The spatial scale of model errors and assimilated retrievals in a terrestrial water storage assimilation system

Abstract

Synthetic satellite observations (or retrievals) of terrestrial water storage (TWS) in the Mackenzie River basin located in northwestern Canada were assimilated into the Catchment land surface model to evaluate the impact (i) assimilating TWS retrievals at subbasin (∼105 km2) or basin (∼106 km2) scales and (ii) incorrectly specifying the model error correlation length that is used for the perturbation of model forcing and prognostic variables in the ensemble‐based assimilation system. Specifically, a total of 16 experiments were conducted over a 9 year study period using different combinations of the spatial scale of the assimilated TWS retrievals and the horizontal model error correlation length. In general, assimilation of the TWS retrievals at the subbasin scale (∼2.7 × 105 km2 on average) yielded the best agreement relative to the synthetic truth. Greater improvement in TWS and snow water equivalent, in general, was witnessed as the (designed) horizontal model error correlation length increased. Conversely, subsurface soil water, evaporation, and runoff estimates typically improved (or remained unchanged) as the horizontal model error correlation length decreased. As the scale of the assimilated TWS retrieval decreased, more mass was effectively transferred from snow water equivalent into the subsurface, thereby dampening the hydrologic runoff response in the study area and correcting for improper model physics related to the runoff routing scheme. In general, TWS retrievals should be assimilated at the smallest spatial scale for which the observation errors can be considered uncorrelated while the specification of the horizontal error correlation length scale is of secondary importance.