Application of Deep Learning for Wildfire Risk Management: Preliminary Results

Due to ongoing climate change and the related increased risk of drought, wildfires in the Mediterranean are expected to become more frequent and intense. This note introduces the use of Deep Learning to produce susceptibility maps for wildfire risk assessment to tackle this scenario. Indeed, thanks to the improved computational capabilities of modern GPUs, recent literature has pointed out that it is possible to train Deep Learning models on big datasets by outperforming standard approaches in several application domains, including the one object of this study. An Artificial Neural Network (ANN) model has been developed and applied to the Calabria region (southern Italy) study area. The network has been trained and validated over ten years of data by considering both meteorological and geomorphological data, and human factors. Preliminary results are promising, evidencing a higher network’s accuracy to the operational state-of-the-art Fire Weather Index (FWI) method, currently applied for forecasting and mitigation purposes in the study area.