A bayesian zero-inflated exponential distribution model for the analysis of weekly rainfall of the eastern plateau region of India
Statistical modelling of rainfall data has been a major research area of the climatologists and agro-meteorologists for quite a long time. Short-period rainfall measurements often include a high percentage of zero values, particularly during the winter weeks. Zero-inflated models are often used in modelling such datasets. In this paper, we attempt to model weekly rainfall data using zero-inflated exponential distribution. Though a frequentist approach (mainly maximum likelihood estimation) is often preferred by meteorologists, it has a few shortcomings discussed in this paper. Hence, we consider a Bayesian model, i.e., we assume the model parameters to be random instead of fixed quantities as in a frequentist approach. As some obvious parts of a Bayesian analysis, we discuss the prior choices, inference based on the posterior distributions of the parameters and calculations of the different percentage probability rainfall amounts based on the posterior predictive distributions. We analyze weekly rainfall dataset for the years 1969-2009 collected at Giridih, India. In the eastern plateau region, agricultural operations depend solely on the rainfall quantities because of the lack of irrigation facilities. We provide 10%, 30%, 50%, 70%, 90% probability rainfall amounts which would help in deciding the accurate week for a particular step of an agricultural procedure.
Hazra, Arnab; Bhattachary, Sourabh; and Banik, Pabitra, "A bayesian zero-inflated exponential distribution model for the analysis of weekly rainfall of the eastern plateau region of India" (2018). Journal Articles. 1600.