Development of ANN models using monthly rainfall for central Telangana region

Abstract

In the present study, artificial neural network technique has been employed to predict monthly rainfall for Medak, Khammam and Warangal stations of Central Telangana, India. The eighty-five years (January, 1901 to December, 1985) of rainfall data were used for training of models and twenty-eight years (January, 1986 to December, 2014) of rainfall data were used for testing of models. Gamma test, autocorrelation function and cross correlation function were used for selection of appropriate input variables. The ANN models were trained using multilayer perceptron with two learning rules i.e. Levenberg-Marquardt and Delta-bar-delta and two transfer functions viz. Sigmoid axon and Tanh axon. It was observed that the better results of monthly rainfall prediction of developed models were observed when rainfall of adjoining stations was used as inputs variable as compared to lagged rainfall of the same station. suggest that the M-8 model, K-7 model and W-5 may be used to predict monthly rainfall of Medak, Khammam and Warangal stations respectively for Central Telangana region.