Grasses fuel most fires on Earth and strongly influence local fire behaviour through traits that determine how flammable they are. Therefore, grass communities that differ in their species and trait compositions give rise to significant spatial variation in savanna fire regimes across the world, which cannot be otherwise explained. Likewise, fire regimes are continuously modified by alterations to savanna grass community traits, through species introductions and climatic changes. However, current representation of grassy fuels in global fire models misses important variation and therefore limits predictive power. The inclusion of grass trait diversity in models, using remotely sensed trait proxies, for example, will greatly improve our ability to understand and project savanna fires and their roles in the Earth system.