Taking advantage of satellite data, large datasets of fire records and cloud computing for modelling potential fire severity useful for better assess fire risk

Fire risk analyses, usually focused on fire hazard (i.e. the probability of fire occurrence), often neglect an important issue such as the sensitivity/vulnerability (i.e., the degree of potential damage, sensus IPCC) of different locations within the area of interest.  Such lack of consideration comes from past data processing constrains that limited fire severity studies to analyse only single or few fire events. Nowadays, online data repositories and processing platforms (e.g. Google Earth Engine) allow to easily integrate and process a vast amount of data from multiple sources that might prove useful for developing tailored tools for decision making. Here, we present an example for predicting potential fire severity based on the analysis of more than 1 000 fire events from southern France and western Italy which integrates climate, topographical and remote sensing variables. Furthermore, we assessed if the model “used” the explanatory variables under a meaningful biophysical sense.   Using the random forest algorithm and the relativized difference of the Normalized Burn Ratio (rdNBR) as proxy of fire severity, we reach to explain up to 75% of the variability in the data with most of the variables showing a clear and interpretable effect. Our results suggests that this type of approach might prove useful for better address fire risk assessments.