Anthropogenically driven declines in tropical savannah burnt area1,2 have recently received attention due to their effect on trends in global burnt area3,4. Large-scale trends in ecosystems where vegetation has adapted to infrequent fire, especially in cooler and wetter forested areas, are less well understood. Here, small changes in fire regimes can have a substantial impact on local biogeochemistry5. To investigate trends in fire across a wide range of ecosystems, we used Bayesian inference6 to quantify four primary controls on burnt area: fuel continuity, fuel moisture, ignitions and anthropogenic suppression. We found that fuel continuity and moisture are the dominant limiting factors of burnt area globally. Suppression is most important in cropland areas, whereas savannahs and boreal forests are most sensitive to ignitions. We quantify fire regime shifts in areas with more than one, and often counteracting, trends in these controls. Forests are of particular concern, where we show average shifts in controls of 2.3–2.6% of their potential maximum per year, mainly driven by trends in fuel continuity and moisture. This study gives added importance to understanding long-term future changes in the controls on fire and the effect of fire trends on ecosystem function.