Wavelet-based combination approach for modeling sub-divisional rainfall in India

Abstract

Agriculture in India is highly sensitive to climatic variations particularly to rainfall and temperature; therefore, any change in rainfall and temperature will influence crop yields. An understanding of the spatial and temporal distribution and changing patterns in climatic variables is important for planning and management of natural resources. Time series analysis of climate data can be a very valuable tool to investigate its variability pattern and, maybe, even to predict short- and long-term changes in the series. In this study, the sub-divisional rainfall data of India during the period 1871 to 2016 has been investigated. One of the widely used powerful nonparametric techniques namely wavelet analysis was used to decompose and de-noise the series into time–frequency component in order to study the local as well as global variation over different scales and time epochs. On the decomposed series, autoregressive integrated moving average (ARIMA) and artificial neural network (ANN) models were applied and by means of inverse wavelet transform, the prediction of rainfall for different sub-divisions was obtained. To this end, empirical comparison was carried out toward forecast performance of the approaches namely Wavelet-ANN, Wavelet-ARIMA, and ARIMA. It is reported that Wavelet-ANN and Wavelet-ARIMA approach outperforms the usual ARIMA model for forecasting of rainfall for the data under consideration.