Development of additive form of time series models for rainfall

Abstract

Using monthly, weekly and monsoon rainfall data (1988 to 2016), additive form of time series models were developed. Turning point and Mann-Kendall (M-K) tests were carried out for identifying the trend component and Fourier series analysis were used for modelling the periodic component. Modelling of dependent stochastic component was done using ARMA (4,0,8) model for monthly rainfall and AR (1) model for both weekly and daily monsoon rainfall. The Portmanteau test was used to check the independence of stationary stochastic independent component. Using Box-Cox transformation, independent stochastic component was normalised and its modelling was done using normal distribution function. Time series models were evaluated using several statistical measures, and they indicated a good model fitness. Developed time series models were also validated with 8 years rainfall data.