Founded by the Swiss National Science Foundation ( Project number IZSEZ0_186483)
The main objective of the study is to elaborate a wildfire susceptibility map for Liguria region (Italy) by applying Random Forest, an ensemble Machine Learning algorithm based on decision trees. RF gives as output a prediction value, expressed as the probability for each pixel of burning under the assumption of a set of predisposing factors. The model implements the following: i) comparison between the “standard model” and the “neighboring vegetation model”; ii) comparison between the random selection of testing dataset (in space) versus five- and nine-folds spatial cross validation; iv) prediction values as main output of random forest, allowing to elaborate susceptibility maps; v) evaluation of the contribution of each predisposing factor to the model.
Mirko d'Andrea and Marj Tonini