A Generative Model for Surrogates of Spatial-Temporal Wildfire Nowcasting

Recent increase in wildfires worldwide has led to the need for real-time fire nowcasting. Physics-driven models, such as cellular automata and computational fluid dynamics can provide high-fidelity spread simulations but they are computationally expensive time-consuming. Much effort been put into developing machine learning models prediction. However, these often region-specific require a substantial quantity of simulation data training purpose. This results significant amount different ecoregions. In this work, generative model is proposed using three-dimensional Vector-Quantized Variational Autoencoders generate spatial-temporal sequences unseen wildfire burned areas given ecoregion. The tested ecoregion recent massive event California, known Chimney fire. Numerical show that succeed generating coherent structured scenarios, taking account impact from geophysical variables, vegetation slope. Generated also used train surrogate predicting dissemination, which on both real event.