The systematic errors in the downward fluxes of precipitation and radiation over land from atmospheric model contributed to strong drift of soil wetness to extreme values in the climate model, reducing the sensitivity of the atmosphere to anomalies in initial soil wetness and thus potentially reducing the predictability of the climate system. We replaced the downward fluxes with observationally-based time series identical to those used to drive the land model in the uncoupled mode. The replacement of fluxes improves the temporal and spatial correlations with the offline results of the soil wetness and surface temperature. Evidence is also presented that a better simulation of interannual variability is produced after flux replacement.

However, flux replacement still cannot overcome the climate drift in a coupled land-atmosphere model. Therefore, we also pursue development of an empirical correction of land and atmospheric errors in a global prediction system to improve the numerical simulation and seasonal-to-interannual prediction of the land-atmosphere branch of the global water cycle, with the ultimate aim of application of empirical correction to further improve climate prediction capability. This procedure adds forcing terms to the prognostic equations of both the atmosphere and land models to render the tendencies more consistent with observations. The correction method can eliminate not only state-independent (“systematic”) errors, but also certain flow-dependent forecast errors. The new method is significantly reduced the climate drift and improves the seasonal prediction skill.