Space-Time Trend Detection and Dependence Modeling in Extreme Event Approaches by Functional Peaks-Over-Thresholds: Application to Precipitation in Burkina Faso

Abstract

In this paper, we propose a new method for estimating trends in extreme spatiotemporal processes using both information from marginal distributions and dependence structure. We combine two statistical approaches of an extreme value theory: the temporal and spatial nonstationarities are handled via a tail trend function in the marginal distributions. The spatial dependence structure is modeled by a latent spatial process using generalized $\ell$ -Pareto processes. This methodology for trend analysis of extreme events is applied to precipitation data from Burkina Faso. We show that a significant increasing trend for the 50 and 100 year return levels in some parts of the country. We also show that extreme precipitation is spatially correlated with distance for a radius of approximately 200 km.