Application-based evaluation of multi-basin hydrological models

Hydrological models are generally calibrated and validated using a suite of well-known statistical metrics, which sometimes lack clear connection and tailoring to the local users’ need and therefore limits the evaluation, especially in the case of global climate services. Therefore, in this study, two types of application-based evaluation metrics are introduced, addressing (i) temporal matching of quantile extremes and (ii) relative bias in flow signatures, which supplements commonly used model performance assessment metrics. The introduced metrics are compared to conventional statistical metrics, at seven case study areas across the world, with three model settings representing different datasets and calibrations, generated from the global hydrological model World-Wide HYPE (WW HYPE). The results suggest that different performance results can occur when comparing application-based metrics to conventional ones. This implies that different evaluation metrics reveal models’ capability in various aspects, supporting their application under the corresponding circumstances. Finally, these metrics enable us to propose two model applicability scenarios: generally applicable models and conditionally applicable models. For instance, the WW HYPE with global dataset and local calibration can yield optimal estimates concerning the timing of quantile extremes and temporal variations in flow signatures, despite its suboptimal performance in conventional evaluation metrics. Therefore, it may be considered as a conditionally applicable model which can be used in regions with limited local datasets, supplying reliable information for decision-makers in formulating strategic plans for water resources management.