IASI is a nadir-looking high-resolution Fourier transform spectrometer onboard the polar-orbiting sun-synchronous Metop (Meteorological Operational) satellites. The first two IASI sounders were launched in 2006 and 2012 (Metop-A and -B, respectively). A third instrument is scheduled for launch in 2018 and will ensure at least 18 years of consistent measurements (2006–2023). IASI covers the entire globe twice daily (09:30 and 21:30 LT when crossing the Equator), with a relatively small elliptical footprint on the ground varying from 12km×12km (at nadir) up to 20km×39km (off nadir), depending on the viewing angle . Its large and continuous spectral coverage of the thermal infrared band region (645–2760 cm-1), its medium spectral resolution (0.5 cm-1 apodized), and its low instrumental noise (∼0.2 K at 950 cm-1 and 280 K) make it an invaluable instrument for monitoring atmospheric composition . CO is retrieved from IASI measurements using the FORLI (Fast Optimal Estimation Retrievals on Layers for IASI) software , which has been extensively described in . The retrieval of NH3 is based on a new and flexible retrieval algorithm, which relies on the calculation of a so-called hyperspectral range index (HRI) and subsequent conversion to a NH3 total column (moleccm-2) using a neural network . The retrieval also includes a full uncertainty analysis, performed by perturbing the input parameters (temperature profile, HRI, NH3 a priori profile, etc.) of the neural network. In this paper we use the ANNI-NH3-v2R-I version of the product, which relies on ERA-Interim ECMWF meteorological input data, along with built-in surface temperature . For a detailed description of the NH3 retrieval methods and the parameters used in the ANNI-NH3-v2R-I dataset, we refer the reader to  and . The validation of FORLI CO profiles and columns has shown good agreement overall using in situ, aircraft, and satellite observations . For NH3 columns, the validation has started but is more difficult considering the important spatial and temporal variability in NH3 and the scarcity of correlative ground- and airplane-based measurements in many regions of the world . Two studies, based on a previous NH3 retrieval algorithm also using the HRI but relying on two-dimensional look-up tables for the conversion into a NH3 total column (moleccm-2) , have shown fair agreement between IASI NH3 observations and other measurements (generally within the uncertainties of the IASI NH3 retrieved columns), with differences of about 60–80 % reported in  and of 30 % on average in .

This work makes use of 8 years (2008–2015) of daily global NH3 and CO total columns (moleccm-2) from the measurements of IASI onboard Metop-A. Only daytime satellite observations have been considered as they usually show a better sensitivity, especially to NH3. We have also assumed a similar sensitivity for IASI to NH3 and CO in the lower layers of the atmosphere. This is not expected to introduce a significant bias since it has been shown for both CO and NH3 that the peak sensitivity was in the lower layers of the atmosphere in case of positive thermal contrast, which generally prevails during daytime in the studied regions . A more important bias may result from the use of a unique vertical profile shape in the retrieval scheme of NH3 total columns, which is therefore not representative of the large variety of profiles observed above biomass burning plumes.  have calculated that the use of an alternative profile could affect the retrieved column by up to 50 %. This is important to keep in mind for the analyses presented next.

In support of the selection of the studied regions and the NH3 and CO columns, we also used active-fire detection data and nitrogen dioxide (NO2) total columns (moleccm-2). Detected active fires are taken from the Global Monthly Fire Location Product (MCD14ML, Level 2, Collection 5) developed by the University of Maryland from the measurements of the MODerate resolution Imaging Spectroradiometer (MODIS) onboard the NASA Terra and Aqua satellites . Active fires are monitored at a resolution of 1×1 km2, with fires as small as 100 m2 detected. NO2 that total columns are retrieved from the measurements of the Global Ozone Monitoring Experiment (GOME-2) also onboard the Metop satellites and working in the UV–visible spectral band region .

From the IASI NH3 and CO total columns (moleccm-2), we have derived NH3 ERs relative to CO (ERNH3/CO) defined as the ratio of the number of emitted molecules of NH3 (here the NH3 total column) to the emitted molecules of the reference species CO (here the CO total column) . The choice of CO as the reference species is particularly suitable here as it is a dominant species emitted by fires and has a lifetime of several weeks in the free troposphere. One main advantage of the ERs compared to the EFs is that ER calculation only requires simultaneous measurements of the studied (NH3) and the reference species (CO), while EF calculation requires fuel information that is not always available or completely reliable . In fire plumes, ERs can be estimated following  ERNH3/CO=[NH3]smoke-[NH3]ambient[CO]smoke-[CO]ambient. When a lot of measurements are available, which is often the case for IASI-derived measurements owing to its excellent spatial and temporal resolution, the average ERNH3/CO ratio can be estimated from the slope of the linear regression of NH3 vs. CO . The ERs can also be derived directly from the EFs by multiplying the EFNH3/EFCO ratio with the ratio of the molar masses MCO/MNH3 . This will be used here to convert the reported EF values from ground-based and airborne studies into ERs in order to allow comparison with our IASI-derived ERNH3/CO ratio.

One of the key steps in this study is the selection of the areas of interest for the calculation of the ERNH3/CO ratio. To be relevant, ERNH3/CO ratios need to be calculated for areas where fires are the dominant source of emissions of NH3 and CO. The selection has been done on a pixel basis. We have first calculated the linear regressions, globally on a 1∘×1∘ grid, between the monthly means of the pairs NH3–CO total columns (moleccm-2), NH3–NO2 total columns, and NH3 total columns–number of active fires (#fires). We have next selected the pixels for which a correlation coefficient (r) higher than 0.3 was found for the three pairs of regression (NH3–CO, NH3–NO2 and NH3–#fires). These are shown in Fig.  (colored pixels) and constitute the areas considered for the calculation of the ERNH3/CO ratio. Pixels with an r value higher than 0.3 for the considered pair but not for (at least) one of the two other pairs are shown in gray. The idea behind this selection procedure is that a good correspondence between the monthly means of NH3, CO, and NO2 total columns provides an indication of a dominant contribution of the fires to their emissions since biomass burning is indeed the only major common source of emissions of these three species. A significant positive correlation between the NH3 total columns and the detected number of fires adds an additional argument in favor of the contribution of fires and ensures keeping only those areas that are close to the source of emissions, making the comparison with ground-based and airborne-derived EFs and ERs easier. In general, the largest correlations are found between NH3 and CO total columns (Fig. , panel a), with correlation coefficients ranging from about 0.6–0.7 up to 0.9 in Africa south of the Equator and Indonesia. The fact that these two species are measured simultaneously from IASI could contribute to this. For the two other pairs (NH3–NO2 and NH3–#fires), the correlation coefficients are in the range of 0.3–0.8. Note that in general, significant positive correlations between NH3 and NO2 (Fig. , panel b) are only found close to the source of emissions due to the relatively short lifetime of NO2 of a few hours;. With a lifetime of typically 12–36 h in the studied regions , NH3 is more likely to be transported over longer distances. This can be seen on the NH3–CO correlation map on which positive correlations are also found over seas downwind of the source areas.

For each of the selected pixels, we have next calculated an ERNH3/CO ratio per year between 2008 and 2015 from the slope of the linear regression between NH3 and CO retrieved columns (moleccm-2). The method considered here for the calculation of the regression line was the ordinary least square fit. To take into account the NH3 and CO columns most likely related to fire emissions, we have only considered IASI measurements located within 50 km from a fire. We have also included a quality filter on the NH3 and CO measurements: only total columns with a relative error lower than 100 % for NH3 and 25 % for CO were retained for the regression. Finally, as a post-filtering, for the analysis we have only kept the ERNH3/CO ratios for which the linear regressions between NH3 and CO columns show a correlation coefficient larger than 0.3 and for which we have more than 10 measurements. The impact of these pre- and post-filters on the calculated ERNH3/CO ratio is discussed in Sect. . An example of a linear regression between NH3 and CO for one of the selected pixels (evergreen broadleaf forest – EBF – in Indonesia) is given in Fig. .

For this study, we focus on the four dominant biomes in the selected pixels. These were identified using the MODIS Land Cover Type product (MCD12Q1) with the 17-class International Geosphere–Biosphere Program (IGBP) classification  (Fig. ). The four selected classes are (1) EBF, (2) the woody savanna (WS), (3) the savanna (S), and (4) the crop together with the cropland/natural vegetation mosaic (C+CNVM), here denoted as C. Figure  also shows the distribution of the mean yearly ERNH3/CO ratio averaged over the time period 2008–2015 for the selected pixels. A first analysis of the distribution of the ERNH3/CO ratio reveals a variability between the four biomes, especially in Africa north of the Equator and in central South America, where a gradient is observed between EBF and WS and between EBF and S, respectively, with a higher ERNH3/CO ratio found for EBF. A clear gradient is observed as well in Africa south of the Equator from the northwest to the southeast.

The pixel-based ERNH3/CO ratios have next been grouped by biome to analyze their regional and temporal variability. In addition, to facilitate the study of the spatial distribution of the ERNH3/CO ratio, we have defined six main regions, which include the majority of the pixels of interest (see Fig. ). Two are in Africa, one north (AFR.NEQ.) and one south (AFR.SEQ.) of the Equator. One corresponds to the central part of South America (S.AM.). A second region in America (C.AM.) is located north of the S.AM. region and includes the region around the Gulf of Mexico, Central America, Colombia, and Venezuela. The last two regions are in Asia; one is for South-East Asia (SE. ASIA) and the second is for Indonesia (INDO.).

Here we analyze the spatial and biome variability in the ERNH3/CO ratio. For each of the four biomes (EBF, WS, S, C) and each of the six regions, a mean ERNH3/CO ratio was obtained by averaging all yearly pixel-based ERNH3/CO ratios in the time period 2008 and 2015 (Fig. , solid error bars). Mean ERNH3/CO ratios for the six regions globally are shown as well (horizontal lines). Overall, the highest mean ERNH3/CO ratio is found for EBF (15.3×10-3), while S and WS show mean ERNH3/CO ratios about 40–50 % lower, with values of 9.1×10-3 and 10.4×10-3, respectively. The larger ERNH3/CO ratio for EBF compared to WS and S is in agreement with previous studies e.g., and is mainly attributed to differences in fuel size and density: EBF, characterized by dense fuel, is indeed dominated by smoldering combustion, which emits more reduced or incompletely oxidized products (among them NH3 and CO) than grassland . One should note, however, that  reported higher ERNH3/CO ratios for S than for tropical forests. For C, the mean ERNH3/CO ratio (12.8×10-3) is close to the EBF ERNH3/CO ratio but is more difficult to interpret because the biome probably includes different types of fuel. Figure , representing the cumulative frequency of the pixel-based yearly ERNH3/CO ratio per biome, also shows the biome-trends in the ERNH3/CO ratio. EBF has, for example, about 40 % of the calculated ERNH3/CO ratios above 0.015, while this value corresponds to only about 5–10 % for S and WS. These differences in the ERNH3/CO ratio between biomes are, however, not necessarily found when looking at the average ERNH3/CO ratios on a regional scale. For SE.ASIA in particular, the differences between ERNH3/CO ratios are low (of the order of 5–10 %). For S.AM., the EBF ERNH3/CO ratio is about 60 % higher than the S ERNH3/CO ratio but close to the WS ERNH3/CO ratio (within 10 %).

When comparing the ERNH3/CO ratios by biome between the six regions in Fig.  (solid error bars), we find good agreements but also large differences, in line with what has already been reported by, for example, , , and . Among the most noticeable differences, we find that the EBF ERNH3/CO ratio for AFR.NEQ. is between 20 and 65 % higher than for the S.AM., C.AM., INDON., and SE.ASIA regions. Similarly, a large variability in the ERNH3/CO ratio is found for the WS and S biomes, ranging between about 7×10-3 for the AFR.SEQ. region and 19×10-3 (14×10-3) for S.AM. (C.AM.) for WS (S). For the C biome, almost no variability is observed, with ERNH3/CO ratios ranging between 11×10-3 for SE.ASIA and about 14×10-3 for C.AM and AFR.NEQ. Note that this intra-biome variability is also found within a given region, as observed in Fig.  and as evidenced by the sometimes large SD associated with the mean ERNH3/CO ratio (e.g., EBF in the AFR.NEQ. region with a SD higher than 0.01). As mentioned in Sect. , these differences can be explained by changes in the fuel type (size and density) but also the climate, weather, topography, moisture and N content, and fire practices. In addition for EBF, different regional deforestation practices could also lead to variation in the ERNH3/CO ratio . It should finally be mentioned that for the AFR.NEQ. region, the measured NH3 columns at the end of the fire period probably originate from the combination of both biomass burning emissions and another source, possibly agriculture as suggested in ; this might therefore introduce a bias in the ERNH3/CO ratio. Overall, these results clearly highlight the need for developing new regional-dependent EFs in order to improve the representativeness of estimations from bottom-up inventories.

The comparison of the regional IASI-derived ERNH3/CO ratios (Fig. , solid error bars) with the values reported in the literature from ground-based or airborne studies (see Table ) shows a good correspondence, especially for the EBF and the S–WS biomes in which ERNH3/CO ratios are generally within the range of values given in the literature. The only exception is for EBF for , who measured a ERNH3/CO ratio of about a factor 2–3 higher. The latter was however derived from tropical dry forest and is likely not representative for the complete EBF class. Note that for WS, the ERNH3/CO ratios are compared here to values reported for S, which are usually denoted as simply S in the literature in the same biome. For croplands, values reported in the literature are in contrast about 2–3 times higher than the one derived from IASI measurements. As mentioned before, the C biome probably includes different type of fuels, and results are therefore more difficult to interpret.

When looking at the mean ERNH3/CO ratios averaged over the six regions (Fig. , horizontal lines) for the four biomes, we find that the latter generally fall in the lower bound of the range given by the ERNH3/CO ratio reported in the literature. While an overestimation of the average ERNH3/CO ratio (or EFNH3) in the literature is possible, other reasons are likely to play a role. First, the differences with the IASI-derived ERNH3/CO ratio could also be (at least partly) explained by the consideration in our work of IASI measurements within 50 km of an active fire, while ground and airborne measurements are done in the direct vicinity of the fire. Second, another possible reason might lie in the difficulty for MODIS to detect smoldering fires, causing the IASI-derived ERNH3/CO ratio to preferentially reflect the flaming phase of the vegetation fires. Third, an accumulation of CO in the region during the fire period (due to its much longer lifetime compared to NH3) might introduce a bias in the calculated ERNH3/CO ratio. Finally, the differences with the reported ERNH3/CO ratio could also be due to the chosen methodology for the calculation of the ERNH3/CO ratio. To verify this, we have recalculated mean biome-specific ERNH3/CO ratios for the six regions (not shown) by varying one by one the pre- and post-filters considered before (see Sect. ). We have performed four tests: (1) with a maximum distance of the NH3 total column to a detected fire of 30 km and (2) 100 km (instead of 50 km), (3) with a maximum error on the NH3 total column of 75 % (against 100 %), and (4) by filtering out the ERNH3/CO ratios for which the linear regressions between NH3 and CO columns show a correlation coefficient (r) of the linear regression lower than 0.6 (instead of 0.3). We find a very limited impact of the distance to a fire and the error on the NH3 column allowed, with differences of only about 3–8 % (interestingly, an increase (decrease) in the tolerance of the maximum distance to a fire systematically slightly decreases (increases) the mean ERNH3/CO ratio). In contrast, an increase to 0.6 of the threshold for the correlation coefficient introduces a large increase in the mean ERNH3/CO ratio of about 13–28 %. Taking into account this increase, we find mean ERNH3/CO ratio closer to the middle range of what is reported in the literature, especially for WS and S. Finally, as we mentioned in Sect. , the retrieval of NH3 could be biased by the use of a constant NH3 vertical profile not representative of the variety of profiles observed above biomass burning plumes. Note that despite the impact of the pre- and post-filters chosen, the analysis on the regional and inter-biome variability in the ERNH3/CO ratios remains valid.

On a regional level (all biomes combined), a comparison with the satellite-derived ERNH3/CO ratios based on TES measurements  again shows an excellent agreement with our calculated ERNH3/CO ratio (Table ).  also derived ERNH3/CO ratios from simulations of the GEOS-Chem global chemical transport model. A good agreement is found between IASI and GEOS-Chem for the regions of AFR.NEQ. and S.AM., with ERNH3/CO ratios in the range of values calculated for north-central Africa and South America. For south-central Africa, in contrast,  reported ERNH3/CO values of about 2 times higher compared to our AFR.SEQ. region.

In this second part, we focus our analysis on the temporal variability in the ERNH3/CO ratio. Figure  shows the mean ERNH3/CO ratio averaged by biome and by year (2008–2015). The solid line represents the 8-year average for each biome. We find an interannual variability in the mean ERNH3/CO ratio up to a factor of 2 for the four biomes studied. The minimum ERNH3/CO ratio is found in 2013 for the S and WS biomes, while, for EBF, a minimum is observed in 2012. Interestingly, the highest mean ERNH3/CO ratio is observed in 2010 for all biomes (especially for EBF) except for C for which the maximum is found in 2011 (despite an ERNH3/CO ratio for 2010 also above the 8-year average). When analyzing the variability in the yearly averaged ERNH3/CO ratio for each region separately (Fig. ), we find that the high mean ERNH3/CO ratio of 2010 for EBF is exclusively carried by the AFR.NEQ. region, with a mean ERNH3/CO ratio of 35×10-3 (compared to about 20×10-3 for the other years in the region). For the WS biome, the peak in 2010 is mainly due to the S.AM., AFR.NEQ., and SE.ASIA regions, with a ERNH3/CO ratio about a factor of 1.5–2.5 higher compared to the other years. This important variability in the ERNH3/CO ratio is probably due to differences in the burning conditions from one year to another. One possible reason for the high mean ERNH3/CO ratio for 2010 in the different regions is the El Niño–Southern Oscillation (ENSO) event that occurred that year and that was responsible for severe droughts and increased fire activity in the regions studied . This is however probably not sufficient to explain the 2-fold increase for EBF for 2010 in the AFR.NEQ. region, but no clear evidence of other processes influencing the ERNH3/CO ratio was found for that year. Surprisingly, the same increase in the ERNH3/CO ratios is not observed for the year 2015, which was the strongest El Niño year since 1997 . For WS, high ERNH3/CO ratios are also observed for 2011 for South and Central America (S.AM. and C.AM.). However, this has a small impact on the global yearly ERNH3/CO ratio, which is mainly driven by the two regions in Africa (AFR.NEQ. and AFR.SEQ.), representing about 20 and 52 % of all the calculated ERNH3/CO ratios for WS, respectively (see Fig. ). For the S biome, the yearly ERNH3/CO ratio is largely dominated by the AFR.SEQ. and the S.AM. (52 and 35 % of the ERNH3/CO ratios, respectively). A peak is observed in the ERNH3/CO ratio for 2010 for the AFR.NEQ., the S.AM., and the C.AM. regions. Note that here the AFR.NEQ. and C.AM. regions also show high ERNH3/CO ratios for the year 2015, which tends to support the hypothesis of the influence of El Niño on the ERNH3/CO ratio. Finally, the C biome is mainly driven by the AFR.NEQ. and SE.ASIA regions (45 and 35 % of all ERNH3/CO ratios, respectively). For the AFR.NEQ. region, a peak is observed in 2010 and 2012, while for the C.AM. region, a maximum is reached in 2011, with a ERNH3/CO ratio about 50 % higher compared to the other years. This important variability in the ERNH3/CO ratio in time and space again highlights the importance of using dynamic EFs datasets in the fire emission inventories in order to better take into account the local fire conditions.

Finally, we investigate the temporal variability in the ERNH3/CO ratio from a seasonal perspective. For this, for each pixel selected in Sect.  we have calculated a separate ERNH3/CO ratio for the early and for the late fire season. The separation early or late fire season has been performed by analyzing the daily time series of the number of fires between 2008 and 2015 for each region and biome studied (not shown). The results are shown in Fig.  (dashed error bars). In general, we do not find a systematic difference in the ERNH3/CO ratio between the early and late fire season except for the AFR.NEQ. region, for which the late ERNH3/CO ratios are higher by about 20–40 % for the four biomes. This is in agreement with the hypothesis made in Sect.  of the presence of a secondary source of NH3 (possibly agriculture) towards the end of the fire season. The same difference in the ERNH3/CO ratio was also observed by , who found a 60 % increase between the beginning and the end of the fire season for north-central Africa. Finally, note that the early and late fire season ERNH3/CO ratios are generally close to the corresponding yearly ERNH3/CO ratios (within 10–30 %), which tends to support our methodology for the calculation of the ERs.