Drivers of spatial and temporal variability in savanna fire emission factors

Roughly half of global fire emissions originate from savannas, and emission factors (EF) are used to quantify the amount of trace gases and aerosols emitted per unit dry matter burned. It is well known that these EFs vary substantially even within a single biome but so far quantifying their dynamics has been hampered by a lack of EF measurements. Therefore, global emission inventories currently use a static averaged EF for the entire savanna biome. To increase the spatiotemporal coverage of EF measurements, we collected over 4500 EF bag measurements of CO₂, CO, CH₄ and N₂O using an unmanned aerial system (UAS) and measured fuel parameters and fire severity proxies during 129 individual landscape fires. These measurements spanned various widespread savanna ecosystems in Africa, South America and Australia, with early and late dry season campaigns. We trained random forest (RF) regressors to estimate daily dynamic EFs for CO₂, CO, CH₄ and N₂O at 500×500-meter resolution based on satellite and reanalysis data. The RF models reduced the difference between measured and modelled EFs by 60-85% compared to static biome averages. The introduction of EF dynamics resulted in a spatial redistribution of CO, CH₄ and N₂O emissions compared to the Global Fire Emissions Database version 4 (GFED4s) with higher emissions in higher rainfall savanna regions. While the impact from using dynamic EFs on the global annual emission estimates from savannas was relatively modest (+2% CO, -5% CH₄ and -18% N₂O), the impact on local EFs may exceed 60% under dry seasonal conditions.