Development of an 8-Parameter Probabilistic Artificial Neural Network Model for Long-Range Monsoon Rainfall Pattern Recognition over the Smaller Scale Geographical Region -District

Attempts to recognize pattern of monsoon rainfall over the smaller scale geographical region (district) 8-Parameter Probabilistic ANN model have been developed. Eleven neurons in input layer to input eleven years rainfall data time series. Eleven neurons in hidden layer, one output neuron for observation of twelve year rainfall, 132 trainable weights in three layers, transfer function sigmoid f (x) = 1/1+e
<sup>-deltax+

eta</sup>
and network training error level ranging from 0.0005 to 0.0001 have been used in the model development. 54 years data for 1951-2004 have been used, of which the first 41 years (1951-1991) of data are used for training the network and data for the period 1991-2004 are used independently for validation. The performances of the model in pattern recognition and prediction have been found to be extremely good and better evaluated over statistical models. The model developed and their evaluations have been presented in this paper.