The increased frequency and spread of tropical peat fires over the last two decades have attracted global attention because they cause significant environmental and health impacts at local to global scales. To understand the relative importance of key factors controlling tropical peatland burning events, we developed PeatFire, an agent-based model simulating the interaction between human-induced ignitions, fire and peat characteristics. The model describes (1) above- and belowground fires, which spread independently but interact with each other; (2) above- and belowground biomass; and (3) the watertable determining peat dryness and susceptibility to fire. We applied PeatFire to a region in South Sumatra that has experienced profound natural rainforest loss due to peat fires. Sensitivity analysis of the model suggests that fire sizes depend mostly on watertable depth, peat-dry-index and number of dry days before ignition. Using pattern-oriented modelling, these factors were parameterised so that the model output matches spatiotemporal fire patterns observed in the study region in 2015. Our results emphasise the risk of a sudden shift from moderate fire occurrence to complete burning and highlight the importance of local context to peatland regulation, which should consider both biophysical and socioeconomic factors and strategies for peatland fire management.