In the realm of climate change and water scarcity, accuracy of hydrological modeling becomes more important as it helps to understand how precipitation over a watershed turns into runoff in time and space. In addition to numerical models, complicated hydrological processes can be tackled by data-driven methods. This study explores how different Machine Learning (ML) algorithms perform over three different living labs in the TRANSCEND project. The results obviously demonstrate that Graph Neural Networks, which can consider how sub-basins are connected within a watershed, outperforms traditional ML models in terms of Kling–Gupta efficiency. While further studies are recommended to provide a broader perspective, the high performance of ML models highlights their potential in climate change impact assessment studies.
