EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Computing wildfire Susceptibility Maps at the national level in Italy: a Machine Learning approach

Andrea Trucchia1, Giorgio Meschi1, Marj Tonini2, and Paolo Fiorucci1
Andrea Trucchia et al.
  • 1CIMA Research Foundation, Italy (
  • 2Université de Lausanne, Institute of Earth Surface Dynamics (IDYST)

Wildfires are a serious social and environmental issue in the Mediterranean basin, menacing human lives, infrastructures and ecosystems. Italy, due to land cover, orography  and climate, expresses a complex wildfire regime that is worth investigating. Static maps, such as susceptibility, hazard and risk maps, are valid allies for wildfire management and land use planning. In particular, the wildfire susceptibility is defined as the spatially distributed probability of experiencing wildfire at a certain point, depending only on  the intrinsic characteristics of the territory. In the presented work, a Machine Learning  (ML) model  is built following a similar approach of [1], to produce different National Scale susceptibility maps for Italy. The adopted algorithm is Random Forest, an ensemble ML method.

Since Italy exhibits two different wildfire seasons, the summer and the winter one, two maps are produced, to identify the different regimes. The presented analysis at the national scale allows the experts and the decision makers to have a deep understanding on the wildfire regimes, and may constitute a solid paradigm for wildfire risk management. The Random Forest associated  a data-set of geographic (orography, land cover), anthropic (distance from crops, roads and urban features) and climatic information (mean precipitation and temperature) to the database of ground-retrieved burned area polygons.  The classifier is then employed to evaluate each pixel of the study area, producing the susceptibility map. The performance of the adopted frameworks are evaluated via spatial cross validation and the evaluation of mean squared error and Area Under the RoC curve  on a test dataset.  A subsequent analysis of the importance of each input factor through the Gini impurity method allows to spot the most important variables, paving the way for further improvements in the dataset.




[1] Tonini, M.; D’Andrea, M.; Biondi, G.; Degli Esposti, S.; Trucchia, A.; Fiorucci, P. A Machine Learning-Based Approach for Wildfire Susceptibility Mapping. The Case Study of the Liguria Region in Italy. Geosciences 2020, 10, 105. 

How to cite: Trucchia, A., Meschi, G., Tonini, M., and Fiorucci, P.: Computing wildfire Susceptibility Maps at the national level in Italy: a Machine Learning approach, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3981,, 2022.


Display file

Comments on the display

to access the discussion