Hurricane forecast accuracy

The metrics

The track improvement story

• 1990: 24-hour forecast error 100 miles average.
• 2000: 24-hour forecast error 60 miles.
• 2010: 24-hour forecast error 40 miles.
• 2020: 24-hour forecast error 30 miles.
• 2025: 24-hour forecast error 25 miles.
• 5-day track forecasts today are as accurate as 2-day forecasts were in 1990.
• Improvement rate: 1% per year.

What drove track improvement

1. Better initial conditions from Hurricane Hunter dropsondes.
2. Higher-resolution models (HWRF, HAFS).
3. Better data assimilation.
4. Ensemble forecasting.
5. Satellite advances (GOES-R series).
6. AI-augmented forecasts (increasing).
7. Better representation of steering flow.

The intensity problem

• Intensity error has barely improved since 1990.
• 2025 24-hour intensity error still 10-15 knots.
• 5-day intensity error still 25+ knots.
• Rapid intensification detection: significantly harder.
• Rapid weakening similarly.
• The physics of eyewall replacement, rapid intensification, and inner-core convection remain challenging.

Rapid intensification

A hurricane that gains 35+ mph of wind speed in 24 hours is undergoing rapid intensification (RI).

• Increasingly frequent since 2017.
• Katrina (2005), Michael (2018), Harvey (2017), Ian (2022), Otis (2023), Milton (2024) all underwent RI near landfall.
• RI near landfall gives less evacuation time.
• Predicting RI 24-48 hours out is the frontier.
• HRRR, HWRF, HAFS all trying to solve this.
• AI-augmented forecasts show promise.
• Ocean warmth is a driver.

The Cone of Uncertainty

NHC issues track forecasts with a 'cone of uncertainty' — the area within which the storm center is expected to fall 2/3 of the time.

• Cone widens with forecast time (more uncertainty).
• Cone width has SHRUNK over decades as track forecasts improved.
• Cone is misunderstood — does NOT bound impact area.
• Hurricane winds and rain extend far outside cone.
• Storm surge extends farther than the eye.
• Deaths often outside the cone track center.

The cone of uncertainty problem

• Public thinks: "if I'm outside the cone I'm safe."
• Reality: hurricane impacts extend 100-300 miles from center.
• Wind field, rain field, storm surge zone are separate maps.
• NHC has explored replacing/supplementing the cone.
• Some communities have added "wind swath" and "surge inundation" maps.
• Still: the cone is what most people see.

The models

The specific improvements needed

• Better ocean coupling.
• Better representation of moist processes in inner core.
• Ensemble intensity forecasting.
• AI acceleration of hurricane physics.
• Multi-model consensus.
• Real-time dropsonde integration.
• Better representation of asymmetric storm structure.

For emergency managers

• Track improvements have enabled better evacuation planning.
• Intensity uncertainty means evacuation for weaker forecast is often prudent.
• The 2018 Michael forecast underestimated the Cat 5 landfall.
• The 2022 Ian forecast underestimated the strengthening.
• Modern practice: plan for Cat +1 for forecast Cat 3.
• Storm surge planning must consider intensity uncertainty.

For public safety

• Do not focus on Category — total impact matters.
• Watch NHC updates (every 6 hours during hurricane season).
• Public advisory + track map + wind field map.
• Storm surge maps for coastal.
• Local emergency management app.
• WEA alerts.
• Assume storm may be worse than forecast.

Learn more

• Hurricane hunter aircraft
• Hurricanes in US history
• Hurricane season preparedness
• How forecast accuracy improved