The Role of the Land Surface Background State in Climate Predictability

Abstract

Skill in ensemble-mean dynamical seasonal climate hindcasts with a coupled land?atmosphere model and specified observed sea surface temperature is compared to that for long multidecade integrations of the same model where the initial conditions are far removed from the seasons of validation. The evaluations are performed for surface temperature and compared among all seasons. Skill is found to be higher in the seasonal simulations than in the multidecadal integrations except during boreal winter. The higher skill is prominent even beyond the first month when the direct influence of the atmospheric initial state elevates model skill. Skill is generally found to be lowest during the winter season for the dynamical seasonal forecasts. This is in contrast to the multiyear integrations, which show some of the highest skill during winter?as high as the dynamical seasonal forecasts. The reason for the differences in skill during the nonwinter months is attributed to the severe climate drift in the long simulations, manifested through errors in downward fluxes of water and energy over land and evident in soil wetness. The drift presses the land surface to extreme dry or wet states over much of the globe, into a range where there is little sensitivity of evaporation to fluctuations in soil moisture. Thus, the land?atmosphere feedback is suppressed, which appears to lessen the model's ability to respond correctly over land to remote ocean temperature anomalies. During winter the land surface is largely decoupled from the atmosphere due to increased baroclinic activity in the land-dominated Northern Hemisphere, while at the same time tropical ocean anomalies have their strongest influence. This combination of effects neutralizes the negative impact of climate drift over land during that season and puts all of the climate simulations on an equal footing.