Model parameter transferability under a sensitivity analysis clustering approach

Abstract:

In seeking to identify connections between the structural features of catchments and their hydrologic functioning, several metrics and correlations have been studied based on catchment physical characteristics, hydrologic signatures, and climatic features under the expectation that such catchment attributes should be useful for transferring model parameters from gauged to ungauged sites. However, similarity may not adequately satisfy usual assumptions of classification/regionalization approaches leading to poor model appropriateness, even when transferring well-calibrated hydrologic parameters among basins deemed similar by regular classification procedures. In this study, we evaluate the relationship between physical catchment characteristics, hydrologic signatures, and hydrologic models for a data set composed of 92 catchments located in three distinct climate zones in the USA. We propose a two-level hierarchical clustering approach with the first level based on the sensitivity of the hydrologic model parameters and the second level based on the model parameters. Model performance, in terms of parameter transferability, is compared between the hierarchical clustering approach and four reference scenarios in which clusters are based on climatic region, physical descriptors, hydrologic signatures, and an unclassified control. A transferability metric is developed to compare the performance of parameter transfer according to each study scenario. The proposed clustering approach based on the similarity of the optimized model parameter set demonstrated slight improvements in model performance gains when doing parameter transfers using the proposed strategy.