Impact of the number of donor catchments and the efficiency threshold on regionalization performance of hydrological models

Abstract

Over recent decades, hydrologists have proposed a variety of methods to predict discharge in ungauged catchments, and significant progress has been made in the field of hydrological model parameter regionalization. However, uncertainties from both hydrological models and regionalization methods make it a challenge to draw clear conclusions for some questions in regionalization (e.g., the best performing regionalization method, the optimal number of donor catchments, and the optimal efficiency threshold of donor catchments). In this study, for the first time, we made an attempt to address such questions in one paper through a comprehensive evaluation of model performance by using five regionalization methods with two weighting schemes, two averaging options, five efficiency thresholds, and four lumped hydrological models over a broad set of 3444 catchments under varying hydroclimatic conditions in North America. The results show that: (1) Spatial Proximity with the Inverse Distance Weighting method and the output average option (SPI-out) generally performs better than or comparable to other regionalization methods for different climate regions and hydrological models, while the global mean method performs the worst. (2) The rank of different regionalization methods is similar among different climatic regions and hydrological models. Compared to catchments with other climates, regionalization methods perform worst in the arid regions. (3) The selection of five donors is relatively efficient for distance/attributes-based regionalization approaches with the output averaging option disregarding the efficiency threshold. (4) The differences of median Kling-Gupta efficiency values among thresholds of “all”, 0.6 and 0.7 are no more than 0.05 for each regionalization method. However, the regionalization performance significantly deteriorates from using efficiency thresholds of 0.7 to 0.9 due to the significant reduction of available donor catchments. Thus, the poorly calibrated catchments may need to be included in the regionalization process, especially when the number of catchments is insufficient.