Bayesian spatial modeling of extreme daily rainfall return levels in Honduras

Abstract

Modeling extreme values in precipitation is very important, and one way to quantify them is by means of return levels. The complete model used for the estimation is developed in a first stage associated with the parameters of the Generalized Pareto Distribution(GPD) and a second stage associated with the exceedance rates. The models are approached by means of hierarchical models. The spatial component of the phenomenon is taken into account using covariates. The best model is chosen using the marginal log-likelihood criterion, and the estimates of the chosen models are used jointly in order to construct return maps with the same estimates. This study is being developed for Honduras taking daily precipitation values from 1972 to 2012 in 59 meteorological stations, showing that elevation is a covariate that influences the estimation of the GPD parameters and the exceedance rates have a constant behavior for all stations.