Effect of weather conditions, geography and population density on wildfire occurrence: A Bayesian Network model

Occurrences of wildfires are related to weather conditions and human intervention and can only be predicted probabilistically. In this paper, the potential of Bayesian Networks for such predictions is investigated. A Bayesian Network is constructed, which expresses the effect of weather conditions, land cover and human presence on the rate of wildfire occurrences. The model is based on both temporal and spatial data. The parameters of the model are inferred from data obtained for the Greek Mediterranean island of Rhodes. Initial results show a dependence between human population density and wildfire occurrence. The selected indicator for weather conditions, a commonly used fuel moisture index, is found to be ill-suited for predicting wildfire occurrence on Rhodes, possibly due to the specifics of the Mediterranean climate. Future work is needed to identify and include relevant influencing factors, which is facilitated by the Bayesian network modeling approach.