Calibration of a Parsimonious Rainfall-Runoff Model: a Sensitivity Analysis from Local to Regional Scale

Using a 4.25-year calibration period and 9 sub-basins (7 to 166 km2) located in the Alzette river basin (Grand Duchy of Luxembourg), an analysis of relationships between optimal at-site parameters (OMP) of the conceptual HRM model and physical basin descriptors (PBD) was carried out in order to compare the model efficiency obtained for four regionalization procedures. The first procedure (P1) consisted in a spatial classification of basin response into ‘physical’ homogeneous clusters according to the OMP-PBD relationships. The second procedure (P2) is a regression-based approach which uses regional equations between OMP and PBD. The third procedure is a lumped regional procedure (P3) which estimates simultaneously a regional parameter set for all the basins. The last procedure is based on a spatial regional approach (P4) which used the semi-distributed version of the HRM model and fits simultaneously a regional parameter set for all the basins according to their geological heterogeneity. Significant correlation with some basin characteristics and noticeably, the permeability of geological formations and land uses (forest, grassland, cropland), could be found for two of the three free model parameters. The goodness-of-fit for the procedure P1 was slightly weaker than the calibration performs on each basin individually. Among the two procedures meaningful for transposition to ungauged basins, the spatial approach (P4) was close to the individual calibration procedure, and outperformed the regionalization of lumped parameters (P2), which was nearly as poor as the lumped regional model (P3). Although these results were obtained for calibration mode only, procedure P4, with few parameter values, should provide good predictions in validation mode.