Long-Range Monsoon Rainfall Pattern Recognition and Prediction for the Subdivision 'EPMB' Chhattisgarh Using Deterministic and Probabilistic Neural Network

Abstract

Attempts to predict long-range monsoon rainfall over the subdivision EPMB, Three layer perception feed forward back propagation deterministic and probabilistic artificial neural network models have been developed. 61 years data for 1945-2006 have been used, of which the first 51 years (1945-1995) of data are used for training the network and data for the period 1996-2006 are used independently for validation. We have found that the mean absolute deviation (% of mean) of the model is less than and half of the standard deviation (% of mean) in the independent period (1996-2006) of the subdivision in deterministic forecast. Correlation between actual and particular model predicted values are more than 0.8 for the districts and 0.7 for the whole subdivision in deterministic forecast. However performance of the model in probabilistic forecast is better evaluated over deterministic forecast. The models developed and their evaluations have been presented in this paper.