The role of initial conditions and forcing uncertainties in seasonal hydrologic forecasting

A series of hydrologic forecasts with lead times up to 6 months are conducted to investigate the relative contributions of atmospheric forcing and hydrologic initial conditions (IC) to the overall errors in hydrologic forecasting during cold and warm seasons. These experiments are known as the ensemble streamflow prediction (ESP) and the reverse-ESP (R-ESP). Analysis of these hindcasts suggests that IC uncertainties outweigh forcing uncertainties thus dominating forecast errors in a short lead time up to about 1 month; at longer lead times, forcing uncertainties become a more important contributor. Further investigation shows that forecast errors at short lead times due to uncertain ICs are mainly determined by the prescribed IC variability, while the evolution of forecast errors due to imperfect atmospheric forcings mainly corresponds to the interannual variability of precipitation. With respect to difference in forecasts initialized in winter and summer times, ICs tend to have longer impacts on warm season forecasts than on cold season ones, due mainly to drier initial moisture state in the summer time. As far as the basin size is concerned, we find that the larger the basin, the stronger the impacts from ICs at short lead times. Small basins are more sensitive to forcing fields. Regardless of basin size, forcing uncertainties dominate relative forecast errors for long lead times. In order to see whether statistically downscaled forcing fields from dynamic climate models are more skillful than traditional ESP, we conducted additional ESP-type experiments using the statistically downscaled climate forecast system (CFS) fields to drive the hydrological model. In comparison to traditional ESP, the IC errors show a larger impact on the forecasts when forced by the CFS fields, which suggests that the latter contains more skill than the traditional ESP approach.