Archive replay · live forecasts → CPHC
Every official forecast in this archive is paired with the track the storm actually took. Measured across all of them: two thirds of today's 24-hour forecasts put the storm's centre within 36 nautical miles (67 km) of where it ends up. In the era that produced Iniki, that circle was 78 nautical miles (144 km) — 2.2 times wider.
| Lead time | 1988-1999 | 2000s | 2010s | 2020s |
|---|---|---|---|---|
| 12 h | 41 nm | 36 nm | 26 nm | 21 nm |
| 24 h | 78 nm | 61 nm | 42 nm | 36 nm |
| 36 h | 117 nm | 89 nm | 57 nm | 48 nm |
| 48 h | 153 nm | 121 nm | 75 nm | 58 nm |
| 72 h | 220 nm | 178 nm | 112 nm | 90 nm |
Follow the 1988-1999 line to 24 hours, then find the same error on the 2020s line: you reach it near 64 hours. A family on Kauai in 1992 got one day of forecast at that accuracy. The same accuracy now arrives almost three days out — time to finish the water, the plywood and the plan rather than start them.
The improvement is not a press release; it is measured from 18,181 forecast points across 64 storms in this archive, and it matches the trend NHC publishes in its own verification. The replay of any storm here shows it plainly: watch Iniki's four surviving 1992 forecasts wander, then watch Lane's 56 advisories in 2018 hold their line.
A tighter centre-line is not a smaller storm. The two-thirds radius says nothing about wind extent, rain, or surge, which reach far outside it — Lane never made landfall and still dropped over a metre of rain on the Big Island. At three days the modern miss is still 90 nautical miles (167 km), which is why watches cover whole islands, not neighbourhoods. And these are the errors of this archive's 64 Hawaii-approaching storms (2,203 points in 1988-1999, 2,727 in 2020s), not NHC's official verification set.
When a storm is coming, act on the Central Pacific Hurricane Center's current forecast, not on any historical average.
Every OFCL forecast position in the archive is compared with the NOAA HURDAT2 best track interpolated to the same hour; the two-thirds radius is the 67th percentile of those errors, bucketed by era and lead time. It is the same table that draws this site's uncertainty shading, and the same construction NHC uses for the official cone. Eras with too few storms to trust are pooled; the build prints what it drops.