Hydrogeophysical inversion using a physics-based catchment model with hydrological and electromagnetic induction data

Abstract

Physics-based catchment models of mountain environments can suffer from equifinality when solely calibrated against streamflow data. Inclusion of intra-catchment data such as soil moisture or groundwater levels in model calibration can reduce equifinality problems, though physical demands of installation and remote field sites can limit their availability. Non-invasive geophysical surveys such as electromagnetic (EM) induction have become practical alternative sources of information on the subsurface. As such, we are interested in addressing the applicability of EM data to directly calibrate hydraulic parameters in physics-based catchment models in hydrogeophysical inversions. This study explores the interrelationships between calibration data, hydraulic parameters, and calibrated model dynamics for a headwater catchment in the Reynolds Creek Experimental Watershed, Idaho, USA. Five calibration scenarios and a global sensitivity analysis are performed to quantify the ability of different combinations of hydrological (streamflow, groundwater levels, soil moisture) and EM data (airborne and ground-based surveys) to predict both streamflow and intra-catchment dynamics. Results indicate that calibrating against streamflow data alone yields accurate streamflow but inconsistent intra-catchment predictions (streamflow, groundwater level, and soil moisture average Kling-Gupta efficiency values of $KGE$ = 0.89, −0.53, and 0.44, respectively). Calibrating against all hydrological data yields reasonable predictions of hydrological dynamics (streamflow, groundwater level, and soil moisture average $KGE$ = 0.91, 0.23, and 0.62, respectively), though some calibrated parameter values do not match expectations from literature values. Reasonably accurate hydrological predictions were obtained when including EM data with either streamflow data alone (streamflow, groundwater level, and soil moisture average $KGE$ = 0.83, 0.29, and 0.52, respectively) or all hydrological data (streamflow, groundwater level, and soil moisture average $KGE$ = 0.87, 0.39, and 0.51, respectively) during calibration. However, EM data alone yields hydraulic parameters that overpredict saturation throughout the catchment (streamflow, groundwater level, and soil moisture average $KGE$ = 0.09, −0.57, and 0.37, respectively). These results highlight potential advantages of collecting EM data in catchments with existing streamflow data but poor coverage of intra-catchment hydrological data sets. Additional work regarding petrophysical model parameterizations, objective function definitions, and data set weighting schemes is needed to ensure that the contribution of EM data to hydraulic parameter identification is maximized.