Machine learning and satellite data aim to give flood forecasters an earlier warning

On a June morning in rural Indiana, Laura Lin was caught off guard by a deluge that dropped over 8 inches of rain in just hours. By the time flood warnings arrived, the water was already overwhelming her neighborhood, and she had nowhere safe to go. For countless people in flood-prone areas, this race against rapidly rising water remains a deadly reality.

The stakes of fast-moving water#

Flash floods rank among the most dangerous weather events in the United States and globally. The physics are unforgiving: just 6 inches of swift-flowing water can knock down an adult, while 12 inches can lift a car. As the climate warms, extreme rainfall events are becoming more frequent, amplifying the risk. A recent catastrophic flood in Nepal caused by a glacier collapse underscored how devastating water can be to entire communities.

The National Weather Service (NWS) responds to this threat by issuing three types of alerts—watches (possible flooding within 12–48 hours), advisories (nuisance-level flooding), and warnings (life-threatening danger). The distinction between a regular flood and a flash flood hinges on timing: if an area floods in under six hours, it's classified as a flash flood. This narrow window leaves little room for error.

How forecasters currently track threats#

Weather offices across the country rely on a patchwork of tools to spot incoming floods. Rain gauges and stream monitors report water levels in real time, while satellites and radar track developing storms. Meteorologists combine this data with knowledge of local soil types, terrain steepness, and flash flood guidance thresholds to decide whether to issue an alert. Yet the system has blind spots: satellite imagery is less detailed over land than oceans, and rural areas often lack sufficient rain gauges. Once rain is already falling, forecasters have lost precious lead time.

TACLS: a new layer of detection#

Researchers at UC San Diego, the NWS, and NASA developed Transient Artifact and Continuous Learning System (TACLS) to address these gaps. The technology leverages the Global Navigation Satellite System (GNSS)—a network normally used for earthquake prediction—to measure atmospheric moisture.

The key insight: the amount of water vapor suspended in the atmosphere, or "precipitable water," reflects conditions that precede dangerous storms. When moisture levels are high, radio signals between GNSS satellites and ground sensors experience measurable delays. TACLS compares this real-time moisture data to weather forecasts, helping forecasters verify whether storms are actually moving as predicted and whether conditions favor flash flooding.