Present forecasting system
At present, the forecast for the South-West monsoon rainfall is issued in two stages. The first stage forecast for the seasonal (June to September) rainfall over the country as a whole is issued in April and the update of the April forecast is issued in June. Along with the update forecast, forecast for seasonal rainfall over four broad geographical regions of India and July rainfall over country as a whole are also issued.
For issuing the forecast for the seasonal rainfall over the country , a new statistical forecasting system based on the ensemble technique is introduced in 2007. The 8 predictors considered for the new ensemble forecast system are given in the Table. For the April forecast, the first 5 predictors given in this table are used. For the updated forecast in June, 6 predictors that include 3 predictors (first 3 predictors in this table) are used. (Details of the method used in the forecasting system is given here)
In the ensemble method, instead of relying on a single model, all possible models based on all the combination of predictors are considered. For April (June) forecast With 5 (6) predictors, 31 (63) different models were developed. Out of all the possible models, best few models were selected based on their skill in predicting monsoon rainfall during a common period. Ensemble mean was computed as the weighted average of the forecast from the selected models. The weights are proportional to the multiple correlation coefficients of the models during the training period. For developing the models, two different statistical techniques namely, Multiple Regression (MR) and Projection Pursuit Regression (PPR) were considered. While the MR technique is a conventional linear model technique and more commonly used, the PPR technique is a non-linear technique, used for the first time in forecasting Indian monsoon rainfall. The PPR method is known for its superiority in capturing the non-linear relationships between the predictors and rainfall. Verification of the results with the past data showed that the ensemble method performed better than the individual models.
The model error of the April forecast system is 5% and for the June forecast system, it is 4%.
For the forecast of July rainfall over the country as a whole, a statistical model with 6 predictors was developed using Principal Component Regression (PCR) technique. The predictors used are: North Atlantic Sea surface temperature (December of previous year)), NINO 3.4 Sea Surface Temperature (May +June), North Pacific mean sea level pressure (March), East Asia mean sea level pressure (February + March), North Atlantic mean sea level pressure (May) and Equatorial Indian Ocean mean sea level pressure (November of previous year). The model error of the model for July rainfall is 9%.
For forecasting of South-West monsoon season rainfall over the four broad geographical regions of India (NW India, Central India, South Peninsula and NE India), multiple regression (MR) models based on separate set of predictors are used. All the four multiple linear regression models have model errors of 8% of LPA.
In view of its great potential, IMD has implemented a dynamical prediction system at the National Climate Centre under a collaborative research project between India Meteorological Department and Indian Institute of Science, Bangalore. The Seasonal Forecast Model of the Experimental Climate Prediction Center (ECPC) was adopted for this purpose. 20 year model climatology (1985-2004) was prepared by prescribing observed sea surface temperature (SST) data as boundary conditions. The validation of model hindcast suggests promising skill (Details here) over the Indian region. At present, this model is used for preparing experimental long range forecasts for the southwest monsoon rainfall over India.
National Climate Centre also prepares an extended range forecast for the onset of southwest monsoon rainfall over Kerala . This forecast was first issued in 2005. The forecast is issued based on indigenously developed statistical model with 6 predictors (Details here). In 2005, the forecast indicated that monsoon onset over Kerala would be on 7th June. Actual onset over Kerala was on 5th June. In 2006, the forecast was 30th May, whereas the actual onset was on 26 May.
In addition, IMD prepares operational long range forecasts for the Winter Precipitation (Jan to March) over Northwest India and Northeast Monsoon rainfall (October to December) over South Peninsula. For this purpose, separate statistical models have been developed.