Do internal flow measurements improve the calibration of rainfall-runoff models?

This paper compares four calibration strategies for a daily semidistributed rainfall-runoff model. The model is applied over 187 French catchments where streamflow data are available at the catchment outlet and at internal gauging stations. In the benchmark calibration strategy, the model parameters were optimized against the outlet flow only, with internal points considered as ungauged. In the three multisite alternative strategies, the parameters were optimized against the flow at the outlet and at one internal gauge. On 53 catchments, a second interior gauge was used as an independent validation point. The four methods were compared for their ability to estimate flow at the two internal points and at the catchment outlet, in calibration and validation modes, and considering three performance metrics. The results in validation indicate that interior flow data provided limited improvement in model performance. When the performance was evaluated at the outlet point, multisite calibrations led to nearly identical performance as the single-site calibrations, regardless of the number of calibrated parameters. Unexpectedly, similar results were obtained for most performance statistics when the model was evaluated at interior points. A sensitivity analysis performed on streamflow data confirmed that this conclusion still holds in presence of errors in flow data. Last, the comparison between lumped and semidistributed parameterizations clearly favored the lumped schemes, which show more stable parameters and equivalent performance for the simulation at independent interior points. The finding from this study provides confidence in lumped parameterization schemes, even for predicting the flow at interior gauges in a catchment.