Machine Learning


Fang Research Group conducts advanced applied machine learning research in water resources. Machine learning provides a data-driven automatic pattern detection approach as an alternative to other data-driven approximation approaches to physical modeling. Many phenomena in water resources are non-linear, which is a fruitful environment for machine learning performance gains over existing approaches. Particular attention has been paid to hydrologic/hydraulic modeling. A machine learning based Flood Alert System (FAS) methodology has been developed by the group which accurately nowcasts stage using only rain and stage gauges as input, thus providing a quick (<1 second run time) and accurate (KGE > 0.90) approximation to the hydrologic/hydraulic processes.

The team has gained hard-won practical experience with a variety of models including, but not limited to, neural networks, deep learning, LSTM, decision tree, random forest, and XGBoost. In addition to practical experience, the team has a deep mathematical understanding of machine learning to ensure rigor in scientific results. A variety of related papers have been published or are in the publishing process, with future work (especially in model interpretability) expected.