Strength in many: Ensemble-based approach for data-driven Fire Danger Index forecast

Machine learning has been applied to several aspects of forest fire management, particularly to estimate the danger associated with forest fires. An important aspect of fire danger assessment is the successful detection of fires and a low rate of false positives in the daily fire danger index forecast. We present an ensemble-based approach for forecasting the data-driven fire danger index (FDI) using a Convolution LSTM architecture, which combines elements of both CNN and LSTM. Our approach is driven by operational considerations, which require not only high fire recall but also low number of false positives flagged by the model. In this talk, I will demonstrate our results from our ensemble approach in forecasting the ensemble-average FDI.

This work is partly supported by the ARCA project which is funded by Interreg IPAADRION programme under the Interreg Funds (European Regional Development Fund and IPA III), agreement number IPA-ADRION00107.