Ensemble spread and systematic errors in the medium-range predictions during the Indian summer monsoon

Abstract

For preparing medium range weather forecasts, two global coarse resolution models at different resolutions were used at the National Centre for Medium Range Weather Forecasting (NCMRWF), India. In order to improve the forecasting skill, an ensemble prediction system (EPS) was implemented on experimental basis. For generating initial perturbations a breeding method was implemented. Experimental forecast runs with 8-member ensemble were carried out and results are analyzed for a monsoon season. The ensemble mean of rainfall forecasts shows that over the broad region of Gangetic Plains, the EPS brings out the monsoon activity (active and weak spell) reasonably well six days in advance. However, over the eastern parts of India, the ensemble mean rainfall is good only in short-range. The ensemble spread becomes quite large from about day-4 forecast and beyond. An examination of the rainfall pattern from day-1 to day-6 forecasts by the model and the ensemble spread shows there is no linearity in the increase of spread with the rainfall amount. The model has a systematic tendency to enhance rainfall activity over the central Bay of Bengal and eastern parts of India as the length of forecast is increased. At the same time, the model tends to dry up over the equatorial Indian Ocean region, however, in the high-resolution model, the same tendency is not seen. In circulation fields, the model also has large systematic errors. These results suggest that to obtain maximum benefit from the ensemble prediction system, the systematic biases in the model must be reduced as the breeding method only takes care of the uncertainties in the initial conditions.