Regional Calibration for a Distributed Catchment Model: an Application in North-Western Italy

One major challenge in large scale modeling is the estimation of spatially consistent distributed parameters, with a robust functional relationship to climate and landscape characteristics. We use here the newly developed PArameter Set Shuffling (PASS) approach, which is able to provide such regionally consistent parameter sets, for the calibration of the SALTO (SAme Like The Others) distributed hydrological model for about 80 catchments in North-Western Italy. The PASS method is a machine learning technique that does not require a priori assumptions on the relationship between model parameters and catchment descriptors. It instead derives these relationships from observed patterns of calibrated parameters and available catchment descriptors. The application demonstrates that the performance of the regionally calibrated distributed parameter sets is consistent with the one obtained locally, calibrating each catchment individually, implying robust results also for ungauged catchments in the area. To allow the reproducibility and repeatability of experiments, and to ease the application of the PASS approach to other case studies, an R package is under development which will be soon made available in GitHub.