In brief: The initial release of ChatGPT in November 2022 sparked a widespread recognition of the promise and peril of artificial intelligence. We’ve been watching these developments closely, both in the broader public context as well as in weather modeling. For the sake of transparency, here’s how we plan to use AI now and in the future.
As artificial intelligence plays an increasingly prominent role in society, I want to clarify what this means for Space City Weather now, and in the future. The central message I wish to convey is that Space City is, and always will be the creative product of human beings. What is written and forecast here is done by people. At the same time, Matt and I are not burying our heads in the sand. As technology changes, we embrace that which improves our ability to forecast the weather and communicate that information.

For example, I started in newspaper journalism more than 25 years ago. At the beginning of my career, printed newspapers still largely set the news agenda. But in 2005, while still at the Houston Chronicle, I started blogging. With the arrival of hurricanes Katrina and Rita that summer, I soon realized the power of immediate communication on the Internet, the value of sharing links to credible information, and the hunger of people for this change. In the two decades since, I have made the majority of my living writing for online publications, without a paywall. This is one reason why I am so committed to keeping Space City Weather free and open to all. We started as a web site, grew on social media, and now many of you are reading this on an app. I cannot foresee where things will be in 10 years, but wherever people want to read Space City Weather, we will be there.
The rise of AI and weather
Artificial intelligence has been lumbering along in the background for several years, but it has really broken through recently with large language models such as ChatGPT. These artificial intelligence systems can be quite good at some tasks. One of these turns out to be forecasting the weather. There are repositories of meteorological data that go back more than 50 years that can be used to ‘train’ designated models for weather prediction. There are now about 10 major groups out there developing AI-based models, and some of them have gotten quite good. Further improvements are likely.
Traditional weather models, which are based on complex physical equations, attempt to simulate the atmosphere and crunch through those equations to predict what will happen next. These physics-based models have gotten steadily better over the years, especially because of more powerful supercomputers and sophisticated tools to ingest more real-world weather conditions (i.e. the temperature at 10,000 feet above the surface of the Atlantic Ocean) into the model before it is run. AI-based models perform none of these physical computations. They can, therefore, be run incredibly quickly, on fairly simple computers.
How we will use AI
Matt and I have been following the output of AI-based weather models for more than a year now, and they definitely have their strengths. Such models are quite good at three- and five-day forecasts to the point where they sometimes outperform physics-based models at tasks such as hurricane tracks. They also have their weaknesses. For example, AI-based models are not (yet) good at high-resolution modeling and predicting the development of thunderstorms. So no, we still won’t be able to tell you whether it’s going to rain at your house at 3 pm on some days.
The bottom line is that we are using AI-based modeling tools as a part of our forecasts here at Space City Weather. They’re not a panacea, but they are another tool in our arsenal that runs the gamut from hand-drawn isobar maps to sophisticated models on supercomputers. I suspect they will be an even more useful tool over the next five years. But that is where our use of artificial intelligence will end. We have not, nor will we use any AI-based service for the writing of our forecasts. Very occasionally we may use an AI-based illustration, but if we do it will be clearly labeled as such.
We want you to know that at Space City Weather our commitment is to show up every day and make each forecast with our best effort, without influence or intervention from anyone else. We don’t win them all. Humans are fallible. But you can rest assured that they are honest mistakes rather than AI-induced hallucinations.
— Eric Berger, January 2025
