Gauch, M., Mai, J., Gharari, S., & Lin, J. (2019a). Data-driven vs. physically-based streamflow prediction models. In Proceedings of 9th International Workshop on Climate Informatics. https://cs.uwaterloo.ca/~jimmylin/publications/Gauch_etal_CI2019.pdf

Climate change leads to more frequent and severe floods and droughts. Precise water flow forecasts for rivers and streams help mitigate damage and are direly needed. We evaluate physically-based and data- driven models on the task of streamflow prediction in the Lake Erie region: Physically-based models capture simplified representations of the physical processes that underlie streamflow, while purely data-driven models encode no such knowledge explicitly. Experiments show that data-driven approaches can provide more accurate predictions than a physically-based model, suggesting potential in hybrid approaches that combine hydrological understanding with high prediction accuracy.

https://cs.uwaterloo.ca/~jimmylin/publications/Gauch_etal_CI2019.pdf