On the date of the project revealed by NASA, the agency announced a new system that uses machine learning to improve flash flood warnings. The effort combines satellite data with advanced algorithms to help meteorologists make flash flood predictions faster and more accurately. This research is a collaboration between NASA’s Jet Propulsion Laboratory, the University of California, San Diego, and NOAA’s National Weather Service, with support from NASA’s Earth Science and Technology Office. What’s more, TACLS is designed to speed up alert generation while keeping human judgment at the center of the final decision.
The system, called TACLS, automatically searches satellite information for signs of increased humidity in the atmosphere, a sign that could precede a flash flood. Identify where flooding is likely to occur and display results in an easy-to-read visualization for analysts. Human weather forecasters then use their professional judgment to decide whether to issue a flash flood warning or weather warning. Essentially, TACLS acts as a decision aid rather than a replacement for human analysis.
In practice, TACLS operates in near real-time and can provide predictions within about 15 minutes. This rapid response is intended to help meteorologists respond more efficiently during severe weather events. Dr. Yehuda Bock of UCSD’s Scripps Institution of Oceanography, who is leading the project, emphasizes that the goal is to provide tools that support faster and better-informed warnings.
During testing, TACLS was exposed to data from a variety of severe weather events, including atmospheric rivers, monsoon storms, and tropical cyclone remnants from 2017 to 2023. The system successfully identified 93% of the flash flood warnings issued in these simulations. National Weather Service forecasters are exploring ways to integrate TACLS into the existing Southern California forecast process.
There are two main parts to a TACLS setup. The analytics backend uses machine learning to process satellite data and mark flood risk areas. The visualization component, MGVis, highlights these areas for human review. The backend relies on data from the Global Navigation Satellite System and uses signal delays caused by water vapor in the atmosphere to estimate moisture levels. A machine learning model trained on over 30 years of GNSS data acts as an anomaly detector, distinguishing between artifacts in the data and actual time-sensitive moisture changes.
If the system flags a temporary event, such as heavy rain, the results are sent to the MGVis visualization for human evaluation. Analysts examine the information and determine whether the data indicates the possibility of flash flooding and whether a warning should be issued. This project also leverages existing JPL software concepts to process GNSS data and display results.
Both the TACLS software and training data will be open source, allowing researchers to adapt models or build new models from scratch. The NASA TechPort page provides more information, and the project is led by Dr. Yehuda Bock with support from NASA’s ESTO, JPL, NOAA, and the National Weather Service.
