By Rear Admiral Tim Gallaudet, Ph.d., U.s. Navy ret
The fatal flooding this weekend in Central Texas was a tragedy. As of this writing, at least 80 are dead. After every such crisis, its important to assess how it occurred and identify ways to prevent loss of life in the future.
To begin, lets first look at the criticisms by state officials of the National Weather Service (NWS) for failing to forecast the flooding. This appears valid, as the NWS Austin/San Antonio office called for 3-6 inches of rainfall when well over ten inches actually occurred. NWS did in fact forecast the possibility of heavy rains in the Texas Hill Country on July 3rd, but it was only after the Guadalupe River began to surge on the morning of July 4th that an urgent flood emergency was declared.
By then, it was too late. The river rose an incredible 26 feet in 45 minutes, catching local officials completely off guard.
Of course, many were quick to condemn the Trump Administration for directing the massive layoffs and early retirements that drastically cut NWS staff as the likely cause for degrading performance and forecast failures.
While staff shortages at NWS are a significant challenge which prompted leadership in the agency to prepare to offer degraded forecasting services, I do not believe that was the primary factor with the Texas flood forecasts. Earlier this year, NWS offices in Kentucky brought in extra staff to predict and issue warnings for a severe tornado outbreak. During my visits to dozens of NWS offices, I saw firsthand how the fine public servants at NWS perform their duties with a failure is not option mentality.
I believe a combination of factors led NWS to miss the mark in Central Texas, and all can be fixed with new technology solutions available today. First, the biggest impact from staff cuts has been fewer weather balloon launches. The resulting loss of data has decreased the accuracy of the NWS global model in one particular metric by twenty percentage points. NWS could remedy this shortfall immediately by purchasing commercial satellite data through a program NOAA has had in place for years.
The next action that should be taken is to incorporate more artificial intelligence (AI) into NWSs integrated atmospheric, oceanic, and hydrologic modeling systems. I helped advance this during my tenure at the National Oceanic and Atmospheric Administration (NOAA), NWSs parent organization, by leading the development of an agency-wide AI Strategy and AI Strategic Plan, as well as establishing the NOAA Center for AI. Since then, NOAA and NWS have made great progress with AI, the most recent example being Project EAGLE, which is providing the ability to rapidly test, develop, and demonstrate AI-based models for global and regional forecasting. However, these efforts are not keeping pace with AI weather models in industry, which are becoming more accurate, efficient, and hyperlocal than their federally funded counterparts.
For example, before adopting an AI-based hurricane model, the National Hurricane Center in NWS has elected to undertake a lengthy research and development partnership with Google. Considering that the Trump administration has directed federal contracts to prioritize commercial products and services over non-commercial ones, and that advancing AI is a key focus area for the White House, it might be time for NWS to begin outsourcing model development and upgrades entirely to the private sector.
Finally, and maybe the most transformative action that NWS can take to improve performance is to adopt agentic AI capabilities in the delivery of decision support services. Today, NWS forecasters employ a combination of AI and conventional Earth system models to provide emergency managers and members of the public information and guidance for effective decision making. Agentic AI offers the next level, where the recommendations for action are fully automated.
The transformative aspect of this approach would be removing the role of NWS meteorologists in developing and delivering forecasts and warnings. Instead, they would employ agentic AI tools tailored to the wide range of conditions to predict, from hurricanes and heat waves to wildfires and windstorms. This by no means negates the need for NWS personnel, who still will be needed to manage the standards, configuration, test and evaluation, and operational quality control of those tools.
This would be a big change for NWS, requiring the commercial development of an enterprise agentic AI architecture that would include technologies such as domain-specific small foundation models, Model Context Protocol (MCP) servers to make tools and data available, and vector and semantic databases. It would not require a major budget increase, as NOAA already has high-performance computing resources and a cloud computing environment to support the architecture. These could be re-tooled from their research and development focus, which outsourcing AI model development to industry would enable.
Had these capabilities been in place this weekend, a NWS agentic AI tool for coupled atmospheric-hydrologic forecasting would have had the right data, a better model, and an automated emergency alerting system that may have given the citizens in Central Texas a more specific flood warning 24 hours in advance, allowing them to move to high ground when they could.
The dedicated men and women in NWS want to do their very best to keep the public safe. The Trump administration would be wise to give them the emerging technology of today to make that happen.