How we grade
Track, wind, rain, and temperature are scored when a forecast lines up with what later happened. Television stations are not graded.
This page says, in plain English, what gets a score, what gets left out, and where the numbers come from. The methodology page is the technical reference the scorer reads. Both pages describe the same rules.
What is scored
A hurricane forecast is scored on two things: where the storm was predicted to be, and how strong its wind was predicted to be. Path miss is the distance, in statute miles. Wind miss is the difference in wind speed. The storm page shows that wind in miles per hour. The file stores it in knots.
Daily temperature and rain chance are scored for the National Weather Service and for weather models, at one official airport in each television market. The high is the day’s high. The low is the climate-day low. Rain chance is the chance of measurable rain.
Rain, flood, snow, and severe-weather forecasts are scored only when a forecast and a later observation are both stored. The miss is the distance from the observation to the middle of the forecast range. If the observation falls inside that range, it counts as a hit. A flood damage map of warning areas and local storm reports is a map. It is not, by itself, a model score.
After a severe storm or a flood ends, the page counts how many stored local storm reports fell inside that day's outlook, by risk level, and how many fell outside. The number is kept. A later update does not replace it. If the archived outlook cannot be fetched, the page says so and does not invent a score.
Television stations are not graded. A meteorologist is scored only after they opt in, on public forecasts they post or share, and this version does not publish a meteorologist board.
How a forecast is paired
A hurricane forecast is paired with the National Hurricane Center’s operational best track at the same valid time. The position is not guessed between fixes. A model forecast is included only when the National Hurricane Center’s own official forecast from that same issue and that same lead is also on file, so the comparison uses the same storms and the same hours.
A daily temperature or rain-chance score pairs a model with the National Weather Service forecast from the same issue date, and with a final observation for that climate day. A day missing any of those three is left out. The climate day is midnight to midnight, local standard time.
A rain, flood, snow, or severe forecast is paired with the observation for that same place and that same report. If only the forecast is stored, or only the observation is stored, that row is not a score.
How far ahead
The hurricane score table uses forecasts made 24, 48, 72, and 120 hours ahead. The map can also draw a 96-hour point when the deck has a paired fix. That 96-hour point stays off the score table.
On a storm's own page, a 12-hour, 36-hour, or 60-hour forecast is shown once that time has passed and the best track has a fix then. The count beside it is how many forecasts were checked. A model is included even when that cycle has no official forecast yet. The overall rank still uses 24, 48, 72, and 120 hours.
The daily headline score uses days 1 through 5. Day 1 is the next calendar day. Days 6 and 7 are shown next to the National Weather Service and are not part of that headline score. Next day is day 1 only. Full week counts days 1 through 7, and only for models whose files cover the week. NAM and HRRR do not, so they stay on next day.
Miles
Path miss is statute miles on the score table and on the storm page. The file also keeps that distance in nautical miles. One nautical mile is 1,852 meters, about 1.15 statute miles. A smaller number is a closer forecast.
Ranks and overall rank
On the daily boards, a model is still collecting until it has 90 days scored in a market. At 90 days the rank is an early read. At 180 days it is proven. Fewer than 90 days is not a rank. A national row also needs a published rank in at least 30 markets. The National Weather Service is the baseline, so it is not ranked against itself. Lower miss is better. Beats the NWS means a smaller miss than the Weather Service on the same days.
Overall rank on a hurricane is the average of two ranks: path and wind. Each rank uses the graded leads at 24, 48, 72, and 120 hours, and each of those leads counts once, however many forecasts it holds. A lower miss takes the better place. A tie shares the average of the places those forecasts occupy. A source is ranked only with at least 3 graded forecasts for path and 3 for wind. Fewer than that says Not enough forecasts yet. A source with no graded path, or no graded wind, is left out of both ranks, and overall is left blank.
When more than one storm is on file, each storm is ranked on its own, then those ranks are averaged. Each storm counts the same, whether it had a few forecasts or many.
Where the data comes from
Hurricane positions and winds come from the National Hurricane Center. The operational best track is the b-deck. Model aids in the public a-deck are scored when that file stores them, including GFS, the GEFS mean, Canadian GEM, UKMET, HAFS-A, HAFS-B, HWRF, HMON, and NAVGEM. The post-season HURDAT2 track is not loaded. The best track on the site is preliminary and is revised after the season.
Google DeepMind and WeatherNext 3 are ensemble-mean tracks from Weather Lab. ECMWF AIFS, ECMWF HRES, and the ECMWF ENS mean are tropical-cyclone tracks from ECMWF open data. AI GFS is NOAA’s AIGFS tracker when that file is still posted. Daily temperature and rain for GFS, NAM, and HRRR come from NOAA open data. The National Weather Service point forecast is the daily baseline. Damage maps use National Weather Service local storm reports and warnings.
Known limits
- The operational best track can move after the season. This site does not swap in HURDAT2.
- Positions are not interpolated. A forecast with no best-track fix at that exact time is not scored.
- The AI GFS tracker directory keeps about two days. A cycle that has aged off is skipped.
- A rain, flood, snow, or severe score appears only when both the forecast and the observation are stored. An empty cell is not treated as zero.
- The severe and flood outlook score counts a report inside the drawn outline only. A report just outside the line is outside. It does not use a neighborhood around the line.
- Landfall is scored only when an explicit landfall row is stored. The track is not scanned for a coastline crossing.
- The daily low is the climate-day minimum. A television graphic often means the overnight low, which can fall on the next climate day. That overnight window is not what this version scores.
- Television stations are not graded, and no one pays for a place on a list.