Mapping natural fire hazards in Middle Urals based on random forest model

In this study, we created the maps of natural fire hazard for Perm krai and Sverdlovsk oblast, based on the Random Forest Regressor machine learning model. It was trained with the materials on wildfires occurred in Perm krai in 2010–2022. Publicly available datasets characterizing forest cover, climate, terrain and the degree of anthropogenic impact on the territory were used as predictors. The most important among them are the distribution of pine forests, bogs (positive correlation), and elevation (negative correlation). For the territory of Perm krai, we found a good accordance of predicted hazard to observed spatial distribution of forest fires and the corresponding losses. The data for Sverdlovsk oblast were used as an independent sample, and here the same above mentioned correlation was also revealed. Over 60 % of the occasions and their area fall into the territories with high estimated fire hazard. The created maps can be used along with weather indices to improve operational forecasting of wildfires