Identifiability of model parameters across processes, models and performance criteria

In selecting a suitable set of performance criteria for model calibration, modellers commonly rely on statistical performance metrics. However, most performance criteria cannot identify all types of parameters. This paper tests 19 statistical performance metrics and hydrological signatures that capture the full spectrum of hydrological conditions regarding their suitability to identify model parameters. Four hydrological models (Raven-GR4J, HBV, SWAT+, and mHM) are applied to six German catchments. For better model comparison, their parameters were classified into five process groups. The newly developed “identifiability quote index” then assesses parameter identifiability for each combination of parameter and performance criterion.The results show that a combination of contrasting statistical performance metrics and hydrological signatures focusing on high, mid, and low flows is needed. Hydrological signatures provide insights that statistical metrics cannot. With average ident_quote values above 0.5, evapotranspiration parameters are better identifiable than subsurface parameters, and capacity parameters better than flux parameters.