Regionalization in global hydrological models and its impact on runoff simulations: a case study using WaterGAP3 (v 1.0.0)

Abstract

Abstract. Valid simulation results from global hydrological models (GHMs), such as WaterGAP3, are essential for detecting hotspots or studying patterns in climate change impacts. However, the lack of worldwide monitoring data makes it challenging to adapt GHM parameters to enable such valid simulations globally. Therefore, regionalization is necessary to estimate parameters in ungauged basins. This study presents the results of regionalization methods for the first time applied to the GHM WaterGAP3. It aims to provide insights into (1) selecting a suitable regionalization method for a GHM and (2) evaluating its impact on runoff simulation. In this study, four new regionalization methods have been identified as appropriate for WaterGAP3. These methods span the full spectrum of methodologies, i.e., regression-based methods, physical similarity, and spatial proximity, using traditional and machine-learning-based approaches. Moreover, the methods differ in the descriptors used to achieve optimal results, although all utilize climatic and physiographic descriptors. This demonstrates (1) that different methods use descriptor sets with varying efficiency and (2) that combining climatic and physiographic descriptors is optimal for regionalizing worldwide basins. Additionally, our research indicates that regionalization leads to spatially and temporally varying uncertainty in ungauged regions. For example, regionalization highly affects southern South America, leading to high uncertainties in the flood simulation of the Río Deseado. The local impact of regionalization propagates through the water system, also affecting global estimates, as evidenced by a spread of 1500 km3 yr−1 across an ensemble of five regionalization methods in simulated global runoff to the ocean. This discrepancy is even more pronounced when using a regionalization method deemed unsuitable for WaterGAP3, resulting in a spread of 4208 km3 yr−1. This significant increase highlights the importance of carefully choosing regionalization methods. Further research is needed to enhance the predictor selection and the understanding of the robustness of the methods on a global scale.