2025 Hurricane Forecasts Continued To Improve, NHC Report Finds

Impacts From Storms Go Well Beyond The Cone

The National Hurricane Center has issued it's final performance report for its own forecasts in the 2025 Atlantic Hurricane Season and the computer models that it uses to forecast tropical storms and hurricanes.

Overall, the season saw the need for fewer forecasts from the NHC, but rapid intensification continues to throw forecasters a curveball. The season saw 240 forecasts for tropical depressions, storms or hurricanes, down from the average of 325 forecasts. This is partially due to the slightly lower number of named storms during the season – 13 in total. The season was notable, however, because of the number of rapidly intensifying hurricanes and the three Category 5 storms that came during the season.

Here's how the NHC performed during those storms:

Track Forecasts

The forecast for the movement of tropical systems was as much as 14% better than the 5-year average. That turns into track errors of 20 miles at 12 hours into the future, growing to 162 miles at 120 hours into the future.

The paths of storms were up to 8% easier to forecast than average during the season for forecasts made up to three days into the future.

Track errors were notably reduced for major hurricanes, including Humberto and Melissa, during the season. This has been common in recent seasons.

There were a couple of trickier storms, though. Much higher errors were seen for Fernand and Imelda. Imelda's track forecasts were mucked up by its Fujiwhara interaction with Hurricane Humberto, which pulled the storm, then a hurricane, away from the Southeast Coast and into the Atlantic, closer to Bermuda. Interestingly, the intensity forecast for Imelda was “excellent.”

(MORE: Hurricanes Humberto and Imelda's Intertwined Recap)

This didn't lead to any records for projected path this year, but forecasts are trending in the right direction. Over the last few decades, track errors at 24-72 hours have been reduced by 75%.

Artificial intelligence (AI) rocked the boat in 2025, and it is was no surprise that the premiere AI model was at the top of the ranks for track forecasts. The Google DeepMind ensemble average outperformed NHC from 12-hr to 72-hr.

The hurricane-specific models were competitive throughout the entire forecast.

As a result of improvements in forecast, the cone will be up to 8% narrower in the 2026 season compared to 2025.

Intensity Forecasts

Unlike the track forecasts, those for intensity were much more difficult to make accurately compared to average. Intensity errors were higher than the 5-year average at all timestamps. Part of this is due to the higher number of rapid intensification forecasts that were made during the season.

There were 23 cases of rapid intensification – a much higher proportion than in an average season, making up 10% of all forecasts during the season. NHC continues to perform better than the models at almost all timesteps in cases of rapid intensification. Larger errors were seen especially during Humberto’s rapid intensification phase.

(RECAP: 2025 Atlantic Hurricane Season)

Overall, there has been a general decrease in errors over the last couple of decades.

Google DeepMind was the best performing model and was on par with the NHC.

As has been the case for years running, the Florida State Super Ensemble (FSSE) continues to perform very well. The best hurricane models are the hurricane-specific HAFS models and the lower-resolution global models, the GFS and European models had little to no use.

Jonathan Belles has been a digital meteorologist for weather.com for 9 years. His favorite weather is tropical weather, but also enjoys covering high-impact weather and news stories and winter storms. He's a two-time graduate of Florida State University and a proud graduate of St. Petersburg College.