A new forecasting tool uses satellite signals and machine learning to help US meteorologists identify conditions that could turn heavy rain into a flash flood. The Transient Artifact and Continuous Learning System, or TACLS, is already operating at National Weather Service offices in Los Angeles and San Diego, with a national rollout planned for the second half of October.
TACLS analyzes delays in signals between navigation satellites and ground stations. Atmospheric water vapor slows those signals, giving forecasters a real-time measurement of moisture building before and during a storm. A machine-learning model trained on years of those readings, atmospheric-river data, rainfall and past warnings looks for patterns associated with flash floods. Nearby stations are compared to suppress isolated sensor errors that could create false alarms.
The system is intended to assist, not replace, the specialists who decide whether to issue alerts. It adds live observations to forecasts that may be wrong and can provide information before rainfall gauges show the full event. Coverage will initially be strongest in the western United States, where many ground sensors were installed for earthquake research. Scientists from UC San Diego, the National Weather Service and NASA developed the project. Its value will depend on how reliably it avoids false positives while giving forecasters enough extra time to act.