← All calculations

Calculation report · proof of concept · SKD-POC-04

New Zealand cyclone wind: the 2023 season

No living mark for this run: it covers one season (two storm windows), and the mark needs at least three comparable seasons.

Peril
Tropical cyclone · wind
Source
ERA5 reanalysis, 10 m wind gust
Period
2023 season (Hale, Gabrielle), hourly
Backtest
Not scored against losses
Cells triggered24 / 11230 km hexagons over land
Season payoutEUR 2.55Mlimit EUR 250k per cell
Share of exposure paid9.1%one season, not a premium
Model to station gap10.03 km/hmean over 3 stations, max 13.7
hex_017_018 peak 131 km/h
T1 · 110 → 40% T2 · 125 → 100% 2023 · T2
Cell hex_017_018, the highest gust of the season (130.9 km/h). Hourly ERA5 gust during the analysed events. Ladder 110 / 125 km/h pays 40 / 100%.

Context

Peak 10 m gust per hex cell across Cyclones Hale and Gabrielle. Checked against the IBTrACS best-track archive rather than assumed, exactly two named storms came within 600 km of New Zealand in 2023: Hale in January, which passed within 2 km of the coastline, and Gabrielle in February, whose damage fell mainly on Hawke's Bay and Tairāwhiti. Both windows are analysed here.

The index is the highest 10 m gust each hex cell recorded during those events. Cells are hexagons of 30 km circumradius (60 km wide and 52 km tall), kept whole where their centre is on land. The payout question is about wind over land, so an ocean cell reading 200 km/h is a picture of the sea, not a claim.

This POC was originally scoped on a parametric windfield: cyclone best-track points driving a Holland vortex, the method used in production for Caribbean hurricane microinsurance. That does not survive contact with the New Zealand record. Of the 24 track points Gabrielle publishes across its approach, exactly two carry both an intensity and a radius-of-maximum-wind, both of them over 400 km offshore, and both reporting a radius of 333 km, which describes a sprawling extratropical field rather than a vortex that could be fitted. At every point where the storm was near enough to New Zealand to matter, both fields are absent, and only eight points in twenty-six years within 300 km of the coast carry both. The cause is institutional, not a data gap: the agencies that publish radius-of-maximum-wind have no responsibility for New Zealand, and Wellington, which holds that responsibility, has never published the field. No alternative feed fixes that, so the hazard is reanalysis instead.

The ladder attaches at 110 km/h and reaches full limit at 125, paying on 24 of 112 land hexes in this season. It replaced a first version set to the Australian/NZ cyclone category boundaries, which are defined on MEAN wind and were being applied to a gust index: 63 km/h is the Category 1 floor for a 10-minute mean, while gusts reach it in ordinary New Zealand weather. That ladder paid on 80 of 112 hexes. There are only two tiers because the index ceilings at 130.9 km/h (a 28 km cell average cannot reach what a headland reaches), leaving 21 km/h above attachment, against a measured 10 km/h gap between the model and real instruments. A third tier would be separated by less than that gap.

Where the method stops

Limits of this calculation.

  1. 01

    ERA5 is a backtest and pricing input. Final ERA5 runs 2-3 months behind and ERA5T about 5 days, so a live product would trigger on station gusts, never on this.

  2. 02

    Values are 28 km grid-cell averages. Exposed sites read higher: Auckland airport recorded 88.9 km/h during Gabrielle while exposed capes recorded 150-159 km/h.

  3. 03

    Peak over the season's analysed cyclone events, not an annual maximum. A winter storm can out-blow a cyclone here.

  4. 04

    No per-cell New Zealand cyclone loss series exists: insured losses are published per event, nationally, and cannot score a hex. Payout-versus-loss is recorded as not applicable rather than estimated.

Want this run on your own trigger?