Remotely Sensed Surface Water Storage Shows Distinct Patterns From SWAT-Simulated Data

Quantifying and projecting the downstream benefits of water stored in lakes and wetlands ( ${\text{SW}}_{\text{storage}}$ ) requires watershed hydrologic models, which often parameterize surface water storage in topographic depressions using static digital elevation model (DEM) data. Calibration and validation of modeled ${\text{SW}}_{\text{storage}}$ dynamics using external data sets is uncommon, particularly across major river basins, with model calibration typically focused on observed discharge. Here, we develop and assess a novel remote sensing-based (RS) ${\text{SW}}_{\text{storage}}$ data set (Sentinel-1 and Sentinel-2) for verifying simulated ${\text{SW}}_{\text{storage}}$ estimates from a Soil and Water Assessment Tool (SWAT) model of the Upper Mississippi River Basin (UMRB; ~440,000 km²). Our results suggest that static DEM-based parameterization as well as model calibration based solely on discharge do not adequately capture spatial and temporal ${\text{SW}}_{\text{storage}}$ dynamics in the UMRB. Mean ${\text{SW}}_{\text{storage}}$ as estimated by SWAT was 74% ± 122% (mean ± standard deviation) higher than RS ${\text{SW}}_{\text{storage}}$ , where SW storage in SWAT was underestimated in wetland-rich subbasins and overestimated in agricultural, tile-drained subbasins. Time series of SWAT ${\text{SW}}_{\text{storage}}$ and RS ${\text{SW}}_{\text{storage}}$ were positively correlated in only 38.8% of subbasins. As RS ${\text{SW}}_{\text{storage}}$ is also vulnerable to error, storage estimates were compared to bathymetric data in select small wetlands. While uncertainty remains in the conversion from extent to storage for RS ${\text{SW}}_{\text{storage}}$ , the method and data set presented here are a promising option for improved parameterization and calibration of ${\text{SW}}_{\text{storage}}$ processes in SWAT and other process-based hydrologic models. Further consideration of these storage processes can potentially improve the accuracy of simulated streamflow in wetland-rich model domains.