Generalized linear model approach for modeling rainfall occurrence

Abstract

Rainfall modeling has been used to identify the characteristics of the rainfall occurrence and rainfall amount. The daily rainfall data was obtained from the Malaysian Drainage and Irrigation Department for six selected rain gauge stations over period of 32 years ranging from 1980 to 2011 in Peninsular Malaysia. The purpose of this study is to model the rainfall occurrence that varies as a function of time of year using Generalized Linear Model (GLM) and compare rainfall patterns between stations or regions based on the smoothing curves. The rainfall occurrence was fitted with a two-state first order Markov chain model in this study. The transition probabilities for first order Markov chain model were calculated using the maximum likelihood method. Since the transition probabilities for all stations are not stationary, smoothing curves to model the transition probabilities using Fourier series throughout the year were suggested. Analysis of deviance table was obtained by using statistical program R project. The result shows that all the stations display a bimodal pattern of rainfall with two distinct peaks except station Hospital Pontian for transition probabilities of dry day to rainy day and station Ipoh for transition probabilities of rainy day to rainy day. Based on the observations, rainfall modeling can be applied in agriculture sector.