Vegetation biomass, woody debris and litter influence the behaviour of wildfires such as flaming or smouldering combustion and hence affect fire emissions. However, the effects of fuels on the combustion process and on the composition of fire emissions, or emission factors, have not yet been quantified at large scales. Current fire emission inventories assume constant emission factors per biome and hence simplify the combustion process. Within the ESA-funded Sense4Fire project, we developed a novel digital twin satellite-based data-model fusion approach (S4F) that integrates various satellite datasets of vegetation, fire and atmospheric composition with a fuel and combustion model to estimate the effects of fire combustion dynamics and emission factors of individual fire events on regional fire emissions. The S4F approach makes use the spatial information from AGB and canopy height maps, the annual temporal information from land cover maps, and the daily to weekly temporal information from LAI, vegetation optical depth and soil moisture time series to retrieve information on the spatial variability and temporal dynamics of fuel loads and fuel moisture content for different fuel types. We additionally use satellite-based fire radiative power observations from VIIRS and Sentinel-3 to calibrate fuel consumption, and of field and inventory databases to calibrate biomass components, combustion completeness, and the dynamic emission factors. Independent field datasets of fuel loads and fuel consumption and of top-down fire emissions of CO and nitrogen oxide (NOx) derived from Sentinel-5p are used to evaluate the retrieved results. The results for the Amazon/Cerrado regions show that the availability of woody debris is the main control on combustion efficiency and on regional total fire emissions. The results demonstrate the need for new satellite observations and in situ measurements of surface fuels and emission factors to better quantify the role of wildfire combustion on atmospheric composition and the carbon cycle.