On the Prediction of India Monsoon Rainfall Anomalies

Abstract

A complex of anomalies in the premonsoon large-scale circulation setting heralds the interannual variability of India summer monsoon rainfall. The most prominent precursors of precipitation anomalies are the latitude position of the upper-air ridge over India, apparently reflecting the persistence of the boreal winter wind regime and its consequence for the establishment of the summer upper-air circulation; the temperature in southern Asia and the adjacent North Indian Ocean waters, a factor instrumental in heat-low development and hence the establishment of meridional pressure gradients and lower-tropospheric airstreams from the Southern Hemisphere; and indices of the Southern Oscillation, capturing pressure departure patterns spanning the global tropics. Stepwise multiple regression is used to extract from this ?anomaly complex? the variance most pertinent to the interannual variability of Southwest monsoon rainfall, observations of pertinent elements being available for the period 1939?81. Regression models developed on a portion of this record are then used to predict the summer monsoon rainfall anomalies of the years 1966?81. The correlations between the various precursors and the rainfall anomalies vary in the course of 1939?81, being, on the whole, strongest in the 1950s and 1960s. While the April latitude position of the 500 mb ridge along 75°E proves to be the strongest predictor, performance is improved by inclusion of other elements representing preseason temperature and the Southern Oscillation. Correlation, root-mean-square error, bias, and absolute error are used as measures of forecast performance. A set of experiments with the dependent dataset, ending in 1965, indicates that a regression base period of about 20 yr is optimal for predictions into the independent portion of the record. Another set of experiments, in which the regression base periods are successively updated to the year immediately preceding the year to be forecast, shows no improvement of predictions over the fixed regression base periods. ?Cross-validation? is not found less demanding than prediction proper. It is demonstrated that about half of the interannual variance in monsoon rainfall can be predicted from antecedent anomalies in the large-scale circulation setting.