Parameter Flexible Wildfire Prediction Using Machine Learning Techniques: Forward and Inverse Modelling

Parameter identification for wildfire forecasting models often relies on case-by-case tuning or posterior diagnosis/analysis, which can be computationally expensive due to the complexity of forward prediction model. In this paper, we introduce an efficient parameter flexible fire algorithm based machine learning and reduced order modelling techniques. Using a training dataset generated by physics-based simulations, method forecasts burned area at different time steps with low computational cost. We then address bottleneck estimation developing novel inverse approach relying data assimilation techniques (latent assimilation) in space. The modellings are tested two recent large events California. Satellite observations used validate identify model parameters. By combining these approaches, system manages integrate real-time adjustment, leading more accurate future predictions.