Quantifying the predictive consequences of model parameter simplification in a <i>SWAT+</i> model coupled with a groundwater module (<i>GWFLOW</i>)

Watershed hydrology models, though often calibrated with limited flow data to acceptable statistical standards, may still fail to capture a wide range of system behaviour. To understand how the calibrated model represents system behaviour, we use linear and non-linear uncertainty analysis methods to evaluate parameter contributions to prediction uncertainties and the effectiveness of calibration in reducing these uncertainties. The same methods were applied to four U.S. watersheds, each representing systems with distinct hydrological characteristics. The results show that parameter contributions to uncertainty are both prediction-specific and model-dependent. Moreover, the transfer of information from calibration data to parameters, and ultimately to predictions, varies with the unique hydrological characteristics of each watershed and the interactions among parameters, observations, and predictions. The results provide insights into the capabilities and limitations of model calibration given the available data. This facilitates the refinement of the data assimilation process and guides further site characterisation efforts.