Distinct hydrologic response patterns and trends worldwide revealed by physics-embedded learning
Published in Nature Communications, 2025
Recommended citation: Ji, H., Song, Y., Bindas, T., Shen, C., Yang, Y., Pan, M., Liu, J., Rahmani, F., Abbas, A., Beck, H., Lawson, K., & Wada, Y. (2025). Distinct hydrologic response patterns and trends worldwide revealed by physics-embedded learning. Nature Communications, 16(1), 9169. https://doi.org/10.1038/s41467-025-64367-1
Rivers worldwide have quietly changed how much rain they pass on as flow rather than return to the air and soil — in some places by more than 20% over twenty years. The models used to forecast floods and droughts had treated that behaviour as fixed, largely because they could not learn from the growing volume of observations available to them. The model built here embeds the physics of the water cycle inside a learning system trained on very large datasets, and tracks observed daily and monthly flows substantially more closely than the systems now in operational use. Mapping the shifts globally ties them to rising flood risk across northern mid-latitudes, deepening water stress in southern subtropical regions, and a measurable decline in the freshwater reaching many European estuaries.