The U.S. Geological Survey (USGS) has launched the River DroughtCast, a new tool that employs machine learning to predict streamflow drought conditions across the contiguous United States. Set to operate from March 17, 2026, this tool utilizes over a century's worth of streamflow data collected from more than 3,000 USGS streamgage locations, some of which have been active for over 100 years.
River DroughtCast forecasts are capable of predicting streamflow conditions up to 13 weeks in advance, with the highest accuracy rates occurring within the first four to six weeks. The tool can accurately identify the onset of severe or extreme drought conditions around 75% of the time during the initial week, though this reliability diminishes to roughly 55% by the 13th week. Each forecast also provides confidence estimates, aiding users in understanding the reliability of the predictions.
John Hammond, the USGS project manager for the system, highlighted the significance of integrating historical data with advanced technology to enhance water resource management. The tool is designed to assist farmers and municipal water managers in making informed decisions regarding irrigation and conservation in anticipation of water shortages.