Use of multiple modules and Bayesian Model Averaging to assess structural uncertainty of catchment-scale wetland modeling in a Coastal Plain landscape

Abstract

Wetland hydrology differs by landscapes and thus a modeling structure that best characterize wetland hydrological processes may vary. However, structural uncertainty of wetland modeling has been assessed in few landscapes. Limited understanding of structural uncertainty remains for a landscape dominated by groundwater. This study used Soil and Water Assessment Tool (SWAT) to assess structural uncertainty in catchment-scale wetland hydrology on the Coastal Plain of the Chesapeake Bay watershed, where landscape hydrology is greatly affected by groundwater. This study investigated the differences in streamflow prediction and quantification of wetland loss impacts on streamflow among three wetland modeling structures. Two lumped structures (SS and RS) and one semi-spatial structure (RG) were examined in this study. Bayesian Model Averaging (BMA) was used to generate the ensemble of three model structures. To reduce parameter uncertainty on model outputs, streamflow parameters were constrained by using the same streamflow parameter set that was mutually acceptable for the three wetland modeling structures. Results showed that the three model structures exhibited different streamflow predictions and hydrologic changes by wetland loss. The RG structure provided the best streamflow simulation results due to its accurate prediction on low-flow conditions relative to the two other structures (SS and RS). The BMA results also indicated less accurate model prediction of streamflow compared to the RG. Hydrologic changes resulting from wetland loss were best captured by the RG structure that mimicked wetland water storage functions such as peak flow changes and annual streamflow variation. In contrast, the results from the two other structures and BMA did not accurately reflect those functions in responses to wetlands loss. The RG structure showed superior simulation of wetland hydrology due to its semi-spatial model structure. In contrast, poor BMA results were attributed to poor predictions from lumped model structures (SS and RS). These results suggest that a spatial model structure is critical to simulate wetland hydrologic processes in a landscape with a high dependency on groundwater processes, and wetland modeling development should be toward an explicit spatial characterization of wetlands.