Upstream Tech has released a major update to its HydroForecast Seasonal streamflow forecasting tool, unveiling a new model architecture and launching the HydroForecast Seasonal-3 system.
The tool, used by water managers, utilities, and hydropower operators, provides long-range streamflow forecasts to help manage water supply and risk. According to the company, the new model architecture enhances the accuracy, flexibility, and scalability of these forecasts.
“Our team of PhD hydrologists, machine learning experts, and software engineers is always working to push the field of hydrology forward,” said Marshall Moutenot, co-founder of Upstream Tech. “We’ve built a new end-to-end, foundational AI hydrology architecture that connects global observations with a wide range of meteorological forecasts to produce probabilistic forecasts with flexible timesteps and horizons.”
The upgraded architecture increases forecast performance through the integration of global data sources, satellite observations, and multiple ensemble weather forecasts. It also allows for faster deployment of new forecasts due to improved scalability.
The newly released HydroForecast Seasonal-3 model is the first to use this upgraded system. It provides forecasts on a daily timestep with the ability to look up to one year ahead. The model also uses a distributed architecture to capture localized variability by ingesting data at the sub-basin level.
Other improvements include enhanced accuracy in cold-weather regions and region-specific insights based on training with Canadian stream gauge data.
“The new Seasonal-3 model brings advancements in accuracy and usability that will make it easier to manage risk and optimize operations with greater confidence and efficiency than ever before,” Moutenot said.
