New AI Tool Helps Weather Services Predict Flash Floods Faster

On a rainy June morning in rural Indiana, Laura Lin watched her barn float away while on a work call. By the time emergency alerts went out, her entire town was already underwater. “All the people in town were already flooded when they started sending out alerts,” Lin recalls. This gap between danger and warning is exactly what a new forecasting system aims to close.

Flash floods kill more people globally than any other weather event, and the U.S. sees them as the second-deadliest weather threat. Just 6 inches of fast-moving water can knock an adult down; 12 inches can lift a car. With climate change intensifying extreme rainfall events, the need for faster, more accurate flood predictions has become urgent.

How TACLS Changes the Game

Enter TACLS: the Transient Artifact and Continuous Learning System. Developed by scientists from UC San Diego, NASA, and the National Weather Service, this tech uses satellite data combined with machine learning to spot dangerous flood conditions before they fully develop.

The breakthrough lies in repurposing an existing satellite network. The Global Navigation Satellite System (GNSS), primarily used for earthquake prediction, actually provides valuable atmospheric data. When moisture accumulates in the air before a storm, it delays signals between satellites and ground sensors. By measuring this delay, forecasters can determine how much water vapor, or “precipitable water,” exists in the atmosphere in real time.

“It helps us to make better decisions in our warning decision process,” says Jayme Laber, senior service hydrologist for the NWS Weather Forecast Office in Oxnard, California. Previously, meteorologists relied on rainfall predictions that weren’t always accurate. TACLS gives them actual, live atmospheric data to compare against forecasts.

The Machine Learning Piece

But raw satellite data means nothing without interpretation. UC San Diego graduate student Bhavik Chandna spent a year developing the machine learning component using long short-term memory architecture, which excels at understanding patterns that change over time, like developing storms.

The system was trained on years of GNSS measurements, atmospheric river data, precipitation records, and historical flash flood warnings. It learned to recognize when atmospheric conditions shift from “rain” to “dangerous flash flood incoming.” The beauty is that it doesn’t replace human forecasters; it gives them another critical tool to make faster, smarter decisions.

False alarms do happen with machine learning, but TACLS has built-in safeguards. “If one station shows a strong signal but its neighbors do not, that detection can be suppressed,” Chandna explains. Real weather systems affect multiple locations simultaneously, so the algorithm filters out isolated anomalies.

Rolling Out Nationally

TACLS is already running in LA and San Diego weather offices, where flash flooding is common. A newer version with improved graphics is being deployed to all 122 National Weather Service offices nationwide in the second half of October.

“We learn what’s going on a little more every time we do a storm,” notes Ivory Small, science and operations officer at the NWS San Diego office. “It’s imperative that you do the research after, especially big events.” This commitment to continuous improvement suggests TACLS will only get smarter.

The system will likely see the heaviest use in the Western U.S., where most GNSS sensors are concentrated near earthquake zones. But the same approach could work globally anywhere sensors and local weather data exist.

For Laura Lin and others caught in flash floods, the question remains: can 15 extra minutes of warning actually save lives? Ivory Small believes it can. “Without TACLS, a storm could kill some folks. With TACLS, you can put out the warning and save some folks.”

Source: Infeeds.com

What if the technology that saves lives is already watching from space, just waiting for us to listen?