Landscape fires show large variability in the amount of biomass or fuel consumed per unit area burned. Fuel consumption (FC) depends on the biomass available to burn and the fraction of the biomass that is actually combusted, and can be combined with estimates of area burned to assess emissions. While burned area can be detected from space and estimates are becoming more reliable due to improved algorithms and sensors, FC is usually modeled or taken selectively from the literature. We compiled the peer-reviewed literature on FC for various biomes and fuel categories to understand FC and its variability better, and to provide a database that can be used to constrain biogeochemical models with fire modules. We compiled in total 77 studies covering 11 biomes including savanna (15 studies, average FC of 4.6 t DM (dry matter) hag-1 with a standard deviation of 2.2), tropical forest (n Combining double low line 19, FC Combining double low line 126 ± 77), temperate forest (n Combining double low line 12, FC Combining double low line 58 ± 72), boreal forest (n Combining double low line 16, FC Combining double low line 35 ± 24), pasture (n Combining double low line 4, FC Combining double low line 28 ± 9.3), shifting cultivation (n Combining double low line 2, FC Combining double low line 23, with a range of 4.0-43), crop residue (n Combining double low line 4, FC Combining double low line 6.5 ± 9.0), chaparral (n Combining double low line 3, FC Combining double low line 27 ± 19), tropical peatland (n Combining double low line 4, FC Combining double low line 314 ± 196), boreal peatland (n Combining double low line 2, FC Combining double low line 42 [42-43]), and tundra (n Combining double low line 1, FC Combining double low line 40). Within biomes the regional variability in the number of measurements was sometimes large, with e.g. only three measurement locations in boreal Russia and 35 sites in North America. Substantial regional differences in FC were found within the defined biomes: for example, FC of temperate pine forests in the USA was 37% lower than Australian forests dominated by eucalypt trees. Besides showing the differences between biomes, FC estimates were also grouped into different fuel classes. Our results highlight the large variability in FC, not only between biomes but also within biomes and fuel classes. This implies that substantial uncertainties are associated with using biome-averaged values to represent FC for whole biomes. Comparing the compiled FC values with co-located Global Fire Emissions Database version 3 (GFED3) FC indicates that modeling studies that aim to represent variability in FC also within biomes, still require improvements as they have difficulty in representing the dynamics governing FC.
