Spatial models for flood risk assessment

The problem of computing risk measures associated to flood events is extremely important not only from the point of view of civil protection systems but also because of the necessity for the municipalities of insuring against the damages. In this work we propose, in the framework of an integrated strategy, an operating solution which merges in a conditional approach the information usually available in this setup. First we use a Logistic Auto-Logistic (LAM) model for the estimation of the univariate conditional probabilities of flood events. This approach has two fundamental advantages: it allows to incorporate auxiliary information and does not require the target variables to be independent. Then we simulate the joint distribution of floodings by means of the Gibbs Sampler. Finally we propose an algorithm to increase ex post the spatial autocorrelation of the simulated events. The methodology is shown to be effective by means of an application to the estimation of the flood probability of Italian hydrographic regions.