Modelling wildfire activity in wildland–urban interface (WUI) areas of Sardinia, Italy

Background Wildfire frequency, magnitude and impacts in wildland–urban interface (WUI) areas are increasing in the Mediterranean Basin. Aims We investigated the role played by socio-economic, vegetation, climatic, and zootechnical drivers on WUI wildfire patterns (area burned and wildfire ignitions) in Sardinia, Italy. Methods We defined WUI as the 100-m buffer area of the anthropic layers. We created a comprehensive and multi-year dataset of explanatory variables and wildfires, and then trained a set of models and evaluated their performances in predicting WUI fires. We used the best models to assess the single variable’s importance and map wildfire patterns. Key results Random Forest and Support Vector Machine were the best performing models. In broad terms, wildfire patterns at WUI were influenced by socio-economic factors and herbaceous vegetation types. Conclusions Machine learning models can be useful tools to predict wildfire ignitions and area burned at WUI in Mediterranean areas. Implications Improved knowledge of the main drivers of wildfires at WUI in fire-prone Mediterranean areas can foster the development or optimisation of wildfire risk reduction and prevention strategies.