A geostatistical approach for extreme precipitations simulations in the Venice lagoon

Abstract

In this work, extreme precipitations in the Venice lagoon will be simulated combining Extreme Value Theory (EVT), Generalized Additive Models (GAM) and geostatistics. Extreme events are identified as exceedances of a high threshold defined as a fixed quantile and fitted through quantile regression based on the Asymmetric Laplace Distribution (ALD). Then the Generalized Pareto Distribution (GPD) is used to model the excesses. Spatio-temporal variation of the parameters of both distributions is captured via the flexible framework of GAM. Obtained marginal models are coupled under a copula-based technique, forming a Gaussian process for high-resolution simulations of extreme events in space and time. The random field is assumed to have zero mean and its covariance is described using a separable parametric correlation function.