude site of the Network for the Detection of Atmospheric Composition Change (NDACC), operational since 2012 and within the programme since 2013. This site produces long-term observations of atmospheric composition in the tropical Southern Hemisphere. Being located on a mountain ridge above the marine boundary layer (MBL), Maido provides conditions suitable for atmospheric composition measurements representative of the free troposphere (FT) and the upper troposphere–lower stratosphere (UTLS) . A high-resolution Bruker IFS 125HR Fourier transform infrared (FTIR) spectrometer acquires direct solar absorption spectra. These observations enable the retrieval of total and partial columns of several trace gases, including CO and HCN .

The Maido Observatory is also appropriately located to observe the outflow of biomass burning pollution from Southern Africa , from South America for long-lived species such as CO and HCN, and from South-Eas Asia . To do this the Reunion Maido site makes regular, year-round measurements of trace-gas species including CO and HCN, using the optimal estimation method as implemented in SFIT4 (v0.9.4.4). The FTIR vertical profiles of volume mixing ratios are weighted by the airmasses in each retrieval layer and integrated to give the total or partial columns in molec. cm⁻². More detail on ground-based the FTIR dataset, retrieval method, and error budget can be found in .

The datasets produced are widely used to investigate long-range transport and the impact of biomass-burning emissions over the tropical Indian Ocean, satellite measurements, and model validation . These CO and HCN FTIR data sets are well suited for the validation of our ULIRS products.

The TOMCAT 3-D chemical transport model (CTM) is a global offline Eulerian model widely used for both tropospheric and stratospheric chemistry studies. Originally developed as two separate models TOMCAT and SLIMCAT , the unified model has since been applied in numerous studies . TOMCAT has been included in several model intercomparison studies of tropospheric chemistry transport and has shown good agreement with other models . The model's meteorological forcings – humidity, temperature, and wind fields – are driven by ERA-Interim reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF), provided at a 6 h temporal resolution . These meteorological variables are linearly interpolated to match the temporal resolution and spatial grid of the model. Given the model resolution, the higher spatial resolution of ERA5 compared with ERA-Interim is not expected to significantly affect the simulated transport. Large-scale vertical advection is not driven directly by the reanalysis vertical velocity fields; rather, it is diagnosed from the reanalysis horizontal divergence after averaging to the model grid. Sub-grid-scale transport associated with convection and boundary layer mixing is represented through dedicated parameterizations. No explicit treatment of pyroconvection is included. Surface emissions from both natural and anthropogenic sources are incorporated at their original resolution and subsequently re-gridded to align with the model's spatial configuration. HCN emissions are derived from several key datasets: anthropogenic and oceanic emissions are sourced from the Coupled Model Intercomparison Project Phase 6 (CMIP6) ; biogenic emissions are taken from a fixed annual dataset provided by the Chemistry-Climate Model Initiative (CCMI) ; and biomass burning emissions are obtained from the GFED Version 4.1 . A complete description of the tracer version of TOMCAT adapted to simulate global atmospheric HCN distributions has been extensively described in and , and evaluated against independent measurements from ACE-FTS and NDACC. This study can be considered a follow-up to previous work; therefore, we used the same setup to ensure consistency with the previous analysis. In the present study, TOMCAT outputs are compared with IASI observations.

This version of the model makes use of the best-fitting ocean uptake scheme defined in , the fluxes reduced by 75 % and the reaction rates proposed by for HCN oxidation by OH radicals and O(¹D). Outputs for 2015 were simulated on a horizontal 2.8° × 2.8° (T42 Gaussian) grid on 60 terrain-following vertical levels (surface to ∼ 60 km) for HCN. In order to compare the model output (T_HCN) with the IASI daily measurements over a short time period, the model was sampled daily and adapted for use with the monthly GFED v4.1 inputs for HCN biomass burning emissions. The TOMCAT model reads the emissions as monthly means, which are then interpolated in time so that the emissions vary smoothly during the model run.

TOMCAT total columns were constructed from the raw output using the approach proposed by , in which the model profiles are interpolated onto the ULIRS retrieval grid and smoothed by applying the ULIRS averaging kernels as xTOMCATsmooth=xa+A(xTOMCATint-xa), where xa is the a priori profile, A the retrieval averaging kernel matrix, and xTOMCATint the model profile interpolated onto the retrieval grid.

The SMAP mission, launched by NASA in 2015, is designed to provide global measurements of surface soil moisture and freeze-thaw conditions every 2–3 d . Although the radar instrument ceased operation shortly after launch, the radiometer continues to function reliably, allowing the generation of global soil moisture maps. SMAP data support a better understanding of the Earth's water, energy, and carbon cycles by quantifying the amount of liquid water in the topsoil layer and distinguishing between frozen and unfrozen ground. These observations are essential for improving weather forecasting, climate modeling, and environmental monitoring by capturing temporal changes in soil moisture that influence land-atmosphere interactions and broader ecological processes.

Recent studies have further expanded the scientific utility of SMAP data. demonstrated that SMAP-derived surface soil moisture variability is a reliable predictor of subsurface water dynamics. validated downscaled SMAP products at 1 km resolution using long-term in situ data, confirming their accuracy across heterogeneous landscapes. Additionally, improved SMAP’s performance in dense vegetation regions by calibrating retrieval algorithms to reduce bias. also enhanced SMAP’s global utility by integrating the data with ASCAT observations through machine learning techniques, resulting in higher resolution and more temporally consistent soil moisture estimates.

The operational SWI is produced by the Copernicus Land Service from the surface soil moisture (SSM) measured by the ASCAT scatterometer onboard the MetOp satellites and from Sentinel-1 C-band SAR. The SWI quantifies from these observations the moisture content of the soil with a spatial resolution of 0.1°. It is a dimensionless index ranging from 0 (dry) to 1 (saturated relative to local climatology), at eight different depths as a function of the characteristic timelength T (1, 5, 10, 15, 20, 40, 60 and 100 d), a parameter expressed in units of time but proportional to the depth of the layer, as described in .

Differently from other products such as SMAP, the SWI is more an indicator of the percentage of relative wetness than an absolute soil water content. It is not limited only to surface soil moisture, as it is able to evaluate water availability at depths typical of plant roots. For this reason, SWI is a particularly suitable parameter for assessing GWL conditions, since it better reflects the integrated moisture status of the soil profile and the overall water availability within the root zone.

The GFED4 database quantifies biomass burning emissions by coupling satellite-derived burned area (MODIS), active fire detections, and biogeochemical models . The GFED framework estimates dry matter (DM) consumption as a function of fuel load, combustion completeness, and fire activity across different biomes on a monthly 0.25° global grid. The trace gas emissions are derived by applying biome-specific emission factors derived from field and lab measurements to the DM consumption . The spatial distribution of the DM burned is then determined by using satellite observations of BA in combination with biogeochemical modelling. Cloud cover and a limited knowledge of the biome distribution and emission factors over Indonesia and other peat-dominated regions can cause possible errors in estimating fire emissions . Another limitation of the approach used to define GFED is its dependence on information about peat consumption by fires – such as the rate, extent, and depth of peat burned – which cannot be easily determined from satellite data .

FINNpeatSM is a bespoke regional dataset for peat fire emissions in Indonesia. FINNpeatSM uses Fire Inventory from NCAR (FINNv1.5) for fire detections of above ground vegetation fires. FINNv1.5 combines active fire detections from MODIS, biomass burned and emission factors (EFs) to provide daily fire emissions at 1 km resolution . However, FINNv1.5 only includes emissions from the combustion of above-ground vegetation. Therefore, it does not include emissions from combustion of peat. added the emissions from the combustion of peat across Indonesia, to create FINNpeatSM. Details on the method can be found in . In summary, to add emissions from peat combustion they use a 2-step process. First, they use a peatland distribution map to identify where a fire occurred on peatland. For each fire occurring on peatland they add additional emissions from the peat burning using Eq : 1Es=BA×BD×ρ×EFs where Es is the emissions of a species (s) from an individual fire that occurred on peatland, BA is the burned area and BD is the burn depth, ρ is the peat density and EF_s is the emissons factor of a species (s).

The second step of the method was to scale burn depth relative to soil moisture (from the ESA CCI Soil Moisture Product New Version Release (v04.4): ESA CCI SMv04.4). This step accounts for burn depth increasing as peat dries out and the water table decreases. In FINNpeatSM burn depth is assumed to increase linearly between a minimum of 5 cm to a maximum of 37 cm. used soil moisture from ESA CCI SMv04.4, which provided soil moisture retrievals up to 2018. To extend the FINNpeatSM dataset to 2023, updated the method to use the SMAP level 4 product. In the Level 4 dataset, SMAP measurements of soil moisture in the top 5 cm of the soil column are combined with estimates from a land-surface model. This provides soil moisture in the top 1 m of the soil column. The SMAP is both spatially (9 km) and temporally (3-hourly) complete, the 3-hourly data were aggregated to generate daily-mean values. Daily-mean SMAP values are used to linearly scale the burn depth between a minimum soil moisture of 0.5 m³ m⁻³ (5 cm burn depth) and 0.1 m3m⁻³ (37 cm burn depth). In this study, we use this version of FINNpeatSM to explore changes in fire burn depth and emissions between 2015, 2019 and 2023.

The Visible Infrared Imaging Radiometer Suite (VIIRS) active fire product (VNP14IMG) provides the latitude, longitude, time, and a confidence flag for a fire detection pixel. The VNP14IMG product is a development of the MODIS thermal anomaly algorithm and several studies have now shown that it is capable of detecting small fires at a 1 km spatial resolution which MODIS does not have the sensitivity to measure. VIIRS fire products are provided at 375 m, a much higher spatial resolution compared to MODIS, and therefore VIIRS is able to detect small wildfires that MODIS is insensitive to. Only nominal and high confidence fires during nighttime were included in this work.

IASI satellite observations are used to estimate the amount of HCN emitted from the Indonesian region during the 2015 wildfire season; the plume of HCN emitted during the wildfire season is clearly visible in the left panel of Fig. . In Sect. , these observations are compared with those from the more recent 2019 and 2023 burning seasons, both of which also occurred under El Niño conditions.

According to the new peatland fire stages classification – stage 1 surface fire, stage 2 shallow peatland fire (depth < 20 cm) and stage 3 deep peatland fire (depth > 20 cm) – proposed by , the Indonesian fires of 2015 were classified as stage 2 for the second half of August and stage 3 from early September to the end of October. This is broadly consistent with the averaged total column of the IASI HCN observations seen in Fig. , where the highest HCN total columns are observed in September and October. The HCN total columns peak in late October/early November, consistent with the work of on CO emissions from the same peat fires. Their results indicate a sudden increase in CO emissions for the latter half of October 2015.

During neutral conditions, Indonesia is situated under one of the rising branches of the Walker circulation, linked to high rainfall. However, during El Niño, the circulation shifts eastwards and the air over Indonesia tends to sink, resulting in below average rainfall, higher surface pressure, and dryness. A plot of vertical winds at 0° latitude (Indonesia spans 95° E–140° E longitude) derived from ERA5 for October 2015 (Fig. ) demonstrates this; the air mass over Indonesia is primarily descending. At the end of October/beginning of November, the circulation pattern suddenly changes to one in which the air is predominantly being uplifted. This is clearly observed in the top panels of Fig. , which shows the monthly mean ERA5 vertical wind vectors averaged along the equator for October 2015 and November 2015. The end result is that in late October, HCN from the peat fires is more readily uplifted to higher altitudes, coinciding with the region of the atmosphere where the IASI sensitivity is greatest, in the upper troposphere between 8 and 12 km. In addition to the increased emissions at the end of October, this contributes to the plume appearing most prominent in the first days of November 2015. The end of October 2015 also coincides with the onset of increased precipitation and an increase in soil moisture. Figure a shows that Indonesian deep soil moisture (SWI at the lowest layer corresponding to the characteristic timelength T=100, described in Sect. ) for August–October 2015 was at its lowest level over the last decade, with a minimum at the end of October. The choice of the parameter T=100, the lowest layer of SWI, allows us to evaluate the water availability at the plant root depth. This strongly suggests that the GWL during the 2015 El Niño season was at a lower level than in any other year over the last decade. Precipitation for the same period was at the lower end of typical values for the given months as shown in Fig. b using IMERG data.

In order to estimate the HCN emissions from the 2015 Indonesian peat fires, retrieved HCN total columns from IASI are compared with outputs of a tracer version of the TOMCAT 3D CTM.

Figure  compares the averaged total column time series of HCN, calculated over Indonesia in the regional box [12° S–7° N, 90° E–127° E] for both IASI measurements (black line) and the TOMCAT model outputs smoothed using the IASI averaging kernels (blue line). The box size was defined to cover Indonesia and, at the same time, capture the wildfire emission plumes. The IASI time series exhibits two gaps in November 2015 due to missing measurements. These same gaps also appear in the model time series, as the absence of IASI averaging kernels during this period prevents the smoothing of the model outputs. The T_HCN model run does not compare very well with the IASI measurements, showing a large overestimation of the HCN amount during the period between September and November 2015. In particular, the large peak simulated for T_HCN during the last week of September is completely absent from the IASI measurements. The left panel of Fig.  shows the HCN total column distribution on 2 November 2015 produced from T_HCN. The HCN amount produced is globally overestimated compared with the IASI measurements reported in Fig. , in particular over the latitude band between 20° N and 40° S. The discrepancy observed between the HCN TOMCAT model and IASI measurements is seemingly caused by the HCN emissions used to drive the model, shown in Fig. , which combine GFED v4.1 (the main component), CMIP6 and CCMI as described in . These emissions indicate an anomalously large peak of more than 4×1011 molec. cm⁻² s⁻¹ in September 2015. This value is more than double the October emissions, during which the largest HCN concentrations were observed over Indonesia during the 2015 wildfire season.

In order to seek better agreement between model and observations, we performed a number of model sensitivity runs. These suggested that the most sensitive component for improving the model simulations was the September GFED HCN emissions. For this reason, a new model run was performed (T_HCN_0.25Emiss) in which the September 2015 emissions were reduced by 75 %, based on the analysis performed in . The orange line in Fig.  indicates the averaged total column (smoothed by the IASI HCN averaging kernels) over Indonesia for the T_HCN_0.25Emiss model run, showing a substantial reduction in HCN concentration during September and October and reaching a very good agreement with the IASI measurements. However, the background HCN values, e.g. in December, are underestimated in the model, possibly due to neglect of additional sources. The global HCN total column distribution modelled on 2 November 2015, as shown in the bottom right panel of Fig. , shows a significantly improved agreement with the IASI measured HCN total columns in Fig. .

The study of CO emissions from the 2015 Indonesian wildfire season by found that GFED is not good at reproducing emissions from large peat fires. GFED estimates monthly fire emissions from satellite-observed burned area and temporally disaggregates them within each month using fire radiative power as a proxy for combustion intensity and timing. For peat-dominated fires, burned area is more sensitive to the initial burning stages and less sensitive to peat burning that occurs underground. Our work supports this conclusion with respect to emissions of HCN from peat fires.

The Indonesian peatland fires from 2015 present an opportunity to estimate the HCN wildfire emission factors from satellite data. The fires burned for several months over a limited area, with the prevailing wind conditions resulting in the emitted plumes being predominantly transported over the Indian Ocean. The relatively long lifetimes of HCN and CO, of the order of 2-5 months, allow us to take advantage of satellite measurements over the ocean, where surface emissivity is reasonably homogeneous, and where the polluted air has been uplifted to higher altitudes, corresponding to the maximum sensitivity of IASI to HCN and CO perturbations. Changes in thermal heating over land between day and night cause a change in vertical sensitivity of IASI to HCN and CO. The effect over the oceans is much smaller due to the larger heat capacity of water. As such, in this work we use an average from both IASI daytime and night time measurements to maximise coverage and reduce the random uncertainty of the results.

The emission factor (EF_X) is a quantity defined as the mass of a trace gas (X) emitted per kilogram of biomass burned (g kg ⁻¹ dry matter). In order to derive EF_X, it is first necessary to derive a related parameter called the enhancement ratio (ER) which can be determined directly from satellite measurements. The ER_HCN∕CO is defined as a ratio of the emitted number of molecules of HCN over the emitted number of molecules of CO , 2ERHCN/CO=[HCN]plume-[HCN]background[CO]plume-[CO]background CO is chosen as the reference gas in this work as it has a similar lifetime to HCN and is emitted from similar sources. The excellent daily coverage of IASI allows a daily estimation of ER_HCN∕CO to be performed. We have chosen the region 12° S–7° N, 90° E–127° E, located downwind of the Indonesian emissions from Sumatra and Kalimantan; only data with a correlation coefficient greater than 0.3 (following threshold methodology of ) are included in our analysis. There is no clear trend in the ER across the September 2015 to December 2015 period (Fig. ), although there is some variability with lower ER in early September, where the correlation between HCN and CO measurements is less than 0.4. The phase of fire is a key factor in the emission of pyrogenic species. HCN and CO emissions are generally higher during the smouldering phase of fire (lower FRP) with smouldering peat fire emission factors being approximately 10 times those of flaming savanna fires . Peat fires generate weakly buoyant smoke plumes that generally accumulate close to the ground, although the emissions can migrate great distances via the prevailing winds. The lower ER in early September may be due to the fact that the fires were only recently ignited and high HCN (and CO) levels were closer to the ground. IASI is more sensitive to near surface CO than HCN, so HCN is likely underestimated at that time. Throughout October, the HCN : CO relationship shows correlation coefficients greater than 0.4. By early November, the fire activity ended abruptly, but the correlation remains greater than 0.3 throughout most of November, until the HCN levels drop below the sensitivity of the IASI instrument.

It is estimated that 90 % of fire emissions from the late 2015 fire events were dominated by emissions from peat soils , with the remaining 10 % from tropical forest. Using a range of ERs derived for differing correlation coefficients, we are able to estimate HCN emission factors, EF_HCN∕CO, from the derived ER_HCN∕CO. Since we are investigating a region close to the emissions, and are not influenced by transport from other regions, the two parameters are related by the equation 3EFHCN=ERHCN/CO×MMHCNMMCO×EFCO. The quantities MM_HCN (27.0253 g mol⁻¹) and MM_CO (28.01 g mol⁻¹) are the molar masses of HCN and CO, respectively. There are a number of datasets of EF_CO for peat over Indonesia (Table ). measured EF_CO in situ (291 ± 49 g kg⁻¹), noting that their measurements were solely from smouldering combustion. also measured EF_CO in situ (255 ± 39), but noted a few smoke measurements involved occasional contributions from small clumps of ignited dry vegetation, which may account for the smaller value of EF_CO. More recently, revisited Kalimantan in 2019, during El Niño conditions, to sample fires burning only peat and measured a 50 % larger EF_CO compared to the work in a similar region. Both and took samples from Sumatra and Kalimantan respectively and burned the samples in the lab, both deriving significantly lower EF_CO values. The final source is from a meta-analyses study by who use the data from the in situ and laboratory studies previously listed to derive an updated EF_CO based on methodological differences in measurements and weighted to account for different sample sizes.

Examining the relationship between HCN and CO total column amounts in Fig. , the correlation coefficient between the two species varies between 0.3 and 0.8 over most of the September to November 2015 period. We use the correlation coefficient as a filtering metric here to understand whether we see a change in the ER depending on how closely related CO and HCN concentrations are in the same air parcel. A summary of the effect on estimating EF_HCN by including only measured data exceeding a chosen correlation coefficient threshold is shown in Table . Generally, the lower correlation coefficients are found to be associated with lower ER values. The highest EF_HCN values are derived where the correlation coefficient exceeds 0.8, ranging between 1.04 and 1.71 g kg⁻¹ depending on which value is used for EF_CO. Where the correlation coefficient HCN:CO is greater than 0.5, we see that the EF_HCN estimates decrease, ranging between 0.76 and 1.19 g kg⁻¹. Both ends of the estimates are lower than the (surface) in situ and lab-based estimates of EF_HCN. We know that the IASI HCN sensitivity is in the upper troposphere between 8 and 12 km, with very low sensitivity to high concentrations below 4 km which may lead to an underestimate of EF_HCN. show that the MLS CO at 215 hPa (∼ 12 km) increased steadily through September with a rapid increase to CO exceeding 400 ppbv at the end of October. The highest correlation HCN : CO in our results is during this late October period, where we see enhanced vertical uplift of both gases and therefore increased sensitivity to the HCN emissions. Before mid-October, we likely underestimate the HCN amounts as plumes are closer to the surface. This alone cannot explain the large discrepancies between the IASI measurements and the TOMCAT simulations observed in Fig. , since the TOMCAT outputs have already been smoothed using the IASI averaging kernels to ensure comparable sensitivity.

Peatland fire emissions from GFEDv4s include a simple representation of burn depth based on soil moisture. However, it is not possible to use the dataset to explore soil moisture and burn depth interactions across different years. Therefore limiting our ability to explore the drivers of differences in HCN retrievals between 2015, 2019 and 2023. To address this limitation in GFEDv4s, we examine fire detections (from VIIRS fire hotpots) and SMAP over Indonesia peatlands. Using the FINNpeatSM method , we can combine these datasets to estimate the burn depth of fires in 2015, 2019 and 2023.

Daily VIIRS temperature hotspots are only used if there is a medium or high confidence in the retrieval and if the fire occurred on peatland in Indonesia. For each retained fire detection, the nearest-neighbour daily mean soil moisture pixel (from SMAP) is determined. The associated soil moisture from an individual fire detection can be used to estimate the burn depth, and emissions, of the fire. The emissions in FINNpeatSM are linearly related to burn depth, so that the deeper a fire burns the higher the emissions.

First, we consider daily mean soil moisture from SMAP across all Indonesian, Kalimantan, and Sumatra peatlands for 2015, 2019 and 2023 (Fig. ). We consider Kalimantan and Sumatra peatlands in addition to all Indonesian peatlands since these regions dominate fire emissions. In 2015 62 % of total emissions were from Kalimantan and 33 % were from Sumatra . Soil moisture is shown for 1 June to 31 December, with the dry season (1 August to 31 October) shaded in grey, since the dry season is when fires are most likely to occur. Across Indonesia and both Kalimantan and Sumatra soil moisture in 2015, 2019 and 2023 were similar from June–July (0.58–0.61 m³ m⁻³). However, from July to August, soil moisture across all regions in 2015 and 2019 begins to decrease rapidly (from 0.52–0.55 to 0.46–0.48 m³ m⁻³), while in 2023 soil moisture remains high (0.59 to 0.54 m³ m⁻³). From August to mid-September soil moisture continues to decrease across Indonesia, driven by reductions in soil moisture in Kalimantan in 2015 and 2019 (Indonesia from 0.46–0.48 to 0.36–0.39 m³ m⁻³ and Kalimatan from 0.39–0.44 to 0.27–0.29 m³ m⁻³). In contrast, over the same period in 2023, soil moisture remains high across Indonesia (decreasing from 0.54 to 0.50 m³ m⁻³), driven by high soil moisture in Sumatra peatlands. From mid-September to mid-October soil moisture continues to decrease in 2015 (from 0.39 to 0.33 m³ m⁻³), while in 2019 there is an abrupt increase in soil moisture (from 0.36 to 0.47 m³ m⁻³), closer to the 2023 values (from 0.50 to 0.54 m³ m⁻³). Overall, soil moisture is lowest throughout the dry season in 2015 (1 August to 31 October mean of 0.40 m³ m⁻³), while in 2019 soil moisture is very low at the start of the dry season but then increases in the late dry season (1 August to 31 October mean of 0.43 m³ m⁻³). In contrast, soil moisture in 2023 is the highest of the 3 years (1 August to 31 October mean of 0.52 m³ m⁻³), in agreement with what was observed in SWI data (Fig. a).

The collocated daily mean soil moisture and daily total number of fire hotspots (from VIIRS) for the same time period (1 June to 31 December 2015, 2019 and 2023) is shown in Fig. . This clearly indicates that 2015 had the highest number of fire hotspots (> 40 collocated pixels with 1000–10 000 fire hotspots) occurring in areas where soil moisture was low (0.15–0.35 m³ m⁻³). However, in 2019, there are fewer fire pixels with a high number of hotspots (24 pixels with 1000–10 000 fire hotspots) and they generally had higher soil moisture (0.25–0.31 m³ m⁻³). In 2023, there were both fewer pixels with a high number of fire hotspots (peaking at 100–1000 fire hotspots), and the soil moisture where fire hotspots occurred was higher (0.21–0.4 m³ m⁻³).

The soil moisture at each fire location can be used to calculate peat burn depth of the fire, using the method developed by . Figure  indicates the burn depth of fires, and the date they occurred. As shown in Fig. , 2015 has the highest number of fires (up to 25 000 fires) that occur in low soil moisture areas (0.3 m). Figure  indicates that these fires burn deep into the peat below (0.25–0.3 m) and occur between September and mid-October 2015 (in line with Fig. ). In contrast, in 2019, there are fewer fires (0–15 000), the burn depth is shallower (0–0.25 m), and the fires that burn the deepest occur earlier in the dry season (mid-September) when soil moisture is higher than 2015 (Fig. ). In 2023, far fewer fires (0–5000 fires) burn deep into the peat below (0.05–0.25 m) and deep burning occurs in mid-September to early-October, in line with increased soil moisture in 2023 (Fig. ).

The accumulated daily burn depth of all Indonesian peatland fires from 1 June to 31 December is shown in Fig. . This plot accounts for both the number of fire hotspots and the burn depth of individual fire hotspots. At the start of the dry season (August to mid-September) the accumulated burn depth in 2015 and 2019 is similar (500–1000 m), compared with < 100 m in 2023. This is likely because soil moisture at the start of the dry season (August to mid-September) is similar in both years, and both years have a similar number of fire hotspots at this time. However, from mid-September, 2015 and 2019 deviate. Accumulated burn depth remains high in October 2015, whereas it substantially decreases in 2019. This is likely driven by the large decrease in both fire hotspots and peatland soil moisture in 2019, compared with 2015 where the number of fire hotspots remains high and the peatland soil moisture remains low.

As illustrated in Fig. , the time series of accumulated daily burn depth, along with the averaged total column of HCN and CO, exhibit similar temporal patterns, a progressive increase from August through October, reaching a peak in late October, and subsequently declining sharply in November.