The Headwaters Hydrology Project (HHP) is a machine learning (ML)-based streamflow
model that provides daily streamflow estimates at the HUC-10
scale across the contiguous United States. Trained on high-quality observed streamflow
data (USGS, MT DNRC), hydroclimatic variables, and basin characteristics, HHP delivers
seamless, natural streamflow simulations—excluding the effects of dams, diversions, and other
human water management influences.
Unlike traditional process-based hydrologic models, HHP leverages ML techniques to improve
streamflow predictions in ungaged basins and headwater watersheds, where
existing operational models often struggle. Benchmarking results show that HHP consistently
outperforms process-based model benchmarks in accuracy, achieving a median Nash-Sutcliffe
Efficiency (NSE) of 0.75, demonstrating its reliability for streamflow estimation.
HHP is updated daily, publicly available, and supports real-time hydrology, drought
assessment, and ecological applications via an open-access API. This dataset advances water
resource management and drought monitoring by providing high-resolution, data-driven
streamflow predictions for the scientific and operational communities.
Data Record Length: 1982–Present (updated daily, ~2-day latency)
Percentile Period of Record: 1995-2024 (most recent 30 years)
