Long lead monsoon rainfall prediction for meteorological sub-divisions of India using deterministic artificial neural network model

The advantages of artificial neural network technique for explaining the nonlinear behavior between the inputs and output is explored to forecast the monsoon rainfall of 36 meteorological sub-divisions of India. The model uses the past years of monsoon rainfall data only to forecast the monsoon rainfall of coming year. Monthly rainfall time series data for each of the 36 meteorological sub-divisions constructed by Guhathakurta and Rajeevan (2007) is used for the present study. The model captures well the input-output nonlinear relations and predicted the seasonal rainfall quite accurately during the independent period. All India monsoon rainfall forecasts were generated by using area weighted rainfall forecasts of all the sub-divisions. For the first time the idea of up-scaling is introduced in monsoon rainfall prediction using neural network technique and it is shown that up scaling helps to capture the variability of the all India rainfall better. This helps to predict the extreme years like 2002, 2004 better than the neural network model developed based on single time series of all India rainfall. However, derivation of smaller scale (sub-divisions) forecast model may be more useful than the all India forecast.