A stochastic generator for high-resolution spatio-temporal extreme rainfall

par Chloé Serre-Combe

mardi 10 févr. 2026, 15:00 → 16:00 Europe/Paris

Stochastic precipitation generators are essential tools for urban flood risk analysis. We propose a framework for modeling the distribution and the spatio-temporal dependence of high-resolution rainfall using data from Montpellier’s Urban Observatory rain gauge network. For univariate modeling, at the point level, the Extended Generalized Pareto Distribution (EGPD) is used to capture both moderate and extreme rainfall without explicit threshold selection, simplifying parameter estimation. Extreme spatio-temporal dependence is modeled with an r-Pareto process based on an underlying Gaussian dependence structure. Unlike max-stable processes, which focus on block maxima, r-Pareto processes offer greater flexibility by using the Peaks Over Threshold (POT) framework. By incorporating a non-separable spatio-temporal variogram with advection, we account for the possible horizontal movement of precipitation clouds, enabling realistic simulations of spatio-temporal rainfall patterns. Variogram parameters are estimated using a novel composite likelihood approach based on bivariate joint exceedance indicators. This methodology will form the core of a stochastic precipitation generator for simulating high-resolution rainfall events on the Montpellier region, which will be integrated into a mechanistic water flow model for flood risk analysis.