Why another best track
Because a best track is a judgment with no error bar
Every official best track records one number per fix — one agency's judgment, made with whatever that agency had at the time — and no uncertainty. Every climatology, trend and record built on it inherits that silence. The missing-intense audit showed where the judgments diverge from the imagery; the CBT is the constructive answer: for every hurricane, every 6 hours, the best estimate the evidence supports, and how sure we are.
How. At each 3-hourly step every voice is an observation of the true 1-minute wind, with its own bias and error: the climatlas satellite estimator and the ADT, each inverted through its measured response to aircraft truth (a single image under-reads a genuine Category 5 by ~16 kt; the inversion removes that, and widens the error bar where the sensor saturates), anchored to each basin's modern book and fused as one observation because their errors are correlated (ρ ); the USA best track, with an error measured basin by basin and era by era; and the WMO regional centre, with its own bias. A Kalman smoother with persistent intensity trends joins them; 300 sampled histories per storm give each lifetime maximum a distribution.
After HURSAT. The HURSAT imagery ends in 2016. From 2017 the satellite voice is rebuilt from the real-time era's archives, each calibrated on aircraft truth by the auto-intensity system: the climatlas estimator on GridSat-B1 (2017 on) and on NOAA's GMGSI mosaic (July 2021 on), the UW-CIMSS ADT and CIMSS SATCON, fused with their measured error correlations. Since v0.3 they are anchored to the same standard as 1978–2016 (the 2005–2016 book) through GridSat-B1, one archive that spans the break (see one instrument across the seam). The CBT's peaks there are calibrated on aircraft-measured peaks, re-run with the aircraft withheld. The satellite-only peaks are mapped onto GridSat's one-instrument record, which is how v0.1 treated 1978–2016. Neither is forced to agree with the book.
Category 4–5 hurricanes by era: the book, the CBT and the satellite alone
Global hurricanes, lifetime maximum ≥113 kt. CBT and objective bars are medians of 300 joint samples; whiskers 5–95%. * 2017 onward: real-time voices anchored to the 2005–2016 standard through GridSat-B1; CBT peaks calibrated on aircraft-measured peaks, satellite-only peaks on GridSat's one-instrument record; 2026 is the season in progress.
Category 4–5 hurricanes per year
Book (grey line) and the CBT median with its 5–95% range (blue band). The dashed rule marks the change of satellite archive after 2016; the last year is the season in progress.
One instrument across the seam
Were the 2017–2025 books written like the 2005–2016 books?
The 2016 line in the charts above is a change of instrument: HURSAT before, the real-time archives after, each side anchored to its own era's book. Measured that way, a change in how the books themselves are written across 2016 could not show. GridSat-B1, NOAA's intercalibrated ISCCP B1 record (the lineage HURSAT was cut from), runs unbroken across it. The climatlas estimator reads every storm in it from to ( images). It is calibrated once on aircraft truth (all years together), anchored once to the 2005–2016 book (v0.1's standard) and run alone through the same smoother. That measures 2017–2025 with the same ruler as 2005–2016. Since v0.3 the CBT's post-2016 voices are anchored to this standard too.
Category 4–5 hurricanes per year: the book and one satellite instrument
Grey: the book. Amber: GridSat-B1 alone, median with 5–95% band of 300 joint samples. Its calibration and peaks are fitted on 2005–2016 only, so those years agree by construction; the years either side are the test. Complete seasons only.
Which side moved? GridSat minus the book, year by year
Median of GridSat minus the book at hurricane-strength fixes, kt, one calibration for all years (years with at least 15 fixes). Dashed ticks: a new-generation geostationary satellite takes over the basin. A step at a tick points at the imagery; a step elsewhere points at the book.
The ruler over the whole record: GridSat and the book against aircraft flight-level winds
Atlantic and eastern Pacific hurricane times with a 700-hPa aircraft fix within 3 h. Median, by year, of GridSat (one calibration) minus 0.9 × the maximum flight-level wind (amber), and of the book minus the same (grey). Only years with at least 8 such times are shown. A line that holds level is a ruler that did not move.
How good was the book?
The official record's own error, basin by basin and era by era
A by-product the CBT needs and nobody has published this way: the error of the USA best track's intensities at hurricane strength, estimated from its disagreement with the aircraft-calibrated satellite voice (var(book − satellite) − var(satellite)), floored at 4 kt. Within 3 h of an aircraft fix it is set to 6 kt regardless. Read the 2017–2026 column as a lower bound: today's books are made from the same ADT and SATCON the satellite voice uses, so where the two agree their errors are partly shared, not independently confirmed. In those years the CBT follows the book closely, and the satellite-only objective reanalysis is the independent check.
Every hurricane
Look one up
Versions and roadmap
- v0.2 — 2017–2026 on GridSat-B1 (2017 to June 2021) and GMGSI (July 2021 on) with the CIMSS ADT and SATCON, calibrated on aircraft truth; 1978–2016 unchanged from v0.1.
- v0.3 — one satellite scale: the estimator on GridSat-B1 from 2005 to the present anchors the post-2016 voices to the 2005–2016 standard, and on its own tests the post-2016 books against it.
- v0.4 (this release) — one instrument, 1980–2026: GridSat re-reads the HURSAT years and reproduces v0.1 era by era, and the aircraft's own flight-level winds show the post-2016 step is in the imagery, not the books.
- Aircraft as winds, not just times — flight-level and SFMR winds from the f-decks as their own observations, and the reduction from 700-hPa flight level made explicit.
- Real time — the objective CBT run forward on every storm, beside the official intensity, on the 2026 Intensity Watch; when the post-season books arrive, the difference is scored.
- Sharper eyes — 2-km GOES-R/Himawari imagery and 4-km AVHRR passes for pinhole eyes, where the 8-km record is a floor.