The potential role of the land surface state in improving predictions of seasonal climate is investigated with a coupled land-atmosphere climate model. Climate simulations for 18 boreal summer seasons (1982-1999) have been conducted with specified observed sea surface temperature (SST). The impact on prediction skill of initial land surface state (interannually varying versus climatological soil wetness) and the effect of errors in downward surface fluxes (precipitation, longwave and shortwave radiation) over land are investigated with a number of parallel experiments. Flux errors are addressed by replacing the downward fluxes with observed values in various combinations to ascertain the separate roles of water and energy flux errors on land surface state variables, upward water and energy fluxes from the land surface, and the important climate variables of precipitation and near-surface air temperature.

Strong systematic errors are found in the model, which are only mildly alleviated by the specification of realistic initial soil wetness. The model shows little skill in simulating seasonal anomalies of precipitation, but it does have some skill in simulating temperature variations. Replacement of the downward surface fluxes has a clear positive impact on systematic errors, suggesting that the land-atmosphere feedback is helping to exacerbate climate drift. Improvement in the simulation of year-to-year variations in climate is even more evident. This suggests that the land surface can communicate climate anomalies to the atmosphere, given proper meteorological forcing.

The evolution of signal-to-noise between ensembles with different soil wetness initialization reveals the properties of land surface memory in the model, and the changes in this evolution under flux replacement suggest pathways through which this memory may be reinforced. The changes in skill under flux replacement suggests the parameterizations within the atmospheric model that need critical attention, and raises the hope that a different coupling strategy (e.g., flux adjustment or anomaly coupling) between land and atmosphere may significantly improve climate prediction.