Evaluation of a Dynamical Basis for Advance Forecasting of the Date of Onset of Monsoon Rainfall over India

Abstract

Abstract The onset of the Indian summer monsoon (ISM) represents one of the most dramatic transitions in the regional circulation pattern. The onset also marks the beginning of the main rainy season for India; advanced and accurate forecast of the date of the onset of monsoon (DOM) thus has application in many sectors. Although the standard deviation (σ) in DOM over the past hundred years is only 7 days, nearly 50% of the cases show large (>1σ) deviations; forecasting of DOM, especially for the extreme years is thus nontrivial and is rarely attempted because of the poor skill of most GCMs in the long-range prediction of daily ISM rainfall. A primary cause for the poor skill in forecasting parameters like rainfall appears to be the loss of predictability due to noise introduced by local synoptic processes. However, sharp transitions in the regional circulation pattern and associated rainfall, which are likely to be less affected by synoptic noise, may have higher predictability, somewhat similar to the way that monthly mean parameters are more predictable. This premise is explored for advanced forecasting of the onset of ISM over Kerala, India, and it is shown that significant skill is possible in advanced forecasting of DOM. A general circulation model (GCM) with a special feature, variable resolution, and an objective debiasing of daily rainfall forecast, is used to meet the special requirements of forecasting DOM. Based on a set of objective and validated criteria, hindcasts of DOM are generated in a complete operational setting from a five-member ensemble for each year for the period 1980–2003. The hindcasts are evaluated in terms of a number of parameters; as well as against a climatological forecast (null hypothesis), for 70% of the forecasts, the mean absolute error is less than that of the climatological forecasts. Furthermore, in contrast to the climate forecasts, these forecasts capture 7 out of 9 large (>1σ in observation) departures from the mean within the mean error, which implies high skill. Implications of the results for predicting certain weather and climate processes are discussed.