Auckland: the Anniversary Weekend flood, 27 January 2023

Context

Peak 72-hour rainfall above each cell's own seasonal normal, across the largest insured weather loss in New Zealand's history.

On the evening of 27 January 2023 Auckland recorded 280 mm of rain in under 24 hours and 211 mm in under six, at Albert Park. NIWA assessed it as at least a 1-in-200-year rainfall. It produced 59,189 insurance claims worth NZ$1.98 billion (ICNZ, to 1 March 2024), the largest insured weather loss in the country's record.

This is PLUVIAL flood: rain falling faster than drainage can carry it away. It is not river flooding, and the two are deliberately not mixed. Cyclone Gabrielle's Hawke's Bay flooding three weeks later was fluvial, a different hazard variable with a different index.

The hazard is IMERG satellite rainfall, chosen on evidence rather than convenience. ERA5 was tested and never ranks this event first in its own year: zero of 35 cells around Auckland put it first, and every one of them ranked Cyclone Gabrielle above it, inverting the loss ordering. CHIRPS cannot express a sub-daily extreme at all: its daily values are a five-day total redistributed by cloud duration. At the cell over central Auckland IMERG ranks the event first at every window tested, as do about two in three cells around the city, and it puts its peak intensity at 18:30 that evening, about ninety minutes after the documented 5 p.m. onset of the flooding.

The index is a departure from normal, not a raw depth. Each IMERG grid cell is compared against its OWN 72-hour normal for that day of the year, built from 15 years of record pooled over a fifteen-day window, which gives 225 samples per calendar day. Thresholds are per grid cell too, fitted to that cell's own annual maxima, so the same rainfall is ordinary in a wet place and exceptional in a dry one. The payout units are 269 hexes of 10 km circumradius, kept whole where their centre is on land. A hex takes its wettest grid cell against the mean of its cells' thresholds, so a hex reaches its first tier in about one year in six rather than one in ten.

The thresholds are fitted on 2008-2022, so any cost quoted from those same years is in-sample and flatters a fitted tail: the ladder already knows how big the biggest years were. IMERG's record starts in 1998, so the ladder was re-scored on 1998-2007, a decade neither the climatology nor the thresholds have ever seen, with nothing refitted. It costs 5.7% of exposure out of sample against 5.9% in sample, a ratio of 0.97. The pricing is therefore not an artefact of the years it was built on, which is the one thing an in-sample number cannot tell you.