Analysing predictability in Indian monsoon rainfall: A data analytic approach

Abstract

This paper examines monthly and annual data to analyse predictability in the Indian monsoon rainfall. The periodic structure in the time series data is extracted using wavelets and the residual random part is separately modeled using artificial neural networks (ANN). Although wavelet and neural network based hybrid techniques have been widely applied in the recent years, the present approach has not been investigated so far. Our results show that the estimated periodic and random components comprise 30 and 15 %, respectively, variance of the total rainfall in case of annual data, whereas the model explains 93 % of variance in case of monthly data. It is shown that the prediction is more accurate when periodic and random parts are treated separately.