Lightning-caused fire occurrence has been modelled for two different Spanish regions, Madrid andAragon, based on meteorological, terrain, and vegetation variables. The model was built on two very contrasting regions, one presenting low number of lightning-caused fires whereas the other presented a high occurrence. The research was conducted between May and September, which happens to be the most lightning-fire prone period in Spain, for a three year interval starting in 2002 up to 2004.
A time-invariant model for lightning-caused fire occurrence was developed for each region at a spatial resolution of 3 km ×3 km. The probabilistic models were based on the logistic regression, aiming to explain the probability of having at least a lightning-fire during the three year period.
Results showed that the number of thunderstorms during the three-year period was the most significantvariable in the model, where an increasing number of thunderstorms leads to a higher probability of occurrence. Validation was assessed through the Receiver Operator Characteristic, showing a good agreement between the modelled probabilities and the reported lightning-caused fires, with an Area Under the Curve around 0.7 in Aragon. However, the model in Madrid showed a poor AOC performance, showing therefore insights that the study period should be larger due to the low occurrence of lightning-fires in
that region.