As in nature, the soil moisture state produced by a land surface model (LSM) reflects the integrated impact of the antecedent meteorological forcing – the impact of the time history of all of the drivers (precipitation, radiation, wind speed, etc.) that can affect the wetness of the soil. If the forcing applied to an LSM is derived from observations, and if the LSM's physical formulations are reasonably realistic, then the LSM-produced soil moisture state can, in principle, serve as an index of the true water content of the soil – and, when the soil gets dry enough, as an index of agricultural drought. The obvious question to ask, then, is to what extent does such an index depend on LSM-specific physical parameterizations? If the sensitivity of the index to the chosen LSM is large, then the value of the index is significantly diminished.

We address this question through an analysis of LSM output generated in phase 2 of the Global Soil Wetness Project, an experiment in which a number of land surface models were driven with the same observations-based meteorological forcing over a period of ten years. The degree of independence in the LSM-produced soil moistures is found to vary spatially across the globe. It is generally small, however, particularly in areas with significant interannual rainfall variability. Indeed, when the soil moistures for each land model are expressed in terms of standard normal deviates relative to that model's seasonally varying climatology, the time series of the different model products are often largely coincident. We thus find that in most areas, the land models tend to produce mostly redundant information, suggesting that an LSM-based drought index in these areas can be considered fairly robust. Uncertainties associated with the inter-model differences that do exist can perhaps be addressed through multi-model averaging.