Vanuatu cyclone trigger

The framework in development defines three triggers, all at ≥64 kt (10-minute) / ≥15,000 people exposed: a readiness trigger (72–120 h leadtime, RSMC Nadi extended track forecast, Shefa/Sanma/Tafea), an action trigger (24–72 h, VMHS track forecast, same provinces), and an observational trigger (no leadtime, VMHS observed track, all of Vanuatu). This tab explores how forecast-trigger thresholds behave on the observed record; the forecast-check tab tests the forecast legs against historical forecasts and shows the observational trigger's national scope.

How to use: set a wind-exposure and/or rainfall threshold, choose whether both (AND) or either (OR) must be met, and pick the wind speed. Storms meeting the rule are triggered (bold labels). Aim for the target number of triggers — staying at or below it — while maximising the total people affected by the triggered storms. Red labels are storms that received a CERF allocation; dashed lines are your thresholds; the shaded region is the trigger zone.

The 64 kt column is read on the 10-minute scale: swaths are rebuilt from the best-track quadrant radii at the ≈73 kt 1-min-equivalent contour (factor 0.88 — see the forecast tab's convention note). The 34/50 kt columns are 1-minute values from the corrected exposure table (adm2_usaradii_exp_dedup.parquet: no boundary double-counting, no dateline tearing). All exposure counts population in Shefa, Sanma and Tafea only. Rainfall is the maximum 2-day rolling mean of IMERG daily rain, averaged over the country. Impact (Total Affected) is from EM-DAT. The return period is Weibull: (20 + 1) / activations. The record is 2006–2025: IBTrACS 64 kt wind radii only exist reliably from 2005 (2003–04 storms, including Ivy 2004 — a direct hit on the AOI at 90 kt — carry none), and 2005 itself is excluded because its forecast archive is a single anomalous year (full JTWC decks survive for 2005 alone among 2003–2011, provenance unclear).

Forecast exposure is computed from JTWC official forecasts in the VMGD historical archive: the 34/50/64 kt quadrant wind radii from each forecast cycle are swept along the forecast track into a wind swath, and WorldPop population inside it is summed.

The two columns mirror the framework's trigger table. The forecast column corresponds to the action trigger: exposure counted in Shefa, Sanma and Tafea only (population —). The observed column corresponds to the observational trigger: exposure counted over all of Vanuatu (population —) — a storm striking Torba or Penama shows its real observed exposure and can activate the framework through the observational leg even though the forecast legs do not count those provinces. The exposure method and population raster are otherwise identical.

Wind-averaging convention: the trigger's 64 kt is read on the 10-minute scale used by RSMC Nadi and VMHS (severe tropical cyclone, Category 3 on the Australian scale). JTWC and IBTrACS-USA radii are 1-minute means, so every 64 kt number on this page — forecast action windows and observed exposure alike — is evaluated at the ≈73 kt 1-min-equivalent contour (factor 0.88; per-quadrant log-linear extrapolation of the R50→R64 decay, gated on intensity). The 34/50 kt values remain 1-minute context. Under the milder 0.93 conversion (69 kt equivalent) the activating set is unchanged and exposures sit between these and the raw 1-minute values (Judy 143k).

Against the framework as written: the forecast column evaluates the action leg only, from each cycle's 24–72 h track segment. The readiness leg (72–120 h) is deliberately not assessed — historical 64 kt wind radii at those leadtimes are too sparse to score it fairly. And the framework's forecast sources are the RSMC Nadi extended track (readiness) and the VMHS track forecast (action), but neither carries historical wind radii — JTWC official forecasts are the proxy here.

VMGD's own forecast tracks carry no wind radii — their Forecast Track Maps give position, intensity category and a position-accuracy figure only, with the wind boundaries drawn on the map image. They are shown here as tracks for context, but they cannot produce an exposure number.

Each storm's peak value is the highest exposure across all of its forecast cycles — i.e. would the trigger have fired at any point during the event. Invest-numbered decks (90–99) are excluded because they duplicate the named storm.

Two artifacts in the original per-adm2 exposure table on blob are corrected here (and in adm2_usaradii_exp_dedup.parquet): per-adm2 all_touched clips counted shared boundary pixels more than once (+23% on AOI totals), and swaths crossing the 180° meridian were torn by the coordinate conversion into shapes that spuriously covered Vanuatu — Fiji storms like Tomas, Yasa, Evan and Keni showed large exposure they never had. All buffers here are handled in a longitude-wrapped frame, so figures read substantially lower than the uncorrected ones.