Toward improved identification of hydrological models: A diagnostic evaluation of the abcd monthly water balance model for the conterminous United States

Continental-scale water balance (WB) assessments are important for characterizing hydrologic systems and understanding regional-scale dynamics and for identifying hydroclimatic trends and systematic data biases. However, it is not clear whether existing models can reproduce the catchment dynamics observed in nature. Nor has our ability to evaluate model results kept pace with computational and data processing abilities. Consequently, methods for diagnostic model evaluation and improvement remain weak. There is a need for well-conceived, systematic strategies to guide model selection, establish data requirements, estimate parameters, and evaluate and track model performance. We examine these challenges in the context of monthly WB modeling for the conterminous United States by applying the abcd model to 764 catchments selected for their comprehensive coverage of hydrogeological conditions. By examining diagnostically relevant components of model error, we evaluate the details of its spatial variability across the continental United States. Model performance, parameters, and structures are found to be correlated with hydroclimatic variables. However, our results indicate a need for the conventional identification approach to be improved. Because they do not constrain models to reproduce important hydrological behaviors, reported values of NSE or r(2) performance can be misleading. Further, we must establish suitable model hypotheses with appropriate spatiotemporal scale for each hydroclimatic region. Until these issues are resolved, such models cannot reliably be used to infer the spatiotemporal dynamics of continental-scale water balance or to regionalize model structures and parameters to ungaged locations.