HydroForecast

HydroForecast is a streamflow forecasting service from Upstream Tech that predicts inflows into dams and rivers for hydropower producers, energy traders, water utilities, mining operations and government agencies. Operators scheduling generation, managing reservoirs and tailings, or preparing for floods need to know how much water is coming; traditional basin-by-basin physical models can take years to deploy and struggle with extreme events and ungauged basins.

HydroForecast uses a theory-guided machine-learning model (an LSTM neural network) trained across hundreds of basins and then tuned to each customer's sites. Its inputs combine several weather forecasts (NOAA GEFS, ECMWF HRES, and GFS and HRRR over North America), daily NASA VIIRS satellite observations of snow cover, vegetation (NDVI) and land surface temperature, static geospatial data, and in-situ gauge observations that are assimilated as they arrive. It issues probabilistic forecasts hourly out to 10 days (reissued every two hours), daily out to 90 days and up to one year ahead, and can reconstruct historical flow records at ungauged locations.

Forecasts are delivered through a web dashboard with configurable alerts, a REST Time Series API, a Delft-FEWS integration and a read-only MCP connector. Upstream Tech's hydrologists set up and tune each deployment; standard offerings go live in under 90 days. Named users include Sonoma Water, Brookfield Renewable, Tacoma Power, Ontario Power Generation and the U.S. Bureau of Reclamation. HydroForecast won a yearlong forecasting competition hosted by CEATI, the U.S. Bureau of Reclamation and North American hydropower utilities, outperforming in 23 of 25 categories, and in 2026 Upstream Tech signed a CRADA with NOAA's Office of Water Prediction.