Headwaters Hydrology Project (HHP) ML-Based Streamflow Estimates

Accurate streamflow estimates in ungaged basins are essential for hydrologic research, water management, and drought/flood monitoring. The Headwaters Hydrology Project (HHP) addresses limitations in existing national-scale models by providing daily, machine learning based streamflow predictions for all Hydrologic Unit Code 10 (HUC-10) watersheds across the contiguous United States. Cross-validation in pseudo-ungaged basins shows that the HHP significantly outperforms the National Hydrologic Model and the National Water Model across multiple performance metrics, including Nash–Sutcliffe Efficiency, Kling–Gupta Efficiency, correlation, and bias. Importantly, the HHP is significantly more accurate in detecting hydrologic extremes such as drought and flood, improving detection at 91-96% of the >1,500 sites tested. The dataset spans 1982-present, updates daily with low latency, and is publicly accessible via an interactive mapping tool and application programming interface. The HHP demonstrates the value of machine learning as a complementary tool to traditional hydrologic models, enabling scalable, real-time insights for monitoring and decision-making.