Critical Overview

The Transient Artifact and Continuous Learning System (TACLS), developed by the National Weather Service (NWS), University of California San Diego (UCSD), and NASA, has entered pilot phase for flash flood prediction. Currently limited to California, the system will expand to all 122 NWS forecast offices nationwide.

Core Facts:

• Technical stack: Global Navigation Satellite System (GNSS) moisture sensing + Long Short-Term Memory (LSTM) neural network
• Developers: UCSD (project lead Yehuda Bock), NASA, NWS (operations)
• Training data: Years of GNSS atmospheric measurements, atmospheric river records, precipitation logs, and historical flash flood warnings
• Scope: California only at launch; nationwide rollout planned

Technical Foundations and an Unexpected Statistic

TACLS leverages a clever repurposing: GNSS networks, originally designed for earthquake monitoring, detect atmospheric water vapor through signal delay between satellites and ground sensors. Higher atmospheric moisture = longer signal travel time, quantified as ‘precipitable water’—a real-time metric of what’s actually happening in the atmosphere before storms form.

This contrasts sharply with legacy systems relying on rain gauges and radar, which only trigger alerts after precipitation begins. As project lead Yehuda Bock explains: ‘Once there’s precipitation, you’re already in the event itself.’

A revealing inefficiency: In areas with sparse infrastructure like deserts, rain gauges are scarce (NWS admits coverage gaps exist), yet GNSS ground stations are more plentiful. This creates accidental parity—low-population high-risk zones gain better monitoring not through new hardware, but by reusing existing geodetic infrastructure.

Senior hydrologist Jayme Laber notes TACLS validates forecast models: rather than trusting predictions blindly, forecasters compare model output with real-time moisture observations to adjust timing and severity assessments.

Integration into Existing Warning Protocols

TACLS augments but does not replace NWS’s established alert hierarchy. NWS defines a flash flood as occurring within six hours of heavy rainfall. Three alert levels guide response:

• Watch: Issued 12–48 hours ahead, prompting emergency teams to review evacuation plans, pre-deploy sandbags and barriers
• Advisory: Nuisance-level flooding unlikely to threaten life or property
• Warning: Imminent danger—risk of being swept away or struck by floating vehicles

The LSTM model processes GNSS-derived moisture trends against historical thresholds calibrated for local soil and terrain. These thresholds, codified in NWS flash flood guidance, specify exactly how much rain in how long triggers danger in a given area. The neural network excels at recognizing patterns in time-sensitive atmospheric changes during storm development.

Practical Guidance for Stakeholders

• Ideal adopters: Western arid/semi-arid states (California, Nevada, Utah) where sparse rain gauges create monitoring gaps but GNSS coverage remains strong; mountainous regions with narrow valleys where flash floods develop in minutes
• Best to wait: General public—no action required, as alerts flow through standard NWS channels regardless of underlying technology; low-risk flat regions with dense gauge networks may see marginal incremental benefit initially

In Closing

Flash floods rank as the second-deadliest weather event in the U.S. (deadliest globally), and current systems often deliver warnings too late for meaningful action. TACLS marks a pivotal shift: From reactive forecasting to proactive detection, buying critical hours instead of minutes. In an era of intensifying extreme rainfall due to climate change, such software-layer innovations—deployable without new infrastructure—offer scalable life-saving potential.