Relevance of soil moisture for seasonal atmospheric predictions: is it an initial value problem?

Ensembles of boreal summer atmospheric simulations, spanning a 15-year period (1979–1993), are performed with the ARPEGE climate model to investigate the influence of soil moisture on climate variability and potential predictability. All experiments are forced with observed monthly mean sea surface temperatures. In addition to a control experiment with interactive soil moisture boundary conditions, two sensitivity experiments are performed. In the first, the interannual variability of the deep soil moisture is removed during the whole season, through a relaxation toward the monthly mean model climatology. In the second, only the variability of the initial soil moisture conditions is suppressed. While it is shown that soil moisture strongly contributes to the climate variability simulated in the control experiment, an analysis of variance indicates that soil moisture does not represent a significant source of predictability in most continental areas. The main exception is the North American continent, where climate predictability is clearly reduced through the use of climatological initial conditions. Using climatological soil moisture boundary conditions does not lead to strong and homogeneous impacts on potential predictability, thereby suggesting that the climate signals driven by the sea surface temperature variability are not generally amplified by interactive soil moisture and that the relevance of soil moisture for seasonal forecasting is mainly an initial value problem.