Parthasarathy, B., H. F. Diaz, and J. K. Eischeid, 1988: Prediction of all-India summer monsoon rainfall with regional and large-scale parameters. J. Geophys. Res., 93, 5341-5350.
By using multiple regression, several statistical models were developed to predict Indian monsoon rainfall from conditions antecedent to the beginning of the monsoon season (defined as the period June-Sept.). The predictors were selected in a stepwise fashion from a set of 11 predictors. These predictors were, in turn, chosen from a larger pool of 31 potential predictors on the basis of the strength of their correlation with all-India rainfall and whether the predictor variable exhibited a significant difference in predictor mean values between wet (seasonal rainfall exceeding one standard deviation from the long-term mean) and dry (seasonal rainfall lower than one standard deviation below the mean) years over the period 1951-1980. The models account for 70-83% of the interannual variance of all-India rainfall, depending upon the calibration period. The regressions were tested on sets of independent years. In these trials, the amount of predictand (rainfall) variance realized varied from near zero to ~60%, depending upon the period. The models with the highest skill (except one) actually included data from years prior to the calibration period. Hence, these models were really performing hindcasts, and, as such, would be useful only in a diagnostic rather than a prognostic mode. However, the sliding calibration model, where the regression coefficients are recalculated annually, did show some predictive skill. Many of the model predictors, such as the mean April latitude of the 500-mb ridge over India, have been previously identified in the work of other investigators. New predictors, such as the preceding winter-to-spring sea level pressure change at Bombay, India, which was a leading predictor in all the models, could alone account for 50-60% of the predictand variance. The latitudinal position of the 500-mb ridge axis in April along 75°E longitude was generally the second predictor to be entered. The regression model based on the earliest development sample (1951-1970), however, had the Southern Oscillation Index of Tahiti-Darwin winter-to-spring sea level pressure tendency as its number 2 predictor.