NHC Atlantic track verification · 2005 to 2025
The Forecast That Publishes Its Score
The National Hurricane Center issues a track, keeps the simple CLIPER5 baseline beside it, and publishes the comparison every year. This is a working operational system being honest about its own skill.
Official NHC mean track error is lower in every available Atlantic year-and-lead cell in the 2005 to 2025 reports.
2024 official 24-hour error, against 88.4 n mi for CLIPER5 across 276 paired forecasts.
CLIPER5 usually loses, and here it loses decisively. The official forecast is not being caught out by a simple rule. The baseline exists to define what skill means, and the annual scorecard shows substantial skill.
The complete annual run
Open any published cell
Each button is one NHC annual-table cell. The large number is skill computed live as 100 × (CLIPER5 error − official error) / CLIPER5 error. The smaller pair is official / CLIPER5 mean error in nautical miles.
| Year | 12 h | 24 h | 36 h | 48 h | 60 h | 72 h | 96 h | 120 h |
|---|
Select a cell. The 2024 24-hour cell is ready.
Control 3 · honest counterfactual
Where the simple rule really wins
These exceptions do not overturn the annual result. They show why the baseline remains a real comparator. Both are storm-level cells published by NHC, and both expose the sample size.
The advisory replay · Hurricane Beryl, 24 hours
Close the record, then open the score
Advance through all 50 archived Forecast/Advisories. The firewall uses each product's printed UTC issue time, not the nominal model initialization. Three advisories lack a homogeneous 24-hour pair and remain visible as exclusions.
24-hour score through this issue
At issuance, neither arm had the coordinate required to calculate these errors.
The complete Beryl run
All 47 paired advisories, including every CLIPER5 win
| Advisory | Issue UTC | Official | CLIPER5 | Winner |
|---|
Control 4 · what the record could not tell them
The scoring coordinate did not exist yet
Issue: 3 July 2024, 1500 UTC
Post-season archive copy: 18 April 2025
At issuance, both the official forecaster and CLIPER5 could know their own forecast positions. Neither could know the later best-track coordinate at the verifying horizon. NHC reconstructs best track after the season from observations that were unavailable at the cutoff, and the reconstruction can differ from the operational center estimate.
The eventual great-circle error, the winner of a forecast pair, and the whole scorecard cell were therefore unavailable. Pressing Open the later score makes that change of information state explicit. It does not pretend the score was sitting unseen in the 2024 advisory.
The check
The official 2024 anchor is recomputed live from all 276 individual 24-hour errors in NHC's error database. The matrix skill is recomputed from the published annual error pairs. Beryl distances are recomputed from the archived forecast and best-track coordinates with the haversine formula.
Uncertainty: the public ATCF forecast coordinates are rounded to 0.1 degree, so this page does not claim to reconstruct the annual CLIPER5 error exactly from those deck positions. The 88.4 n mi annual value is transcribed from NHC Table 2. Beryl's coordinate-derived errors are correspondingly approximate. The exact first publication time of the post-season Beryl best track was not established; the instrument uses the archived b-deck file's 18 April 2025 last-modified time as a conservative availability marker.
Free choices: 2005 starts the CLIPER5 era; the matrix is Atlantic; the storm replay is Beryl at 24 hours; Earth radius is 3,440.065 n mi; zero-hour equality is omitted; missing 60-hour cells remain missing; a tie is any displayed difference below 0.05 n mi.
Data and licence: the annual cells are transcribed from NHC annual verification tables. Advisory times, a-deck forecasts, b-deck best track, and the individual error file are NOAA/NWS federal forecast records. They are public-domain material. The shipped snapshots and their derivation program are in research/nhc-forecast-versus-cliper5/.