Hydrological Model Performance Assessment across several Moroccan Catchments: Investigating the Effect of Model Attributes, Catchment Features, and Precipitation Inputs

Hydrology research can benefit significantly from large-sample hydrology studies by offering the possibility for better hydrological models assessment and by providing a suitable ground for identifying catchment characteristics that influence model performance. In our study, we conducted a performance assessment of eight monthly lumped rainfall-runoff models (GR2M, XM, WM, VUB, abcd, DWBM, GR5M, and Wapaba) in 30 Moroccan catchments, forced by rainfall data from 34 rain gauges. During the study period 1983-2019, we investigated the relationship between model performance (quantified with KGE) and both model complexity and structural attributes. Furthermore, we conducted correlation analysis to explore possible connections between this performance and catchment features (more than 180 features were considered), and we additionally examined how the models respond to three precipitation input data, namely ERA5, CHIRPS, and PERSIANN-CDR. Our findings revealed that no hydrological model was the best (or the worst) across the entire set of catchments. The model performance was found to be more influenced by model structure than by its degree of complexity, and more by hydro-climatic characteristics, particularly those related to calibration and calibration relative to validation, than by non-hydro-climatic factors. Among the investigated features, the Pearson correlation between observed rainfall and runoff was the strongest characteristic influencing model performance. Furthermore, this study (i) emphasized the essential role of rainfall and runoff data richness, in terms of wet and dry years, in enhancing model performance even if the calibration data is only relatively richer than the validation data and (ii) showed that dry periods are more beneficial to model performance than wet ones. Finally, our study revealed a consistent pattern in the models responses to the different rainfall forcings; with ERA5 consistently yielding the best model performance and PERSIANN-CDR consistently resulting in underperformance. This consistent behavior of the models was best explained by the linearity between the employed rainfall products and the catchments' observed runoff.