Modelling Extreme Rainfall in Northern Thailand with Estimated Missing Values

Abstract

We analyzed annual maximum daily rainfall in upper northern Thailand using some generalized extreme value (GEV) distributions. The data consists of 16 locations for the period from 1957 to 2012. But some of observations are missing for some locations. Thus we propose a heuristic method to impute the missing values. It is a modified EM algorithm with robust estimation of conditional expectation. The method takes account into the spatial dependency between locations of weather stations. After the estimated missing values are included to complete data, we fitted stationary and non-stationary Gumbel and GEV distributions. Among these models, the best model is selected based on the likelihood ratio test for each station. Using the selected model, return values for several return periods are obtained and contour plots are drawn. A cluster analysis is conducted based on the estimated return levels. The information obtained from our study is useful for the agencies involved in water management in terms of strategic planning and the prevention of flood, related to public safety.