Precipitation and net radiation in an atmospheric general circulation model (AGCM) are typically biased relative to observations. As a result, the simulated evaporative regime of a region may be biased, with consequent negative effects on the AGCM's ability to translate initialized soil moisture anomalies into improved seasonal predictions. These potential problems are investigated through extensive offline analyses with the Mosaic land surface model (LSM). We first forced the LSM globally with a 0.5-degree, 6-hourly, 15-yr (1979-93) "observational" dataset, a dataset consisting of ECMWF reanalysis data corrected with observation-based precipitation and radiation. We then repeated the simulation after imposing a representative set of GCM climate biases onto the forcings - the observational forcings were scaled so that their mean seasonal cycles matched those simulated by the NSIPP-1 (NASA Global Modeling and Assimilation Office) AGCM over the same period (1979-93). Evaporative regime in both the "observational" and "GCM-biased" simulations is characterized, as in previous studies, by the slope of the relationship between evaporation efficiency (evaporation divided by net radiation) and soil moisture content.
The AGCM's climate biases do indeed lead to significant biases in evaporative regime in certain regions, with the expected impacts on soil moisture memory timescales. Furthermore, the offline simulations suggest that the biased forcing in the AGCM should lead to overestimates of feedback in certain parts of North America -- parts already identified in previous studies as having excessive feedback. The present study thus supports the notion that the reduction of climate biases in the AGCM will lead to an improved ability to translate soil moisture initialization into increased seasonal prediction skill.