The relative influence of soil moisture and SST in climate predictability explored within ensembles of AMIP type experiments

Three ensembles of AMIP-type simulations using the Arpege-climat coupled land–atmosphere model have been designed to assess the relative influence of SST and soil moisture (SM) on climate variability and predictability. The study takes advantage of the GSWP2 land surface reanalysis covering the 1986–1995 period. The GSWP2 forcings have been used to derive a global SM climatology that is fully consistent with the model used in this study. One ensemble of ten simulations has been forced by climatological SST and the simulated SM is relaxed toward the GSWP2 reanalysis. Another ensemble has been forced by observed SST and SM is evolving freely. The last ensemble combines the observed SST forcing and the relaxation toward GSWP2. Two complementary aspects of the predictability have been explored, the potential predictability (analysis of variance) and the effective predictability (skill score). An analysis of variance has revealed the effects of the SST and SM boundary forcings on the variability and potential predictability of near-surface temperature, precipitation and surface evaporation. While in the tropics SST anomalies clearly maintain a potentially predictable variability throughout the annual cycle, in the mid-latitudes the SST forced variability is only dominant in winter and SM plays a leading role in summer. In a similar fashion, the annual cycle of the hindcast skill (evaluated as the anomalous correlation coefficient of the three ensemble means with respect to the “observations”) indicates that the SST forcing is the dominant contributor over the tropical continents and in the winter mid-latitudes but that SM is supporting a significant part of the skill in the summer mid-latitudes. Focusing on boreal summer, we have then investigated different aspects of the SM and SST contribution to climate variations in terms of spatial distribution and time-evolution. Our experiments suggest that SM is potentially an additional source of climate predictability. A realistic initialization of SM and a proper representation of the land–atmosphere feedbacks seem necessary to improve state-of-the-art dynamical seasonal predictions, but will be actually efficient only in the areas where SM anomalies are themselves predictable at the monthly to seasonal timescale (since remote effects of SM are probably much more limited than SST teleconnections).