Whole-season view — every storm's track at once, colored by peak intensity. Click any track to open that storm.

Live forecast mode. The gold marker replays the storm's whole life on one timeline — drag the slider or hit play. The past leg is JMA best-track (position, wind & the real storm-force wind radius per observation, via Digital Typhoon); from now on it's the JMA forecast — growing wind-radius circle, expected wind & Dvorak number, with violet probability circles (position uncertainty). Nothing here is simulated.

Track, intensity, and wind-radius data: IBTrACS v04r01 (NOAA NCEI), Western Pacific basin. JTWC stops issuing on a system that dies over land, so a storm that crosses China and re-forms is filed by IBTrACS as two records; this archive rejoins the halves when they share a JTWC ATCF id, which across all 82 seasons happens once — Saudel 2026 (WP172026), one storm in CMA’s record too, filed as 2618. The overland stretch such a break leaves is filled from the agencies that kept fixing the depression after JTWC stopped, each supplying only what the one before it could not: first the CMA National Meteorological Center track service, which over the Chinese mainland is both the responsible agency and the detailed one — for Saudel it carries the storm 3-hourly across Fujian, Jiangxi and Guangdong — and then JMA’s analyses (via NII’s Digital Typhoon archive), which resume over Zhanjiang on the Leizhou peninsula where CMA’s stop. Borrowed points say CMA fix or JMA fix on hover; their winds are unit-converted and nothing more — CMA reports a 2-minute mean, JMA a 10-minute one — so they stay out of peak wind and ACE, which remain on the 1-minute JTWC basis the rest of the archive uses. Together they cut Saudel’s unfixed stretch from 78 hours to 15. What is left, where no agency has a fix at all, is drawn as a dotted bridge rather than track, because a straight line between two distant fixes is an artefact of joining them, not an observed path. Storms marked LIVE are fetched straight from NOAA NCEI's IBTrACS active-storms feed in your browser, then topped up from the JTWC working best track itself (the ATCF b-deck, openly mirrored by UCAR/RAL, updated several times a day) — so an active typhoon's history runs to the latest fix, and its 50/64 kt radii appear the moment JTWC publishes them (the working best track reports the 34 kt radii first; 50/64 kt usually arrive in post-analysis, which is also why 2025-onward storms show only Level 7 (34 kt) circles for now). Taiwan (CWA) classes use CWA's official 10-minute thresholds (17.2 / 32.7 / 51.0 m/s) applied to JMA's real 10-minute wind analysis for each point (carried in IBTrACS as TOKYO_WIND — the same averaging convention CWA uses, so storms class the way CWA's own record does, e.g. Doksuri 2023 as mostly 中度 rather than the inflated 強烈 a converted 1-minute wind gives); points between JMA's 6-hourly analyses interpolate, and only storms with no 10-minute record at all fall back to 0.88-converted 1-minute winds. CWA's and JMA's own analyses can still differ slightly — both are real agency data. Wind circle radii (level 7/10/12 — gale/storm/hurricane-force, JTWC's 34/50/64kt radii, per quadrant) are JTWC (USA agency) reports, which is why coverage starts in 2001 — systematic quadrant reporting didn't exist before then, for any storm in this archive, so it can't be extended back to 1985. Radius shapes interpolate between the four reported quadrant extents; treat them as an approximation, not a precise wind-field boundary. The Dvorak T-number is estimated from each point's wind speed via the standard Dvorak CI-number/wind-speed table — a standard approximation, not an independent satellite reading. ACE, rapid intensification (≥30 kt/24 h), and the season climatology are all computed from this same best-track wind record. The forecast cone is a persistence baseline — it simply projects the storm's current heading and speed forward, with a cone sized to typical average track error; it is the naive benchmark real forecasts are measured against, shown to contrast with the actual track, not a skilful prediction. ENSO tags use real NOAA CPC ASO ONI values.