Evaluating satellite-derived soil moisture data for improving the internal consistency of process-based ecohydrological modelling

Abstract

Soil moisture is a key variable controlling the energy and water balance at the land surface and satellite-derived soil moisture (SDSM) products can therefore be helpful to constrain parameters of (eco)hydrological models. Due to coarse resolution of many SDSM products (>25 km), previous studies mostly focused on large catchments (>500 km²). However, recent developments in soil moisture remote sensing resulted in SDSM products with improved spatio-temporal resolution, which are potentially better suited for model calibration in small to meso-scale catchments. This study therefore aims at assessing the value of three recent SDSM products (SMAPL3E, SCATSAR and ASCAT DIREX SWI) with high spatio-temporal resolution (spatial sampling 0.5–9 km; temporal resolution 1–3 d) for the calibration of the process-based ecohydrological model EcH2O in a 66-km² catchment in NE Germany.

Satellite-derived soil water index (SWI) data agreed well with in-situ soil moisture observations (Pearson correlation coefficient: 0.67–0.86), with a slightly better performance for SMAPL3E than for the other two products (SCATSAR, ASCAT DIREX SWI). Calibrating the ecohydrological model EcH₂O to SWI data based on SMAPL3E improved the dynamics of simulated soil moisture (increase of Pearson correlation coefficient by 0.03), while model performance for streamflow slightly deteriorated (decrease of Nash-Sutcliffe efficiency by 0.02). Including volumetric soil moisture data based on in-situ data or SDSM scaled to in-situ data in the model calibration was necessary to further improve the model with respect to absolute soil moisture levels (increase of Nash-Sutcliffe efficiency for scaled soil moisture from 0.23–0.25 to 0.56–0.59). Comparing spatial patterns of simulated soil moisture and SDSM revealed shortcomings of the simulated and the SDSM data. Simulated soil moisture spatial patterns are influenced by artefacts of the interpolated precipitation patterns, while spatial patterns of SDSM seem too strongly damped. Based on this study, the following recommendations may be derived for practitioners who consider including soil moisture data for model calibration: 1) integrating SDSM in model calibration is beneficial for improving model internal consistency; 2) where available, in-situ soil moisture data should be included in order to better constrain simulated absolute soil moisture levels; 3) at the scale of small to meso-scale catchments (∼100 km²), the temporal dynamics of the SDSM products evaluated in this study are likely more helpful for model calibration than their spatial patterns.