2.1 Field measurements

2.1.2 Sample coverage and literature studies

The dataset obtained using the abovementioned UAS methodology includes both previously published data collected in Mozambique, South Africa, and Brazil (Russell-Smith et al., 2021; Vernooij et al., 2021, 2022b) and new measurements from xeric and mesic savannas in Botswana, Zambia, and Australia measured during the fire seasons of 2021 and 2022. The measurements cover three continents and the full length of the dry season, ranging from early dry season (EDS) campaigns in which fuel conditions sometimes prevented successful ignition to late dry season (LDS) campaigns with high-intensity fires. The 129 fires that we measured using the abovementioned methodology were supplemented with 85 previous savanna fires for which EFs of the measured species were reported in the updated database by Andreae (2019). This literature compilation only includes samples taken within minutes after emission to avoid significant chemical changes during atmospheric ageing. For the comparison with geospatial data, we only included fires for which the fire date and coordinates were provided, a prerequisite to get relevant satellite features. These criteria mean that laboratory studies, satellite studies covering wider regions, and most aircraft campaigns were excluded. Figure 1 provides an overview of the UAS (red for previously published and orange for our new measurements) and literature (blue) sample locations included in the study.

Figure 1Overview of sampling locations used for the analysis. The previously published (red) and new (orange) UAS measurements and the locations of the included literature studies of savanna fire emission factors listed in Andreae (2019) (blue) are shown. The shaded green area shows the distribution of savanna and grassland fires over the 2002–2016 period according to GFED4s.

2.1.3 Fuel measurements

During more recent fieldwork campaigns, we measured not only EFs but also other parameters including fuel characteristics and fire severity indicators. Before the fire, we collected fuel load and fuel composition from various classes (e.g. grass, litter, coarse woody debris, shrubs, and trees) and meteorological parameters. After the fire, we revisited the plots and recorded the combustion completeness of various fuel classes and fire intensity proxies (e.g. patchiness of the fire and scorch and char heights) following the methodology outlined by Eames et al. (2021) and Russell-Smith et al. (2020). Table 1 lists the individual UAS EF measurement campaigns and whether fuel was collected following the abovementioned methodology. Fires were lit on the windward side of the plot and generally burned through two to six individual randomly scattered 50×10 m fuel transects covering the fuel of a homogenous vegetation type and equal time since the last fire. We took the average of the affected fuel transects as the fire-averaged value to correspond to the fire-averaged EF, which is calculated over all the bag samples taken from that specific fire. Although the measurements were linearly correlated using the calibration bags for the individual fires, the standard deviations between the calibration samples were 2.58 % for CO₂, 7.06 % for CO, 2.32 % for CH₄, and 4.04 % for N₂O, indicating larger measurement uncertainties than reported by the manufacturers, which possibly arises from the bag methodology. The difference in the mean calibration value compared to the calibration gases was −4.75 % for CO₂, −1.32 % for CO, −3.97 % for CH₄, and −1.28 % for N₂O.

Table 1Measurement campaigns, including the number of fires for which emission factors were measured and the number of corresponding fuel transects.

Download Print Version | Download XLSX

2.2 Regression analysis

Field measurements provide the most accurate description of the vegetation conditions during the fire and yielded the most reliable insights into the drivers of EF dynamics. However, these measurements are sparse and thus unsuitable for spatiotemporal extrapolation. We therefore built machine learning algorithms, for which we selected a subset of satellite and reanalysis features with global coverage and temporal data availability for at least the past 20 years.

Table 2Satellite and reanalysis features assessed for the prediction of savanna biomass burning emission factors.

Download Print Version | Download XLSX

2.3 Spatial extrapolation for global savanna emission estimates

To assess the impact of EF dynamics on emission estimates and study global spatiotemporal patterns, we developed gridded EF layers that can easily be incorporated into existing emission inventories. The remote sensing proxies (“features”) were resampled to the required spatial resolution by simply averaging the values of the relevant grid cells. For example, to compute the 0.25^∘ fraction tree cover feature, we averaged the fraction tree cover of all 500 m pixels classified as savanna or grassland. When computing to a higher resolution, e.g. 500 m EFs, only the higher-resolution (MODIS-based) features exhibit pixel-to-pixel variability, while meteorological conditions (derived from ERA5-Land at 0.10^∘ resolution) remain consistent across many adjacent 500 m pixels. However, due to MODIS-derived features like FTC EF, estimates remain distinct between the grid cells. In contrast, temporal resolution within the models is more influenced by ERA5-Land-derived fluctuations. While FTC retrievals remain constant throughout the year, variations in factors like VPD, temperature, and FWI cause EF estimates to fluctuate on a daily basis.

Figure 2Estimation of the CO EF at 500 m resolution for MODIS tile “h20v10” on 1 June 2019 (g) using a random forest regression based on (a) fractional tree cover (FTC), (b) fraction of absorbed photosynthetically active radiation (FPAR), (c) the fire weather index (FWI), (d) vapour pressure deficit (VPD), and (e) soil moisture. For grid cells containing biomes other than savanna (f), GFED4s static EFs for the respective biome were imposed replacing the savanna EFs. Sources of the individual features are listed in Table 2.

Figure 2 provides an example for the estimation of the CO EF at 500 m resolution for MODIS tile “h20v10” (covering parts of Zambia, Botswana, Angola, Namibia, Zimbabwe, Mozambique, and the Democratic Republic of the Congo) on 1 June 2019, using the features shown in Fig. 2a–e. The temporal resolution of the computed gridded EFs in the example of Fig. 2 is daily, in which the day-to-day EF dynamics are being driven by daily variations in VPD, FPAR, FWI and soil moisture. Burned area products cannot differentiate the time of the day at which a grid cell was burned. For features with a typical diurnal pattern, we therefore weighed the hourly meteorological data by the average diurnal fire profile for the grid cell in the respective month of the year. This diurnal fire profile was based on the 3-hourly fractions of daily emissions obtained from GFED4.1s, which is based on the timing of active fire detections from both MODIS and geostationary satellites (Mu et al., 2011; van der Werf et al., 2017). To study the impact of EF dynamics in savannas, we calculated monthly global savanna emissions by multiplying the dynamic EFs computed by our models with dry matter consumption from GFED4s (Randerson et al., 2012; van der Werf et al., 2017) at 0.25^∘ spatial resolution for the 2002–2016 period (the period for which MCD64A1C5 as used in GFED4s was available). To classify the land cover type of the cell (Fig. 2f) we used the International Geosphere-Biosphere Program (IGBP) classification (Loveland and Belward, 1997), obtained from the MODIS annual MCD12Q1C6 product (Friedl and Sulla-Menashe, 2019), where the savanna biome comprised land cover type classes 6–11. We then calculated the dynamic monthly MCE and the EFs for CO, CH₄, N₂O, and CO₂ at 0.25^∘ spatial resolution for the savanna biome using the RF models. For burned grid cells that were partially classified as savanna, the EF of the cell was obtained by averaging the EFs of the different biomes in the underlying 500 m grid cells, weighted by their dry-matter consumption. We ran GFED4s using both static (original) and dynamic (this study) EFs for the savanna biome to determine the impact on seasonal and spatial emission patterns using our approach.

Figure 3EFs (g kg DM⁻¹) measured in the sampled vegetation types during the EDS and LDS and the EFs from savanna measurements listed in savanna literature based on the Andreae (2019) compilation. The green diamond represents the arithmetic mean, and the red cross represents the EMR weighted-average value. The colours correspond to the savanna subclasses at the bottom of the figure. Table 1 lists the time frames of the individual field campaigns, while Table A1 in the Appendix provides a broad floristic description of the dominant vegetation types.

Download

3.1 Variability of savanna EF measurements

During six fire seasons we have collected over 4500 bag samples containing emissions from 129 fires in a variety of savanna ecosystems under different seasonal conditions. Figure 3 shows the ranges, averages (green diamond), and WA EFs (red crosses) measured during the campaigns listed in Table 1. For the calculation of the WA N₂O EF we excluded samples that contained total carbon emissions of less than 10 mol following the findings described by Vernooij et al. (2021). Table A1 provides a short geomorphological and floristic description of the savanna ecosystems included in Fig. 3, including the seasonal behaviour of the dominant vegetation. The relatively small range in the boxplot describing previous savanna literature (Fig. 3, red box based on studies listed by Andreae, 2019) may be attributed to the fact that most studies report either fire averages, vegetation type averages, or even study averages, whereas the other boxplots based on our measurements show the variability observed between individual samples.

We observed substantial variability within EF bag samples from different savanna ecosystems, which was strongly linked to tree-cover density and mean annual rainfall. EFs of CO and CH₄ were lower (i.e. higher MCE) in xeric open savannas compared to woodland savannas. Fire-WA EF measurements for CO, CH₄, and N₂O using the UAS method were on average 13 %, 29 %, and 44 % lower, respectively, than estimates listed in previous inventories. However, this may be largely attributable to the fact that xeric savannas were overly represented in our measurements in terms of annual biomass consumption (i.e. sample bias). Our measurements in higher-rainfall savannas were much closer to the previous averages (Fig. 3). In humid areas like dambos (seasonally inundated grasslands) and riverine forests, we found large intra-seasonal differences in N₂O, CO, and CH₄ EFs. Water availability in these landscape features is often strongly soil type and geomorphology related (Bullock, 1992; Gonçalves et al., 2022), making the correlation with seasonal rainfall less direct and drying patterns over the dry season more diverse. The grasslands with the highest EFs (found in high-rainfall savanna dambos) were uncharacteristically green for the time of the season, and under those conditions fires in these landscapes would therefore not be representative of more xeric grasslands.

Table 3Consumption of fuels prone to residual smouldering combustion in the EDS and LDS for xeric open savannas measured in Botswana and Australia and Miombo woodlands measured in Mozambique and Zambia.

¹ Weighted average over the consumed contribution of each individual fuel subclass. ² Weighted average over the dominant shrub types found in the plots.

Download Print Version | Download XLSX

3.2 EF seasonality, fire intensity dynamics, and fuel consumption in xeric and mesic savannas

Table 3 lists the EDS and LDS pre- and post-fire fuel characteristics averaged over all the transects we measured in the respective vegetation type and season. In both xeric and mesic savannas, the moisture content of the fuel and the relative humidity were substantially lower in the LDS compared to the EDS. This resulted in increases in fire intensity proxies over the dry season. Particularly during measurement campaigns in the Miombo woodlands in Mozambique and Zambia, the fine fuel in the EDS plots predominantly consisted of tree litter and became even more litter-dominated with the progression of the dry season. EDS fires were patchy and generally did not consume coarse woody debris and shrubs. As the dry season progressed, there was a clear shift towards the combustion of more live foliage and fuels prone to residual smouldering combustion (RSC) like coarse woody debris, stems, and densely packed litter, which after months of drying had become more receptive to combustion. RSC occurs after the passage of a flame front, and its emissions are not lofted by strong fire-induced convection (Bertschi et al., 2003). The increase in the consumption of live and coarse fuels towards the end of the dry season coincided with higher EFs for CO and CH₄ in the LDS. This shift in combusted fuels also results in a seasonal increase in the WA carbon content of the consumed fuel of woody savannas (Table 3), which linearly scales the EFs of all measured species. For some characteristics (e.g. the total fuel load), it is important to note that the average time since the last fire was not necessarily equal between the listed vegetation types. The higher fuel loads we found in open savannas in Australia compared to Botswana may be partially attributed to the longer fuel build-up.

Figure 4(a) Correlation of the predicted and measured fire-integrated weighted-average MCE for the training (orange) and validation (blue) datasets. The vertical blue and orange lines represent the standard error of the mean within the respective fire. The vertical red line is the static MCE derived from the EFs used in GFED4s. The “improvement” refers to the reduced mean absolute error compared to prediction using this static GFED4 (red line) MCE and compared to the average of the input data (magenta line). (b) The remote sensing and reanalysis datasets used by the model and the feature importance (an indication of how strong each feature is used to differentiate the data) of the respective features.

Download

Overall, our measurements of CO and CH₄ EFs in xeric, grass-dominated, and shrub-dominated savannas (e.g. Australian Spinifex grasslands and open savannas in the Kalahari) were slightly lower in the LDS compared to the EDS campaigns but much lower compared to woody savannas (Fig. 3). Contrary to the mesic savannas, where RSC-prone fuel is readily available and becomes more flammable with the progression of the fire season, fires in xeric shrub and grasslands tended to consume much of the available fuel in the EDS (Table 3). Overall, the WA nitrogen content of the combusted fuel decreased with the progression of the dry season through curing of grasses and litter decomposition. This was somewhat compensated for by an influx of leaf litter and an increased combustion of live shrubs, which were richer in nitrogen than grasses (that had commodiously already cured in the EDS). Overall, fires that consumed more litter emitted more N₂O than grass-dominated fires. Between individual fires, the curing stage of the grasses affected the N₂O EF, with green seasonally inundated grasslands emitting more N₂O compared to fully cured grasslands. In some miombo woodland fires in Kafue, which were measured in November when the vegetation already carried its first green flush, we also measured relatively high N₂O EFs.

3.3 Estimation of BB EFs using random forest regression based on satellite proxies

To extrapolate these relations for use in global emission inventories, we correlated the field measurements to satellite products. Table 4 lists the correlations of the individual field-measured ecosystem attributes to the MCE and fire-averaged EF measurements and global satellite proxies. Direct correlations between fire-averaged EF measurements and global satellite proxies and inter-correlations between the satellite and reanalysis proxies are listed in Table A2. The strongest predictors for the MCE and the CO and CH₄ EF were the tree density in the plots, the grass-to-litter ratio, the combustion completeness, and the WA moisture content of the consumed fuel (Table 4). In turn, these parameters were best correlated to the remotely sensed FTC, FBC, VPD, and FWI. EFs for CO and CH₄ are primarily proportionate to the inverse combustion efficiency (i.e. the not fully oxidized compounds) which had a standard deviation of 90 % relative to the mean. CO₂, on the other hand, is proportionate to the fully combusted carbon fraction, which is much larger and more stable, with a relative standard deviation of 4.5 % compared to its mean. Therefore, the carbon content of the fuel – with a standard deviation of roughly 5 % – becomes a dominant factor explaining the variability in CO₂ EFs. The features that most strongly correlated with the N₂O EF were the nitrogen-to-carbon ratio in the combusted fuel and the percentage of grass in the fine fuel (consisting of grass, litter, and coarse woody debris), which in turn correlated with the FBC and the VPD.

Table 4Spearman correlation matrix for the field-measured ecosystem attributes, the fire-averaged emission factors and MCE, and the satellite products used in the study. Positive correlations are presented in blue, while negative correlations are presented in red.

Download Print Version

Figure 5Pearson correlation of the predicted and measured fire-integrated WA MCE (a), CH₄ EF (b), N₂O EF (c), and CO EF (d) for the training (orange) and validation (blue) datasets using a limited set of features. The boxes in the bottom right of the panels list the remote sensing and reanalysis datasets used by the model and the feature importance (an indication of how strong each feature is used to differentiate the data). The red line represents the static biome average used in GFED4s, while the magenta line represents the average of the training and validation data. The abbreviation “improv. GFED” refers to the reduced mean absolute error compared to the static average used by GFED4s, and the abbreviation “improv. Avg.” refers to the reduced mean absolute error compared to the static average of the input data.

Download

For the global estimation of MCE and EFs, we found that RF models performed well with respective out-of-sample correlation coefficients ranging between 0.80 and 0.99. In Figs. 4 and 5, feature importance represents the mean accumulation of the impurity decrease within each tree and is an indication of how much the variability in each feature is used as a split criteria by the models to explain the variability in the EF data. The average MCE in the measurements was slightly higher compared to earlier assessments that were used by GFED4s (red line in Fig. 4), which may again be attributable to the dominance of relatively dry savannas. Overall, we found that by using only globally available features covering a large (>20 year) time span, we could estimate the field-measured MCE of the fires in the validation set with a mean absolute error (MAE) of 0.006. Using the static MCE in GFED4 (MAE of 0.015 compared to the measurements) as a baseline, this meant a MAE reduction of 60 %.

Although the features listed in Fig. 4 all have sufficient spatiotemporal coverage for global emission modelling, some features exhibited strong co-variation. Other retrievals were hampered by LDS cloud cover (e.g. dNBR and Pgreen), which meant we could not use consistent quality retrievals or had to remove samples from the data. Further simplification using a subset of features that are not directly correlated reduced the data dependency and computational intensity of the model and the loss of training data due to cloud cover without losing much explained variance. When using a five-feature subset, we found that RF regressors still predicted much of the variability in the MCE and EFs. Figure 5a shows the predictive performance of a RF regression model that uses VPD, FTC, FWI, FPAR, and soil moisture (SM) to estimate the MCE, which was relatively similar to the model predicting MCE using all features (r of 0.80 versus 0.86).

Figure 6Difference in savanna and grassland fire emissions for CO (a), CH₄ (b), and N₂O (c) between emission computation using dynamic EFs versus static biome reference EFs (dynamic minus static) calculated using GFED4s for the 2002–2016 period.

We found that spatial variability dominated the total variability in the MCE within the savanna biome with higher combustion efficiency in more xeric and open savannas. To isolate the effect of combustion efficiency in the prediction of individual species and make the model more transparent, we added the computed MCE to the predictor features. Both the models that were trained using the full set of features in Table 1 and the five-feature models identified the computed MCE as one of the primary features explaining of the variability in other EFs. The largest deviation from static EFs (vertical red line in Fig. 5d) was predicted for N₂O. This is partially due to the large number of new fires, which on average (vertical magenta line) were lower than the static reference used in GFED4s. The modelled MCE was the main predictor of the N₂O EF, followed by the soil moisture in the top layer (0–7 cm depth). Somewhat surprisingly, we found soil moisture to correlate more strongly with the tree density in the plot rather than the fuel moisture content (Table 4).

Figure 7Seasonality of fire carbon emissions (black) and the computed CO EF (orange) for different savanna subclasses in Southern Hemisphere Africa, averaged over the 2002–2016 period. The savanna classes are based on the International Geosphere-Biosphere Program (IGBP) classification (Loveland and Belward, 1997). The shaded areas represent the timing of our measurements in Southern Hemisphere African savannas, indicating that our LDS campaigns in particular may not be representative of the bulk of the fires. The horizontal red bar on the right represents the static EF used for savannas by GFED4s.

3.4 Impact on global emission estimates using variable savanna emission factors

Figure 6 shows the relative impact of using variable EFs on annual global savanna fire emissions of CO (Fig. 6a), CH₄ (Fig. 6b), and N₂O (Fig. 6c), averaged over the 2002–2016 period based on GFED4s. The map only shows cells for which the partial coverage of savannas exceeds 50 %. In grid cells that are partially (50 %–99 %) covered by savanna, the total impact on emissions is to some degree diluted as the EFs of the non-savanna biomes remained constant. For CO and CH₄, the dominant effect is a spatial redistribution with higher CO and CH₄ EFs in mesic, high-tree-cover savannas and lower EFs in xeric savannas compared to previous estimates. For CO₂ (not shown), we find the opposite pattern to CO. Relatively speaking, however, changes in CO₂ emission are much smaller because most carbon is emitted as CO₂, even when MCE values are low. Although CO and CH₄ followed the same spatial pattern, we found that MCE affected the CH₄ EF more strongly than the CO EF, which resulted in lower CH₄ to MCE ratios in savannas with lower tree density. Global savanna emissions of CO were 2 % higher compared to the GFED4s reference scenario, whereas CO₂, N₂O, and CH₄ emissions were, respectively, 0.2 %, 18 %, and 5 % lower. N₂O emissions were lower for the entire savanna biome (Fig. 6c).

Figure 7 shows the seasonal patterns in the average CO EF for different savanna vegetation classes in Southern Hemisphere Africa. The IGBP savanna subclasses are only used here to indicate the average patterns and are not involved in the EF calculation. Using the IGBP classification, our samples were classified as “Woody savannas” (24 %), “Savannas” (42 %), “Open shrubland” (21 %), “Grassland” (4 %), “Cropland/Natural vegetation mosaic” (6 %), and “Croplands” (1 %). The latter two classes are misclassifications and were all situated in protected areas with no crops. These classes are listed in the accompanied dataset (Vernooij, 2023). We found a stronger and more persistent seasonal decline in the CO EF in xeric grasslands and shrublands compared to woody savannas. N₂O EFs showed a similar pattern characterized by a decline over the dry season in the more xeric grass and shrubland savannas, while EFs in woody savannas are more stable. The model indicates a reversal of the seasonal trend in woody savannas around August–September, long before these rains start. The coloured areas represent the timing of our field campaigns in this region. Although LDS campaigns were conducted before the first seasonal rains, the graph indicates they may not be indicative of peak-season fires. Figure 8 shows an overview of the relative changes in emissions for the various savanna-rich GFED regions. Many of these regions contain both xeric and mesic savannas with contrasting spatial patterns, meaning local differences may be much larger (Fig. 6).

Table 5Emission factor averages for the global savanna.

¹ Averaged over the fires measured using the drone methodology (skewed towards xeric savannas) ² Averaged over the fires measured using the drone methodology and the included literature studies. ³ Dynamic EFs weighted by the consumed biomass at time and location of fires as calculated using GFED4s.

Download Print Version | Download XLSX

Both our measurements and the savanna biome averages in literature compilations (e.g. Akagi et al., 2011; Andreae, 2019) are subject to sampling bias when representing global savannas. A disproportionate number of field studies are clustered around reactively accessible locations with a well-developed research infrastructure, whereas other fire-prone areas lack direct field measurements. Rather than comparing the average of our savanna measurements to the literature averages, we computed the dynamic EFs globally using the RF model and subsequently calculated the emissions for the entire savanna biome. We then divided these annual emissions by the consumed biomass from GFED4s to get the annual consumed-biomass weighted-average EFs, which we will further refer to as the “effective” EFs. Over the 2002–2016 period, the effective EFs over the savanna biome were 1685 (±5) for CO₂, 64.3 (±0.6) for CO, 1.9 (±0.0) for CH₄, and 0.16 (±0.00) for N₂O, with the number in the parentheses indicating the interannual standard deviation. In Table 5, we compare the effective average EFs over the 2002–2016 period calculated by our model to the static average EFs for savanna and grassland vegetation used by GFED4s and those suggested by Andreae (2019) and Wiedinmyer et al. (2023). Table 5 also lists the average EFs of the UAS-measured fires and the average EFs of all included fires (including literature studies). Except for N₂O, the differences between the effective EFs compared to more recently updated static EFs from Andreae (2019) were larger (+1.3 % for CO₂, −7.1 % CO, −31.4 % CH₄, and −3.7 % N₂O) than when using GFED4s EFs as the static reference.

Figure 8Relative difference in the landscape fire emissions of CO₂, CO, CH₄, and N₂O for the 2002–2016 period when using dynamic EFs versus static EFs using GFED4s (dynamic minus static) over the different savanna-rich GFED regions. Note that many of these regions encompass both xeric and mesic savannas with contrasting patterns that balance each other out. Differences may therefore be much larger on a regional scale.

Download

4.1 Comparison with previous studies

The largest difference compared to previous savanna burning emission estimates is the reduction in N₂O emissions. Rather than being the effect of including spatiotemporal dynamics, this reduction resulted from a substantial influx of new N₂O EF measurements that exhibited significantly lower values than the averages found in EF compilations. Our field measurements yielded an average EF of 0.11 g kg⁻¹, while EF compilations reported averages of 0.21 g kg⁻¹ (Andreae and Merlet, 2001), 0.20 g kg⁻¹ (Akagi et al., 2011), and 0.17 g kg⁻¹ (Andreae, 2019). However, in our measurements, xeric savannas are over-represented. When using the global RF model to extrapolate the measurements over the entire savanna biome, the effective average N₂O EF – for savanna grid cells at the time of their burning – was 0.16 g kg⁻¹, which is similar to the value listed in Andreae (2019). It is known that older studies might overestimate N₂O, due to N₂O formation in stainless steel sample containers (Muzio and Kramlich, 1988). Particularly compared to more recent studies, our EFs were in line with other savanna measurements from South America (0.05–0.07 g kg⁻¹; Hao et al., 1991; Susott et al., 1996), Australia (0.07–0.12 g kg⁻¹; Hurst et al., 1994; Meyer et al., 2012; Surawski et al., 2015), and Africa (0.16 g kg⁻¹; Cofer et al., 1996). In accordance with Winter et al. (1999b), we found N₂O EFs to be closely correlated with the nitrogen content of the fuel. Through this relation, we can explain both the spatial distribution observed in Fig. 6c and the different seasonal trends. In line with Susott et al. (1996) and Ward et al. (1992), we found that woody vegetation has higher nitrogen content contained in the foliage (Table 3), causing higher N₂O emissions from tree-dominated areas. We found relatively low nitrogen content for Australian open woodland savannas, which was in line with previous studies (Bustamante et al., 2006). The seasonal reduction in the nitrogen content of the fuel as the vegetation cures (Table 3) coincides with a reduction in the N₂O EF over the dry season (Yokelson et al., 2011; Vernooij et al., 2021). This tends to happen quicker in xeric grasslands and shrublands compared to more mesic and tree-covered areas. On the other hand, as fuels get more receptive over the dry season, fires consume increasingly greater amounts of litter, coarse fuels, and live foliage, provided these fuels are available (Table 3). This increases the WA carbon and nitrogen contents of the fuel.

For carbonaceous species our model predicts a spatial redistribution, characterized by higher combustion efficiency in lower-tree-cover savannas and lower combustion efficiencies in more woody savannas. Previous research by van Leeuwen and van der Werf (2011) identified multi-linear correlations between EFs of CO₂, CO, and CH₄ and environmental drivers resulting in coefficients of determination (r²) ranging from 0.48 to 0.62. In accordance with their study, as well as many other field studies (e.g. Laris et al., 2021; Sinha et al., 2004), we found the FTC to be a strong predictor of the MCE and the EFs of CO and CH₄ (Fig. 9). When denoted in grams per kilogram of dry biomass consumed, EFs of carbonaceous species are dependent on both the combustion efficiency and the carbon content of the fuel. The carbon content is often fixed in global studies of EFs, e.g. 45 % in Andreae (2019) and Andreae and Merlet (2001) or 50 % in Akagi et al. (2011), with the latter forming the basis of the EFs used in GFED4s that represent the static EF references in this study. However, both the combustion efficiency and the carbon content have a spatial component with higher carbon contents in shrubs and trees compared to grasses (Table 3). For the studied fires, the WA carbon content of the fuel ranged from 40.3 % to 49.3 %, which linearly scales to a 22 % difference in EFs between those extremes. In line with Andreae (2019), we assigned a carbon content of 45 % to literature studies for which the carbon content was not reported that was close to our average measured value of 45.8±2.3 %. Contrary to previous research, which indicated that dryer conditions in the LDS would lead to higher-MCE fires in both grasslands and savanna woodlands (Korontzi, 2005), we found lower MCE in these regions under late-LDS conditions (Fig. 3). One potential explanation is that although the LDS fires were more intense, they consumed much more RSC-prone fuels (Table 3), which may explain the higher CH₄ and CO EFs. An alternative explanation to this fuel-driven MCE reduction is that in certain areas our measurement campaigns missed the peak season when fires are driven by stronger winds (Laris et al., 2021; N'Dri et al., 2018) and that fire intensity and MCE in these areas would already be on the decline. Eck et al. (2013) studied seasonal changes in BB particles during 15 annual fire seasons in xeric (e.g. Etosha Pan and Kruger National Park) and mesic (e.g. Mongu) savannas in southern Africa using the Aerosol Robotic Network (AERONET). They found a linear trend in the single scattering albedo (SSA), increasing throughout the dry season, which would support a late dry-season decrease in MCE (Liu et al., 2014; Pokhrel et al., 2016). We found that in the xeric savannas the composition of the fuel in LDS fires did not significantly differ from EDS fires, as most of the available fuel was consumed in both the EDS and LDS fires. In these areas, we did observe a slight seasonal decline in CO and CH₄ EFs.

Figure 9The non-linear regression between the CH₄ EF and the MCE for the individual bag samples (green circles) and the fire-averaged values (orange diamonds). In the box on the bottom left, ρ refers to Spearman's rank correlation coefficient for the bag samples.

Download

In accordance with previous studies (e.g. Korontzi et al., 2003b; van Leeuwen and van der Werf, 2011; Barker et al., 2020), we found steeper CH₄ EF-to-MCE regression slopes in woodlands compared to grasslands. Our data indicated a positive correlation of the CH₄ EF-to-MCE slope with the FTC based on MOD44Bv006. The MCE is a simplified form of the combustion efficiency and only calculated using CO and CO₂ emissions. Being less oxidized than CO (which is still common in flaming combustion), CH₄ emissions have a stronger dependency on the actual combustion efficiency (CO₂ divided by all carbon emissions). While most studies describe the relationship between the CH₄ EF and the MCE as being linear (Korontzi et al., 2003; van Leeuwen and van der Werf, 2011; Selimovic et al., 2018; Yokelson et al., 2003), we found that for individual bag samples it was better described using a non-linear function (Fig. 9), in line with findings by Meyer et al. (2012) for Australian savanna measurements. Figure 9 represents individual bag measurements rather than fire averages (for which the spread in MCE is much lower). Laboratory experiments described by Selimovic et al. (2018) and others showed that the CH₄-to-CO ratio is more complex and variable in real time than at the fire-average level. Individual bag samples sampled over a concise 35 s time frame thus exhibit a broader range and more pronounced variation in comparison to fire averages. Stable carbon isotopes also point to CH₄ emissions being more depleted in heavy carbon (¹³C) compared to CO in both mixed (C₃ and C₄) and single-fuel-type experiments using wooden logs, indicating a stronger dominance of RSC and the pyrolysis of lignin in its total emissions (Vernooij et al., 2022a). Mainly within woody savannas, this clarifies why studies focused on either smouldering- or flaming-phase emissions exhibit diverse slopes for CH₄ EF-to-MCE when employing linear regressions. Additionally, this phenomenon accounts for the inclination of the slope to intensify in fuel types characterized by higher lignin content.

Although higher MAR generally coincides with high FTC, this was not the case for our measurements from Brazil. The measured areas in the Estação Ecológica Serra Geral do Tocantins (EESGT) received relatively high MAR (1250–1600  mm yr⁻¹) compared to 850–1250  mm yr⁻¹ for Miombo woodlands and 890–1100  mm yr⁻¹ for Mozambican Miombo woodlands. Nonetheless, despite being strictly protected from logging and other land clearing practices, the MOD44BV006 FTC in the measured areas in EESGT was very low (1 %–10 %, with an average of 2 %) compared to 7 %–32 %, with an average of 19 %, for Zambian Miombo woodlands and 3 %–43 %, with an average of 22 %, for Mozambican Miombo woodlands. Our measurements in the EESGT being skewed towards open savannas (that typically burn with higher MCE) may explain the relatively low CH₄ EF-to-MCE slope discussed in Vernooij et al. (2021). For the whole Cerrado, the average MOD44BV006 FTC is 17 %, indicating that the measurements in EESGT may be underestimating the MCE in other parts of the Cerrado. According to its classification, MCD44BV006 FTC only includes canopies of trees exceeding 5 m in height (Adzhar et al., 2022), which may be why some common Cerrado species are classified as shrubs. However, the EFs observed from these areas were similar to those observed in low-tree-cover savannas.

Figure 10Detection rate of the fires measured using the UAS methodology by different satellite algorithms in the EDS (green) and LDS (orange). The darker area represents the cases where a fire was observed in the actual pixel within the listed time frame. The lighter areas represent fires that were not detected in the same pixel as the samples but were detected in adjacent pixels. Time frames are listed below the product labels. For the VIIRS detections, the distance limits between the detection point and closest sample of the fire were 1 km for the darker-shaded area and 3.5 km for the lighter-shaded area.

Download

Measurements of fuel loads were higher than previous measurements from African savannas described by Shea et al. (1996). They found average fine-fuel loads (litter and grass) of 3.8 t ha⁻¹ in moist Miombo woodland. In semiarid Miombo woodland they found 3.1 t ha⁻¹. In comparison we found 5.6 t ha⁻¹ in Mozambican Miombo woodland and 5.6 t ha⁻¹ in Zambian Miombo woodland. The percentage of grasses in these fuels was similar; Shea et al. (1996) reported 24 % in moist Miombo woodland and 18 % in semi-arid Miombo woodland, whereas we found 37 % in Mozambican Miombo woodlands and 18 % in Zambian Miombo woodlands. The combustion completeness of these fuels was slightly lower in our fires at 50 %–80 % versus 80 %–92 % reported by Shea et al. (1996), although the lower values in this range occurred in the EDS. Combustion completeness of shrub leaves and coarse woody debris was in the same range. For dambo grasslands our fuel loads were also much higher at 6.2 (±2.16) t ha⁻¹ of which 99 % was grass versus 3.1 t ha⁻¹ from Shea et al. (1996). Although these differences are large, they may be attributed to the significant natural variability in productivity and decay related to water availability, fire frequency, and termite and grazing activities in these natural landscapes.

4.2 Model representativeness

This is the first study to quantify the spatial distribution of GHG EFs over the entire savanna biome using field measurements from a variety of savanna ecosystems and their relation to global data mainly from satellites. Although spatiotemporal coverage has improved, there are still many understudied savanna and grassland areas for which we have derived EFs based on our model. Figure 1 clearly illustrates the gaps in the spatial distribution of the training data. Savannas bordering the tropical rainforest, Northern Hemisphere Africa, Central America, South-East Asia, and temperate grassland ecosystems are particularly understudied. Due to the lack of measurements in these ecosystems, EFs are presently computed based on measurements primarily taken in Southern Hemisphere Africa. Nevertheless, EF trends in other regions might considerably differ from those observed in extensively studied savannas. To guarantee the model's relevance to specific regions, it remains essential to calibrate and evaluate the model using supplementary in situ emission factor measurements.

Most of the fires used to train the models were prescribed fires set by scientists or park rangers in protected areas in order to facilitate collection of data pre- and post-burn on site. It is common practice to extrapolate these measurements in relatively undisturbed savanna vegetation to the wider savanna. Even though these protected natural areas tend to burn more frequently, they represent a minority of the area that is currently modelled using savanna and grassland emission factors by global inventories (e.g. Fig. 1). Most of this area is to some degree affected by humans though cattle ranging, wood harvesting, slash and burn agriculture, etc. This means fires in this study may not always represent the burning practices by local farmers, and representativeness of our work for the larger savanna area therefore remains uncertain. The samples were predominantly collected over heading fires, which in the measured fires typically represented most of the burned area. A common approach for prescribed fires is burning against the wind (backing fire) to minimize both the impact on vegetation and risk of spread. In a heading fire, RSC can be increased because the high rate of spread and patchiness leaves fuels smouldering further from the convection associated with the advancing flame front. In accordance with Wooster et al. (2011) and Laris et al. (2021), we found higher MCE in samples from backing fires, indicating less RSC and thus CH₄ and CO emissions in these types of fires. Another possible explanation for the higher MCE in the backing fire samples is that more slowly lofting RSC smoke does not mix with the flaming combustion emissions in these measurements like it does in heading fires.

4.3 Spatial resolution and model considerations

For this research, we computed the average attributes within the 0.25^∘ grid cell before calculating the savanna EF. This spatial resolution was selected because the GFED4s burned area data, including assumptions for small fires, is generated at a 0.25^∘ resolution. Nonetheless, there are potential advantages to future EF estimations at greater spatial resolutions. Recent studies indicate that higher-resolution modelling yields different emissions than that based on aggregated data, due in part to improved representation of landscape heterogeneity (van Wees and van der Werf, 2019). Enhancing the resolution of meteorological data would further amplify the precision of these models. These advancements anticipate that future global emission inventories will adopt higher spatial resolutions, enabling better representation of local or regional dynamics. We found the highest variability in EFs within smaller landscape features that are bound to geomorphological niches, typically along rivers and valleys. While these features are likely to have low significance for global emission patterns, they represent vital ecosystems that may require special fire protection. In its current form, the model may not always pick up on those landscape features. High-resolution modelling allows for a better understanding of localized fire regimes, especially in areas with relatively heterogeneous land cover.

The model is limited by the accuracy and spatial resolution of the underlying products. Using the features included in the current models, EFs can be calculated up to the native spatial resolution of the included MODIS-based products (500 × 500 m), which is also the resolution of globally available burned area products. New high-resolution burned area products, however, indicate that these global products, including the GFED4s data used for global emission analyses in this study, grossly underestimate burned area due to omission of small fires (Chen et al., 2023; Roteta et al., 2021; Roy et al., 2019). This also pertains to a substantial proportion of the fires we measured. Of the UAS-measured fires in this study, only 5 of the 45 EDS fires (11 %) and 13 of the 65 LDS fires (20 %) were registered by MCD64A1 as burned area (including adjacent pixels and a 4 d time lag). Only 4 (9 %) of the 45 EDS fires and just 32 (49 %) of the 65 LDS fires were detected by VIIRS S-NPP as thermal anomalies, with the hotspot's centre point (accounting for a 1 d time lag) falling within a 3.5 km radius of the sample. Depending on the spatiotemporal nature of these omissions, this may affect some of the results in this study concerning the effects of the EF dynamics on total emissions. Chen et al. (2023) indicate that disproportionately more burned area is added in higher-tree-cover areas when using higher-resolution satellite imagery in the savannas. Giving more weight to these areas would mean our savanna-wide effective EFs of CO, CH₄, and N₂O would increase. The Sentinel-2-based burned area product from Roteta et al. (2021) performed much better and registered 8 of our 14 EDS fires (57 %) and all of our 16 LDS fires (100 %) in Botswana and Mozambique in 2019 (including adjacent pixels and up to a 21 d time lag). Due to there being fewer overpasses present, the temporal allocation of this product is less precise, with an average time lag of 5.5 d. Figure 10 shows the portion of our EDS and LDS fires that were detected by various satellite algorithms.

Fire intensity proxies (dNDVI and dNBR from MODIS) were considered by the models to be poor predictors for the EFs. A potential explanation is that these features were not always representative, as many of the fires only affected part of the pixel. Similar misrepresentation errors can be expected for the NDVI before the fire, FPAR, and Pgreen. Particularly in the LDS, we were often limited to areas that were enclosed by recent fire scars (0–2 years old) or other non-flammable boundaries like roads or bare areas. Although the burnt areas were sizable (several hectares), many of the retrievals in these pixels may poorly represent the burned vegetation. Along with inconsistent retrievals related to cloud cover, this may contribute to these features being deemed poor predictors by the models. Enhanced-resolution features could improve the accuracy of pixel representations for the actual burned vegetation.

The meteorological parameters obtained from the ERA5-Land dataset carry uncertainty. This uncertainty increases when examining earlier time periods or remote regions due to diminished validation data availability. To what extent uncertainty propagates to the EF predictions varies depends mostly on whether there is a bias that was also present in the training data or misinterpretation or uncertainty in general. As this model is trained using specific datasets, these datasets should not be replaced by other sources without evaluating the consistency of that source with the feature training data. FTC and FBC based on MOD44Bv006 were found to be strong predictors of BB EFs. However, intercomparison with Tropical Biomes in Transition (TROBIT) field sites in African, Brazilian, and Australian savannas has shown that this product consistently underestimates canopy cover in tropical savannas by between 9 % and 15 % (Adzhar et al., 2022). Products based on higher-resolution satellite retrievals (e.g. LandSat and Sentinel) have the potential to further enhance the spatial resolution of the EF estimates to include small landscape features and thus become more representative. Although all satellite data come with some uncertainty, we feel the errors are small enough to have high confidence in the key findings, such as lower EFs in dry regions and higher EFs in wetter regions.

The interdependence among features led to varying feature importance scores (depicted in Fig. 4) across different model runs, driven by the test–train data division and bootstrap resampling. For instance, a decision tree split based on VPD might closely resemble soil moisture or RH, and FTC in national parks often exhibits strong correlation with the MAR, with the exception of our measurement sites in Brazil. While we conducted model runs considering different subsets of features and selected the optimal one, it is important to note that various features might also effectively account for a significant portion of the variance. In cases where features had substantial co-variation (such as FPAR and LAI or FWI and ISI), this resulted in the selection of only one feature for the simplified model, even if both features demonstrated high initial scores.

The models are currently trained using meteorological features obtained from ERA5-Land (Muñoz-Sabater et al., 2021), which is available from 1950 to present and has a 2- to 3-month delay. When interested in longer time periods or for near-real-time (NRT) applications, these features may be substituted with ERA5 (Hersbach et al., 2020), which is available from 1940 to present with a shorter latency period of 5 d, or even CMIP climate projections. Although supplementing the datasets on which the models are trained with alternative data always comes with additional uncertainty, we found meteorological parameters obtained from ERA5-Land to be in close accordance with ERA5, indicating the two may also be substituted. This means that the EFs computed using the methodology outlined in this paper could potentially also be used to improve NRT biomass burning emission estimates like those from CAMS-GFAS (Andela et al., 2015; Di Giuseppe et al., 2016).