Short to Medium Range Forecasting Skills of the GFS Model

Abstract

The reliable prediction of the South Asian monsoon rainfall and its variability is crucial for various hydrological applications and early warning systems. This study analyzes Global Fore- cast System (GFS) model generated high-resolution precipitation forecast over the South-Asian region during June-September 2012. This work delineates the error characteristics of the model over land and ocean; how forecast errors vary at different hours of the day; the skill of the model in active and break cycle and clustering of the precipitation events. This study shows that forecast errors are much larger over the land than over ocean. More- over, the rate of increase of errors with lead time is rapid over the oceans where observed precipitation shows large day-to-variability. This is possibly due to the one-way air-sea interac- tion in the atmosphere-only model used for forecasting. Furthermore, over ocean, for a smaller range of RMS error there was not much variation in RMS error growth with lead time but for a higher range of RMS error, there was a rapid growth. Over land for a lower and higher value of RMS error, there was no variation in error growth with lead time. It has been also shown that the model had poor forecasting skills in predicting very heavy (>30 mmdayð€€€1) precipitation over both land and ocean. The error decomposition analysis shows that error by pattern variation was contributing more than 90% of the total mean square error as compared to an error by mean di erence. This can be probably due to an error in daily and diurnal scale variation. On the daily scale, the transition of the occurrence of active and break phases was well captured by the model. However, the model had considerable difficulties in forecasting long intense break and heavy rainfall events. Diurnal cycle of precipitation in the model shows the phase error of about 6 hours over land. On the other hand, over oceans, there was no phase error in precipitation forecast. Moreover, there was a systematic bias over the ocean. This shift and bias in model forecasted phase result in large error over both land and ocean. Thus, efforts should be given to improve the phase and amplitude forecast of the diurnal cycle of precipitation from the model over the South Asian region.