A Q&A with NHC Science Operation Officer Wallace Hogsett

AI model forecast guidance for Tropical Storm Gabrielle (2025). Google DeepMind (orange) and NOAA AI-GEFS (yellow) ensemble forecast systems are both displayed in NHC’s experimental cloud AWIPS system.

1. Artificial intelligence (AI) is a hot topic. How is the National Hurricane Center using AI in hurricane forecasting, especially during last year's hurricane season? 
   
   The use of AI in hurricane forecasting is accelerating. While NHC has used forms of AI tools for many years, new tools emerged more quickly than ever in 2025. Last season was all about experimentation, including first working with our partners to conduct thorough verification and testing prior to using any AI tools for operational decisions. During the season, as forecasters gained experience, NHC began integrating these new AI weather prediction (AIWP) systems as guidance when preparing operational forecasts, alongside all of the other critical tools in our toolbox.  

2. Which specific AI models is the NHC currently using?
   
   We have begun using a few different AI models. NHC has partnered directly with Google DeepMind to develop a new AI hurricane forecast model that was used experimentally during the 2025 season. Additionally, NOAA’s Environmental Modeling Center and the European Center for Medium-Range Weather Forecasting (ECMWF) have developed AI-driven global forecast systems, both of which were also available for evaluation during the 2025 season.  More AI-based tools and models are coming, and we’re evaluating them all thoroughly for potential integration into forecast operations. At NHC, we are just beginning to scratch the surface of how these new models may be used.

3. In simple terms, how do these AI forecasting models actually work?
   
   Essentially, AIWP models are “trained” to learn patterns by analyzing vast amounts of historical data. They learn relationships between different variables and how those relationships vary over space and time. Most of the current AI models are trained on “reanalysis” datasets, which are huge global datasets that incorporate all global observations and span many decades. During the extensive training process, the AI models learn relationships between all of the different variables (such as pressure, wind, and temperature). Then they use those learned relationships to estimate the future state of the atmosphere.  While AI is a very different approach from traditional weather forecast models, the end goal is the same – to produce the most accurate forecast possible. 

4. What's the main difference between the traditional, physics-based weather models (like the GFS) and the newer, learning-based AI models?
   
   Traditional numerical weather prediction (NWP) and the newer AIWP models take very different approaches to produce a forecast. A primary difference is that NWP models start with the current atmospheric conditions (called “initial conditions”) and solve the equations of the atmosphere to predict the future. Solving these equations requires enormous supercomputers for each model forecast. 

   AIWP models also begin with the initial conditions, but instead of solving the equations of the atmosphere, the model uses the relationships it has learned from history to predict the future.  Because the AIWP models learn the relationships among variables during a training process, AIWP models run much more quickly than NWP models.    
   
   Despite these differences, traditional NWP models and AI models are highly complementary. Because they use different methodologies, their forecasts have different types of errors, which can be very valuable to forecasters. These differences can be very helpful for forecasters to understand the full range of uncertainty and communicate the risk more effectively.

5. How is the NHC combining AI models with the existing forecasting process?

   At NHC, we have many decades of experience integrating new tools into our forecast “toolbox”. This process includes a period of evaluation and experimentation, which allows us to understand how the new models perform in a variety of situations, and understand their strengths and weaknesses. In addition to the AI weather prediction models, AI also has potential to help make our forecast process more efficient by sifting through the vast amount of data at forecasters’ disposal. During 2024 and 2025 we were really focused on this process of learning and building trust in the new tools. Now we are ready to begin incorporating the new AI models into our toolbox, although the learning will continue! 

6. What new benefits or improvements is this AI technology bringing to the hurricane forecast?

   Forecasting hurricanes is all about communicating a range of possibilities and the risks associated with hurricane hazards. Because AI models operate differently than traditional NWP models, they provide a new, independent guidance on the range of possible outcomes. And because they run very quickly on supercomputers, we may soon have thousands of possible forecast outcomes to estimate the uncertainty in track, intensity, and hurricane hazards.  Ultimately, this means capturing the range of possibilities and communicating risk to decision-makers and the public more confidently.
    
7. Thinking back to the 2025 Hurricane Season, can you give an example where AI-based tools made the forecast significantly more accurate?
   
   First, I would like to encourage everyone to focus on the NHC forecast, as verification analysis over multiple storms and forecasts shows the official forecast is the most skillful and consistent, surpassing any individual model forecast. It takes a long time to fully evaluate the skill and reliability of new models, and we don’t like to point to a single model’s success or failure as indicative of the overall value of a new tool. However, one storm that stood out in 2025 was Hurricane Melissa, which was a very difficult-to-forecast and impactful storm. The AI models honed in very early on the likely track and intensity and provided very valuable guidance to complement our traditional NWP guidance. It is just one example, and there are other examples where the traditional models performed better. However it was promising to see the new tools performing well.  This is all part of the learning process for forecasters, and we’re still in the early stages. 
    
8. What are the biggest pluses and minuses—the advantages and disadvantages—of using AI for hurricane forecasting?
   
   AI systems are computationally very efficient – they can produce a forecast very quickly. This opens the door to produce very large ensembles, or ranges of possible solutions, to provide hazard risk information and enable more informed decisions for those in the path of the storm.  On the other hand, AI systems are evolving quickly and require one to constantly understand why the model generated the result that it produced. Active research is ongoing to help forecasters understand not only the answer the model produces, but also why it produced that answer. We will likely have a more complete answer in a few years as the technology matures and we gain more experience over many different storms.
    
9. With all the recent progress in technology, do you think AI will eventually replace the human hurricane forecaster?
   
   We often hear this question, and we’re learning that the answer is a resounding “no”. The number of available tools and models is increasing at a pace that is faster than I have ever seen in my career. However, it is critical that we understand the tools’ outputs, strengths, and weaknesses to continue providing forecasts and warnings that are trusted for decision-making. None of the models are perfect, and they never will be.  Now more than ever, we need trusted experts in the loop to observe, synthesize, and make sense of the vast amounts of information. AI will likely assist in efficiently synthesizing information, but without experts in the loop to constantly evaluate and integrate new tools and technologies and communicate a coherent risk-based message, lives and property would be at greater risk. 
    
10. Anything else to add that people should know? 
    
    Every person at NHC takes very seriously our mission to protect lives and property from the risks posed by hurricanes. We will continue to move very quickly to incorporate new tools, which allow us to constantly improve our forecasts, warnings and hazard risk information that we provide to everyone in harm’s way.   

For more information, please contact nhc.public.affairs@noaa.gov