Extreme-wind statistics toolkit

Problem

Offshore sites in typhoon regions need design wind speeds for rare events, the winds expected once in 50 or 100 years, which measured records are too short to show directly. An older R implementation existed but had known bugs.

What I did

I led development of the toolkit and wrote most of it, with co-developers in a shared repository. It builds typhoon wind fields with the Holland (1980) parametric model from IBTrACS storm tracks, with selectable parameterisations for the radius of maximum wind and the profile shape, and estimates return-period winds with extreme-value methods including block maxima. To migrate safely off the R code we kept a “legacy” pipeline that reproduces the old output, known bugs included, so the new code could be checked line by line before switching to corrected physics.

Result

Return-period wind estimates for offshore sites from one tested toolkit, and a migration that could be verified against the old results before anything changed.

Tools

Python · IBTrACS · extreme-value statistics