Assimilation of Spatially Sparse In Situ Soil Moisture Networks into a Continuous Model Domain

Abstract

Growth in the availability of near‐real‐time soil moisture observations from ground‐based networks has spurred interest in the assimilation of these observations into land surface models via a two‐dimensional data assimilation system. However, the design of such systems is currently hampered by our ignorance concerning the spatial structure of error afflicting ground and model‐based soil moisture estimates. Here we apply newly developed triple collocation techniques to provide the spatial error information required to fully parameterize a two‐dimensional (2‐D) data assimilation system designed to assimilate spatially sparse observations acquired from existing ground‐based soil moisture networks into a spatially continuous Antecedent Precipitation Index (API) model for operational agricultural drought monitoring. Over the contiguous United States (CONUS), the posterior uncertainty of surface soil moisture estimates associated with this 2‐D system is compared to that obtained from the 1‐D assimilation of remote sensing retrievals to assess the value of ground‐based observations to constrain a surface soil moisture analysis. Results demonstrate that a fourfold increase in existing CONUS ground station density is needed for ground network observations to provide a level of skill comparable to that provided by existing satellite‐based surface soil moisture retrievals.