Modelling biomass burning emissions and the effect of spatial resolution: a case study for Africa based on the Global Fire Emissions Database (GFED)

Large-scale fire emission estimates may be influenced by the spatial resolution of the model and input datasets used. Especially in areas with relatively heterogeneous land cover, a coarse model resolution might lead to substantial errors in estimates. We developed a model using MODerate resolution Imaging Spectroradiometer (MODIS) satellite observations of burned area and vegetation characteristics to study the impact of spatial resolution on modelled fire emission estimates. We estimated fire emissions for sub-Saharan Africa at 500 m spatial resolution (native MODIS burned area) for the 2002–2017 period, using a simplified version of the Global Fire Emissions Database (GFED) modelling framework, and compared this to model runs at a range of coarser resolutions (0.050, 0.125, 0.250∘). We estimated fire emissions of 0.68 Pg C yr−1 at 500 m resolution and 0.82 Pg C yr−1 at 0.25∘ resolution; a difference of 24 %. At 0.25∘ resolution, our model results were relatively similar to GFED4, which also runs at 0.25∘ resolution, whereas our 500 m estimates were substantially lower. We found that lower emissions at finer resolutions are mainly the result of reduced representation errors when comparing modelled estimates of fuel load and consumption to field measurements, as part of the model calibration. Additional errors stem from the model simulation at coarse resolution and lead to an additional 0.02 Pg C yr−1 difference in estimates. These errors exist due to the aggregation of quantitative and qualitative model input data; the average- or majority- aggregated values are propagated in the coarse-resolution simulation and affect the model parameterization and the final result. We identified at least three error mechanisms responsible for the differences in estimates between 500 m and 0.25∘ resolution simulations, besides those stemming from representation errors in the calibration process, namely (1) biome misclassification leading to errors in parameterization, (2) errors due to the averaging of input data and the associated reduction in variability, and (3) a temporal mechanism related to the aggregation of burned area in particular. Even though these mechanisms largely neutralized each other and only modestly affect estimates at a continental scale, they lead to substantial error at regional scales with deviations of up to a factor 4 and may affect large-scale estimates differently for other continents. These findings could prove valuable in improving coarse-re