An Extreme Value Mixture model to assess drought hazard in West Africa - Université Clermont Auvergne Access content directly
Preprints, Working Papers, ... Year : 2021

An Extreme Value Mixture model to assess drought hazard in West Africa

Abstract

A critical stage in drought hazard assessment is the definition of a drought event, and the measure of its intensity. Actually, the classical approach imposes to all climatic region the same set of thresholds for drought severity classification, hence resulting in a loss of information on rare events in the distribution tails, which are precisely the most important to catch in risk analysis. In order to better assess extreme events, we resort to an extreme value mixture model with a normal distribution for the bulk and a Generalized Pareto distribution for the upper and lower tails, to estimate the intensity of extreme droughts and their occurrence probability. Compare to the standard approach to drought hazard, which relies on a standardized precipitation index and a classification of drought intensity established from the cumulative standard normal distribution function, our approach allows the drought threshold and the occurrence probability of drought to depend on the specific characteristics of each precipitation distribution. An application to the West Africa region shows that the accuracy of our mixture model is higher than that of the standard model. The mixture performs better at modelling the lowest percentiles and specifically the return level of the centennial drought, which is generally overestimated in the standard approach.
Fichier principal
Vignette du fichier
E&D_23.pdf (955.71 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03297023 , version 1 (23-07-2021)

Licence

Attribution

Identifiers

  • HAL Id : hal-03297023 , version 1

Cite

Abdoulaye Sy, Catherine Araujo-Bonjean, Marie-Eliette Dury, Nourddine Azzaoui, Arnaud Guillin. An Extreme Value Mixture model to assess drought hazard in West Africa. 2021. ⟨hal-03297023⟩
278 View
387 Download

Share

Gmail Facebook X LinkedIn More