The development of a neurocomputing technique to forecast the average winter shower in India has been modeled from 48 years of records (1950–1998). The complexities in the rainfall–sea surface temperature relationships have been statistically analyzed along with the collinearity diagnostics. The presence of multicollinearity has been revealed and a variable selection has been executed accordingly. The absence of persistence has also been revealed. For this reason, an Artificial Neural Net Model as a predictive tool for the said meteorological event in the form of a Multiple Layer Perceptron has been generated with a sea surface temperature anomaly and monthly average winter shower data over India during the above period. After proper training and testing, a Neural Net model with small prediction error is developed and the supremacy of the Artificial Neural Net over conventional statistical predictive procedures has been established statistically.

This content is only available as a PDF.