There's a ‘ghost hurricane' in the forecast. It could help predict a real one

CNN

a day ago

Hurricanes
Storms
Social mediaFacebookTweetLink
Follow
A scary-looking weather forecast showing a hurricane hitting the Gulf Coast in the second half of June swirled around social media this week—but don't panic.
It's the season's first 'ghost hurricane.'
Similar hype plays out every hurricane season, especially at the beginning: A cherry-picked, worst-case-scenario model run goes viral, but more often than not, will never come to fruition.
Unofficially dubbed 'ghost storms' or 'ghost hurricanes,' these tropical systems regularly appear in weather models — computer simulations that help meteorologists forecast future conditions — but never seem to manifest in real life.
The model responsible this week was the Global Forecast System, also known as the GFS or American model, run by the National Oceanic and Atmospheric Administration. It's one of many used by forecasters around the world.
All models have known biases or 'quirks' where they tend to overpredict or underpredict certain things. The GFS is known to overpredict tropical storms and hurricanes in longer-term forecasts that look more than a week into the future, which leads to these false alarms. The GFS isn't alone in this — all models struggle to accurately predict tropical activity that far in advance — but it is notorious for doing so.
For example, the GFS could spit out a prediction for a US hurricane landfall about 10 days from now, only to have that chance completely disappear as the forecast date draws closer. This can occur at any time of the year, but is most frequent during hurricane season — June through November.
It's exactly what's been happening over the past week as forecasters keep an eye out for the first storm of the 2025 Atlantic hurricane season.
No weather forecast model is designed in the exact same way as another, and that's why each can generate different results with similar data.
The reason the GFS has more false alarms when looking more than a week out than similar models – like Europe's ECMWF, Canada's CMC or the United Kingdom's UKM – is because that's exactly what it's programmed to do, according to Alicia Bentley, the global verification project lead of NOAA's Environmental Modeling Center.
The GFS was built with a 'weak parameterized cumulus convection scheme,' according to Bentley. In plain language, that means when the GFS thinks there could be thunderstorms developing in an area where tropical systems are possible – over the oceans – it's more likely to jump to the conclusion that something tropical will develop than to ignore it.
Other models aren't built to be quite as sensitive to this phenomenon, and so they don't show a tropical system until they're more confident the right conditions are in place, which usually happens when the forecast gets closer in time.
The western Caribbean Sea is one of the GFS' favorite places to predict a ghost storm. That's because of the Central American gyre: a large, disorganized area of showers and thunderstorms that rotates over the region and its surrounding water.
The combination of abundant moisture and spin in the atmosphere makes it a prime breeding ground for storms, especially early in and during the peak of hurricane season. Given the model's sensitivity, it's quick to pounce on these possible storms.
But this sensitivity has an advantage: By highlighting almost anything that could become tropical, the GFS misses very few actual storms.
Its tendency to cry wolf isn't ideal, but the GFS team found it was worth giving the model a higher chance at catching every storm and better predicting each one's intensity than to prioritize fewer false alarms, Bentley explained.
'It was critical to improve the probability of detection of tropical cyclone formation and tropical cyclone intensity forecasts… and we did achieve that,' Bentley said.
During the 2024 hurricane season, the GFS had the least error when forecasting the intensity of tropical cyclones – tropical depressions, tropical storms and hurricanes – of any other global forecast models in its class: the ECMWF, the CMC and the UKM. However, the ECMWF and UKM outperformed the GFS in tropical track forecasts out more than five days in the future.
Despite how often the GFS conjures ghost systems in its longer term forecasts, it can't be discounted.
'The crucial role of the forecaster is to understand a model's known biases and use that knowledge to their advantage to produce a better forecast,' Bentley said.
Knowing the GFS latches onto anything that could even vaguely become tropical well in advance helps forecasters keep an eye on areas where conditions may ultimately come together to create the next hurricane.
The more reliable solution for predicting tropical behavior more than a few days in advance is to take advantage of ensemble forecasting, according to Bentley.
'A deterministic model like the GFS produces one forecast at a time; it gives one answer,' Bentley explained. 'An ensemble forecast can show you a variety of possible outcomes, as well as which forecast looks like a possible outlier.'
Unlike social media clickbait, no well thought out forecast is made from a single model run. Forecasters use everything at their disposal – deterministic and ensemble models, observations, climatology and more – to predict weather as accurately as possible to give people the time and information they need to stay safe.
The National Hurricane Center, for example, typically uses a blend of different types of models to make their forecasts. That strategy, combined with extensive expertise, led to their most accurate track forecasts on record for the Atlantic last season.