ProbFire: a probabilistic fire early warning system for Indonesia

Abstract. Recurrent extreme landscape fire episodes associated with drought events in Indonesia pose severe environmental, societal and economic threats. The ability to predict months advance would enable relevant agencies communities more effectively initiate fire-preventative measures mitigate impacts. While dynamic seasonal climate predictions are increasingly skilful at predicting fire-favourable conditions Indonesia, there is little evidence that such information widely used yet by decision makers. In this study, we move beyond forecasting risk based on timescales (i) develop a probabilistic early warning system for (ProbFire) multilayer perceptron model using ECMWF SEAS5 (fifth-generation system) forecasts together forest cover, peatland extent active-fire datasets can be operated standard computer; (ii) benchmark the performance of new 2002–2019 period; (iii) evaluate potential benefit integrated Indonesia. ProbFire's event probability outperformed climatology-only 2- 4-month lead times south Kalimantan, Sumatra Papua. central Sumatra, an improvement was observed only 0-month time, while west Kalimantan did not offer any additional over climatology-only-based predictions. We find coupled prediction confer substantial benefits wide range stakeholders involved management provide blueprint future operational systems integrate non-climate features.