Mapping the hazard of extreme rainfall by peaks-over-threshold extreme value analysis and spatial regression techniques

Abstract

The occurrence of rainfalls of high magnitude constitutes a primary natural hazard in many parts of the world, and the elaboration of maps showing the hazard of extreme rainfalls has great theoretical and practical interest. In this work a procedure based on extreme value analysis and spatial interpolation techniques is described. The result is a probability model in which the distribution parameters vary smoothly in space. This methodology is applied to the middle Ebro valley (Spain), a climatically complex area with great contrasts due to relief and the exposure to different air masses. The database consists on 43 daily precipitation series from 1950 to 2000. Since rainfall tends to occur highly clustered in time in the area, a de-clustering process was applied to the data, and the series of daily cluster maxima were used hereinafter. The meanexcess plot and error minimizing were used to find an optimum threshold value to retain the highest records (peaks-over-threshold approach), and a Poisson- Generalised Pareto model were fitted to the resulting series. The at-site parameter estimates (location, scale and shape) were regressed upon a set of location and relief variables, enabling the construction of a distributed probability model. The advantages of this method to obtain maps of extreme precipitation hazard are discussed in depth.