- The River DroughtCast leverages over a century of streamflow records to give water managers and communities earlier warnings of potential shortages.
The U.S. Geological Survey (USGS) has unveiled a machine learning-based forecasting system capable of predicting streamflow drought conditions up to 90 days in advance across the contiguous United States. Known as the River DroughtCast, the tool is designed to bridge a critical gap between short-range weather forecasts and longer-term seasonal water supply outlooks, giving municipalities, farmers, and ecosystem managers additional lead time to respond.
Unlike meteorological drought, which centres on rainfall deficits, streamflow drought refers to sustained periods when rivers and streams fall below normal levels, a condition that can persist even after precipitation resumes. Contributing factors include soil moisture, snowpack, and groundwater dynamics, making this type of drought considerably harder to anticipate but highly consequential for water availability.
"The USGS is putting more than a century of streamflow data to work in a completely new way, using machine learning to predict streamflow drought weeks in advance," said John Hammond, USGS project manager for the drought forecasting system.
The model was trained on data from thousands of USGS streamgages, some of which hold over 100 years of continuous records. Users can select any forecast window from one to 13 weeks. Accuracy is highest in the first four to six weeks, with severe or extreme drought correctly predicted around 75% of the time in week one, declining to approximately 55% by week 13. Each forecast includes confidence estimates to help users interpret reliability across different timeframes.
Currently covering more than 3,000 streamgage locations with at least 40 years of data, the tool has practical applications across multiple sectors. Irrigated agriculture, municipal water supply, and recreation industries could all use early warnings to adapt operations proactively.
A future version is planned to extend coverage beyond gauged locations and improve overall forecast accuracy.
Drought: an ongoing concern across the U.S.
The relevance of tools like River DroughtCast is underscored by current conditions across the country. In California, the Department of Water Resources recorded the second lowest April snowpack on record at the start of this month, with statewide levels at just 18 per cent of average, a direct consequence of record-hot March temperatures that triggered snowmelt weeks ahead of schedule. As snowpack supplies roughly 30 per cent of California's annual water needs, its near-disappearance raises serious concerns about water availability heading into the dry season.
Meanwhile, Florida, typically one of the wettest U.S. states, is experiencing its most widespread and severe drought since 2012, according to U.S. Drought Monitor data. As of early April, nearly 80 per cent of the state faced extreme drought conditions, with shallow groundwater aquifers in northern and central regions at critically low levels, as mapped by NASA's GRACE-FO satellite mission. The dry spell has triggered water use restrictions, threatened crops already weakened by February freezes, and fuelled wildland fires across the state.
