Why hurricane track forecasts keep improving — and what they still miss

Twenty years of cone-of-uncertainty refinements have shrunk average track error by 40%. Landfall intensity at T+72 remains the harder problem, and the 2025 season proved it.

The 2025 Atlantic season produced 18 named storms, 9 hurricanes, and 4 major hurricanes — a hyperactive year by any measure. But what stood out to forecasters at the basin forecast desk wasn't the raw count. It was the widening gap between track accuracy and intensity skill, visible in every 5-day advisory package issued between August and November. Three storms — Franklin, Idalia, and a late-season Ophelia — made landfall at intensities substantially above the T+72 forecast, each by at least one Saffir-Simpson category. The cone had narrowed. The landfall surprise had not.

The track-intensity divergence

Since 2000, the desk's average 72-hour track error has fallen from roughly 240 nautical miles to under 90 — a 62% improvement driven by better global models, more reconnaissance data, and the quiet revolution in ensemble forecasting. Intensity error, over the same period, has improved by barely 15%. The two curves separated around 2010 and have been diverging ever since. Every forecaster I interviewed for this piece used the same phrase: “the track-int gap.” They didn't need to explain it.