Nonstationary Intensification of Global Tropical Cyclone Rainfall Extremes
Per-cell nonstationary extreme-value analysis of tropical-cyclone rainfall across six ocean basins, 1989–2024. M.S. thesis; manuscript in revision for Geophysical Research Letters.
Code and data →My M.S. thesis at Georgia Tech, advised by Dr. Ali Sarhadi (Climate Risk & Extreme Dynamics Lab). The manuscript, Yoo & Sarhadi (2026), is in revision for Geophysical Research Letters.
Question
Are the rarest tropical-cyclone rainfall events getting more frequent, and where? Standard intensity–duration–frequency estimates assume a stationary climate; this work asks what changes when that assumption is dropped.
Approach
- A per-storm tropical-cyclone rainfall dataset covering six basins (3,055 storms, 1989–2024), built by attributing gridded 0.25° precipitation to tracked storms.
- Per-cell Bayesian extreme-value analysis on event-level and annual-maximum rainfall, fitting an ensemble of stationary and nonstationary distributions whose parameters vary with time and with physical covariates such as column water vapor, potential intensity, and ENSO state.
- Out-of-sample scoring with threshold-weighted CRPS to weight the ensemble toward the upper tail, with trend significance assessed by block-bootstrap Mann–Kendall tests.
- Return-level and return-period products, driver attribution, and a stationary versus nonstationary classification for every cell.
The derived datasets and analysis code are archived on Zenodo (CC-BY-4.0) under the link above. The thesis itself is under a one-year embargo and will be released in August 2027.