Page 85 - AWSAR_1.0
P. 85

 Predicting Indian Summer Monsoon Rainfall: Moving from Multiple to Single Forecast
how many days should be taken for computing weights. Further, the forecasts from even the best model can often go quite wrong due to the chaotic nature of the atmosphere. This trait of varying performance of models is ignored by the existing MME approaches, as they select all the participating models which degrade the overall performance thereby underestimating the performance of the better performing models.
The newly developed approach overcomes these limitations by dynamical selection of models based on their performance in the recent past, therefore, 45 days (chosen from the summer monsoon season, namely, June to September). The minimum forecast error was achieved by using 45 days, which became the basis for the selection of 45 days in order to weight the models. Unlike conventional MME approaches, the dynamical model selection approach makes a choice from the combination of only three better performing models from the five participating models. The selection is based on the higher correlation between the rainfall values forecasted by the models and observed values obtained from the IMD rain gauge rainfall product. Due to the varying skill of each model at different places and forecast hours, the model selection procedure is carried out at each region and for each forecast hour, hence, the three selected models change for every region as well as for each forecast hour.
To verify the method’s accuracy, forecasts were checked against IMD rainfall data from 2008 to 2013. Results show an improvement of around 6-10% for 24-120 hour forecasts using dynamical model selection approach compared to the existing approaches. The point worth noting was that the improvement increased with the forecast lead time. The newly developed method has shown sufficiently promising results for real time dissemination of the forecasts to the user community.
Due to the inherent chaotic nature of the atmosphere, the precise prediction of the accurate amount and position of rainfall still remains a challenge to the scientific community, but this uncertainty in nature is what attracts the scientists and researchers towards this research field. It reminds me of the words from Edward Lorenz, pioneer of the chaos theory that, “If atmospheric processes were constant, or strictly periodic, describing them mathematically would be easy. Weather forecasting would be easy, and meteorology would be boring.”
63





























































































   83   84   85   86   87