Most attempts at predicting south Asian monsoon variability have concentrated on seasonally averaged rainfall over the Indian subcontinent some months in advance using regional and remote boundary effects as predictors. Overall, about 30% of the variance of mean seasonal monsoon rainfall can be explained, but the statistics appear to be nonstationary and correlations vary strongly on interdecadal time scales. Model intercomparisons show that climate models have difficulty in simulating even gross-scale features of the monsoon such as mean summer rainfall, and there is little demonstrated skill when the models are used in predictive mode. Even if the statistics were stable and model predictions were skillful it is argued that the information is not readily downscalable because the mean rainfall does not define the timing or number of intraseasonal variations or even the spatial distributions of the seasonal mean rainfall. Based on these concerns, it is argued that skillful and timely forecasts of intraseasonal variability possess a greater potential utility for agriculture and water resource management and should be the highest priority for prediction within the monsoon regions.

A physically based empirical Bayesian prediction scheme is developed for forecasting regional intraseasonal variability of the monsoon. Ten predictors are chosen that depict the morphology of the monsoon intraseasonal mode. The scheme employs a wavelet-banding technique and linear regression to forecast 5-day average rainfall variability over regions of south Asia 15–30 days (i.e., six 5-day lags) in the future. Hindcasts conducted for the central Indian region for the period 1992–2002 show considerable skill out to 30 days in both the timing and amplitude of the intraseasonal oscillations. Skill, albeit reduced, is also found in smaller regions such as the Indian states of Rajasthan and Orissa. The use of wavelet analysis to sort time series and isolate each band from the noise generated in other bands, together with the careful choice of predictors, are the defining elements of the scheme. Anomaly correlations of rainfall in the 28–80-day band in central India are 0.88, 0.76, 0.73, 0.66, and 0.58 for 10,15, 20, 25, and 30 days, respectively. Similar skill is found for forecasting the discharge of the Ganges and Brahmaputra into Bangladesh. The potential utility of these forecasts for applications in agriculture and water resource management is discussed together with the possible use of the empirical scheme as a diagnostic tool and as a guide for the development of a new type of numerical model.

This content is only available as a PDF.

Footnotes

Schools of Earth and Atmospheric Sciences and Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia