Quantifying conceptual model structure uncertainty and assessing relative model performance: findings from a 36-model, 559-catchment comparison study

The choice of hydrological model structure, i.e. a model's selection of states and fluxes and the equations used to describe them, has an important impact on model performance and realism. This work investigates differences in performance of 36 lumped conceptual model structures, calibrated to, and evaluated on, 10 years of daily streamflow data in 559 catchments across the United States. We use three objective functions and find that (i) reasonable model performance can be achieved in nearly all catchments, but that no single model is capable of this. Instead, nearly every model can be one of the best choices or one of the worst, depending on the catchments used. (ii) In most catchments, several models achieve performance very similar to that of the best model for that catchment, resulting in high levels of model structure uncertainty. (iii) We find no relation between the number of model parameters and performance during either calibration or evaluation periods, nor evidence of increased risk of overfitting for models with more parameters. (iv) Correlation analysis shows that the strongest relation exists between model performance and streamflow signatures. The relationship between catchment attributes such as climate, geology, topography, soil type and vegetation remains elusive. (v) Results suggest that certain model structures are inherently better suited for certain objective functions, and that models that share certain structural elements tend to have similar performance across the catchment sample as well.