Modeling soil moisture: a Project for Intercomparison of Land Surface Parameterization Schemes Phase 2(b)

In an intensive investigation of soil moisture simulation in land surface schemes, a number of numerical experiments was conducted with 14 representative schemes and the results compared with Hydrological and Atmospheric Pilot Experiment - Modelization du Bilan Hydrique (HAPEX-MOBILHY) data. The results show that soil moisture simulation in current land surface schemes varies considerably. After adjustment of land surface parameters, the disagreement in soil moisture for a 1.6-m soil layer remains around 100 mm. Correspondingly, the range of variation in predicted annual cumulative evaporation as well as total runoff plus drainage is around 250 mm (annual precipitation being 856 mm for HAPEX-MOBILHY). The partitioning of surface available energy into sensible and latent heat fluxes is closely coupled to the partition of precipitation into evaporation and runoff plus drainage. Although, on average, the range of variation in net radiation is about 8 W m^-2, that of both the latent and sensible heat fluxes is twice as large. These disagreements are related to different causes but attempts to establish the link between the outcome and the responsible mechanism has had only limited success to date because of the complex interactions embedded in the schemes. This study implies that different schemes achieve different equilibrium states when forced with prescribed atmospheric conditions and that the time period to reach these states differs among schemes; and even when soil moisture is fairly well simulated, the processes (particularly evaporation and runoff plus drainage) controlling the simulation differ among schemes and at different times of the year. These results suggest that prescription of land surface scheme physics may have to be a function of the type of predictions (short-term weather forecasting, mesoscale modeling or climate ensembles) required as well as the underlying scheme formulation and that scheme simulations must be validated for all components of the prediction.