Landscape fire is a key but poorly understood component of the global carbon cycle. Predicting biomass consumption by fire at large spatial scales is essential to understanding carbon dynamics and hence how fire management can reduce greenhouse gas emissions and increase ecosystem carbon storage. An Australia-wide field-based survey (at 113 locations) across large-scale macroecological gradients (climate, productivity and fire regimes) enabled estimation of how biomass combustion by surface fire directly affects continental-scale carbon budgets. In terms of biomass consumption, we found clear trade-offs between the frequency and severity of surface fires. In temperate southern Australia, characterised by less frequent and more severe fires, biomass consumed per fire was typically very high. In contrast, surface fires in the tropical savannas of northern Australia were very frequent but less severe, with much lower consumption of biomass per fire (about a quarter of that in the far south). When biomass consumption was expressed on an annual basis, biomass consumed was far greater in the tropical savannas (>20 times that of the far south). This trade-off is also apparent in the ratio of annual carbon consumption to net primary production (NPP). Across Australia's naturally vegetated land area, annual carbon consumption by surface fire is equivalent to about 11% of NPP, with a sharp contrast between temperate southern Australia (6%) and tropical northern Australia (46%). Our results emphasise that fire management to reduce greenhouse gas emissions should focus on fire prone tropical savanna landscapes, where the vast bulk of biomass consumption occurs globally. In these landscapes, grass biomass is a key driver of frequency, intensity and combustion completeness of surface fires, and management actions that increase grass biomass are likely to lead to increases in greenhouse gas emissions from savanna fires.