Event‐Type‐Based Multi‐Dimensional Diagnostics of Process Limitations in Hydrological Models

Abstract Aggregated evaluation metrics and overlooked hydrological process variability in individual streamflow events hinder understanding of how well hydrological processes are encoded in models. This study introduces a novel event‐type‐based multi‐dimensional diagnostic framework to enhance model performance assessment and to identify process limitations. It evaluates the performance variation (in terms of timing and relative magnitude errors) for streamflow events of different types (e.g., snow‐related events, rainfall on dry or wet soils) and using explainable machine learning (XAI) analyzes the relative importance of three possible error drivers: event properties, model process limitations (i.e., model fluxes and states), and initial model errors. The effectiveness of the proposed framework is assessed through a case study of a conceptual hydrological model applied to 340 German catchments. In this case study, the rainfall events on dry soils have higher timing errors, while relative magnitude errors prevail for the snow‐related events. Across all event types, initial model errors before the streamflow event are the primary driver of both timing and magnitude errors. We also find that the hydrograph‐related event properties and the model fluxes representing land surface dynamics are also important for magnitude errors regardless of the event type. The proposed framework provides valuable insights into how and why model performance varies across different error dimensions and under different event conditions, making it a powerful tool for advancing hydrological research and practice.