Everyone is talking about artificial intelligence, and even the dreaded world of hurricanes is already benefiting from it. Here’s an overview of how AI is helping to provide more accurate and detailed forecasts for the cyclones that threaten the Sunshine State every year from June 1 to November 30, the official hurricane season.
A Look Back at the 2025 Season
The National Hurricane Center (NHC) confirms it: although it has been using AI tools for many years, new tools emerged faster than ever in 2025. Last season served as a testing ground, particularly thanks to an initial collaboration with NHC partners to conduct thorough verifications and tests before using AI tools for operational decisions. During the 2025 season, as forecasters gained experience, the NHC began integrating these new AI weather prediction systems (AIWP – AI Weather Prediction) alongside other essential tools in its toolkit.
The devastating Hurricane Melissa, which struck Jamaica and tied the record for the most powerful hurricane to make landfall with winds of 185 mph (298 km/h), demonstrated that AI-based forecasts continue to improve dramatically, said NHC Director Michael Brennan, at the Governor’s Hurricane Conference in West Palm Beach. “These are forecasts we simply weren’t able to make five or ten years ago,” he noted.
According to the Center, Hurricane Melissa was a very difficult storm to predict and one with significant consequences. AI models identified the likely track and intensity very early on, providing valuable insights to supplement traditional forecasts. While the case of Melissa is just one example among many, there are also instances where traditional models have yielded better results, according to the NHC. But for the Center, it remains encouraging to see that the new tools are working well.
AI Models Used
While the NHC began using several different AI models, it partnered directly with Google DeepMind to develop a new AI-based hurricane forecast model, which was used experimentally during the 2025 season. In addition, NOAA’s Center for Environmental Modeling and the European Centre for Medium-Range Weather Forecasts (ECMWF) have developed AI-based global forecasting systems, both of which were made available for evaluation during the 2025 season. Other AI-based tools and models are also under development.
How It Works
The NHC explains that AIWP models are trained to learn patterns by analyzing vast amounts of historical data: “They learn the relationships between different variables and how these relationships vary in space and time.” Most current AI models are trained on reanalysis datasets—huge global datasets that incorporate all global observations and span several decades. During this extensive training process, AI models learn the relationships between different variables (such as pressure, wind, and temperature). They then use these relationships to estimate the future state of the atmosphere.
Difference from Traditional Models
Although AI represents a very different approach from traditional weather forecasting models, the ultimate goal is the same: to produce the most accurate forecasts possible.
Traditional Numerical Weather Prediction (NWP) and the new AIWP models both use these two approaches to generate forecasts: NWP models start with current atmospheric conditions (called “initial conditions”) and solve atmospheric equations to predict the future, requiring massive supercomputers for each model run. AIWP models also start with initial conditions, but the model uses relationships it has learned from historical data to predict the future. Because AIWP models learn the relationships between variables through a learning process, they run much faster than NWP models.
Despite these differences, traditional NWP models and AI models are highly complementary.
Benefits and Improvements with AI
According to the NHC, because AI models operate differently from traditional NWP models, they provide new and independent insights into the range of forecast outcomes. “And because they run very quickly on supercomputers, we may soon have thousands of possible forecast outcomes to estimate the uncertainty regarding the path, intensity, and hazards associated with hurricanes,” meaning we can capture the range of possibilities and communicate risks to decision-makers and the public with greater certainty than before.
Not Without Concern
Despite the enthusiasm for AI-generated forecasts, the Trump administration has once again proposed cutting the NOAA budget by $1 billion and eliminating the Office of Oceanic and Atmospheric Research, which conducts research on hurricanes and the atmosphere. According to WLRN, this could potentially lead to the closure of the Atlantic Oceanographic and Meteorological Laboratory in Virginia Key, just north of Key Biscayne and south of Miami’s Brickell neighborhood.
Meanwhile, the National Center for Atmospheric Research (NCAR) is fighting in court to prevent the center’s dismantling.
Useful Links
During this particularly dreaded time of year, we invite you to follow LeSoleildelaFloride.com and our social media channels to stay informed.
We also recommend checking the following links, if needed:
• Updated forecasts and AI-powered models – National Hurricane Center (NHC)
• Local vulnerability and evacuation zones
Florida Division of Emergency Management
https://www.floridadisaster.org
• National Oceanic & Atmospheric Administration (NOAA)
• What to do before a tropical storm or hurricane
Preparedness guide https://www.weather.gov/safety/hurricane-plan
A calmer-than-normal season
For the 2026 Atlantic hurricane season, NOAA forecasts a 35% chance of a near-normal season,
a 10% chance of an above-normal season, and a 55% chance of a below-normal season, with a total of 8 to 14 named storms (winds of 39 mph–24 km/h or higher). Of these, 3 to 6 are expected to develop into hurricanes (winds of 74 mph–119 km/h or higher), including 1 to 3 major hurricanes (Category 3, 4, or 5 with winds of 111 mph–179 km/h or higher).
Remember that advance preparation is essential to staying safe throughout the season.
Article translated by DeepL.
