Author(s): Juan Farfan-Duran; Luis Cea

Linked Author(s):

Keywords: Regionalization; Ensemble Modeling; Hydrological Model; Ungauged Basin

Abstract: Hydrological modelling is a valuable and widely used tool in water resources management. However, the hydrological data is scarce in many catchments, hampering the calibration of the model parameters. One possible approach to address this problem is to transfer the model parameters from a gauged (donor) basin to the ungauged (receptor) basin. This procedure is referred to as parameter regionalization. The methods for regionalization can be classified in 3 groups: 1) Regression-based, 2) Similarity-based, and 3) Signature-based methods. In the regression-based methods, a hydrological model is calibrated in several gauged catchments, and different catchment descriptors (CD) are used to predict the value of the model parameters in the ungauged catchments. The physical similarity-based methods consists of using the most similar donor catchments to the receptor catchment in order to transfer a set of model parameters. The hydrological model is then run in the ungauged catchment using one or several sets of parameters from the donor catchments. The present work presents a methodology that combines regression-based and similarity-based techniques in three steps. First, the hydrological model parameters that best correlate with the CD were identified. A multivariate linear regression between the CDs and these parameters was established using the data of the donor basins. The regression was then used to predict the value of these parameters in the receptor catchment. Second, the Euclidean distance between the CDs of the receptor and donor basins was calculated in order to select the 15 most similar catchments to the receptor basin. All the model parameters (except those defined by linear regression in step 1) of the 15 most similar basins were directly transferred to the receptor basin. In this way, 15 potential parameter sets are assigned to the receptor catchment. Third, the hydrological model was run in the receptor basin using the 15 potential parameter sets defined in step 2. This produced 15 potential hydrographs were then weight-averaged to obtain the predicted outlet hydrograph in the receptor basin. In this last step, the weights given to each of the 15 hydrographs were determined by IDW, based on the distance between the CDs of the basins. The proposed methodology was tested in 24 watersheds located in the northwest of Spain using cross-validation. The regionalization method obtained NSE values ranged from 0.5 to 0.86 in 79% of the cases, while for the basins calibrated with observed data, the NSE ranged from 0.66 to 0.87. The determination coefficients ranged from 0.67 to 0.87 for the regionalization method in the 24 basins and from 0.68 to 0.87 for the calibration with observed data. Those cases in which the proposed method failed should be further studied in future work.

DOI: https://doi.org/10.3850/IAHR-39WC252171192022871

Year: 2022