––° ––mph ingest HOLD
GOES-12 visible satellite image of Hurricane Katrina on 28 August 2005

Katrina · GOES-12 · 28 Aug 2005

Home Research Hurricane Rapid Intensification

Research · North Atlantic · 1980–2024

Decoding hurricane
rapid intensification

Forty years of 6-hourly North Atlantic cyclone data, cleaned and modeled in a reproducible Python pipeline. Rapid intensification defined as a ≥30 kt wind-speed gain in 24 hours — a threshold that separates forecast-busting storms from everything else.

View on GitHub

40+

Years of record

1980–2024

≥30 kt

RI definition

per 24 hours

6 h

Cadence

HURDAT2 / IBTrACS

2

Datasets

cross-validated

The question

Is rapid intensification becoming more common in the North Atlantic, and can we see it early enough to matter?

Storm surge and wind exposure models depend on intensity forecasts 24–48 hours out. When a storm jumps two category levels in a day, every pre-positioned resource, every evacuation order, and every underwriting assumption based on the prior track goes stale at once. The goal here is empirical: establish the long-run RI frequency baseline, test whether the trend is real and significant, and characterize the synoptic environments in which onset occurs most reliably.

The answer matters beyond meteorology. Insurance pricing, reinsurance treaties, emergency operations, and infrastructure hardening all price in categorical forecasts. A systematic upward bias in RI frequency is a structural risk that current pricing cycles may not fully reflect.

Twenty-four hours

Maria · 17–18 Sep 2017
Terra MODIS satellite image of a disorganized Tropical Storm Maria on 17 September 2017
17 Sep 2017 · Terra MODIS Tropical storm · disorganized
Suomi NPP VIIRS satellite image of Hurricane Maria as a Category 4 on 18 September 2017
18 Sep 2017 · Suomi NPP VIIRS Category 4 · 24 hours later

This is the jump the pipeline is built to flag. Maria went from a ragged tropical storm to a major hurricane in one day — a ≥30 kt 24-hour wind-speed gain, visible from orbit as a closed eye and a symmetric canopy. NASA Earth Observatory / LANCE.

24-hour wind delta · approaching landfall

TanStack Charts · dashed rule is the ≥30 kt RI threshold. Illustrative profile.

The record

Schematic · HURDAT2
95W 85W 75W 65W 55W 45W 35W 25W 15W 45N 35N 25N 15N 5N
All named tracks
RI-experiencing tracks
RI onset location

Storm tracks are schematic representations of North Atlantic basin activity. RI onset points cluster in the western basin and Gulf of Mexico where warm SST anomalies and low vertical wind shear create favorable development environments.

Suomi NPP VIIRS satellite image of Hurricane Maria on 18 September 2017 after rapid intensification
Maria · Suomi NPP VIIRS · 18 Sep 2017 After the jump, Category 4 · NASA EO

Method

01

Data ingestion & cleaning

HURDAT2 best-track records (1980–2024) parsed from NHC fixed-width format into a normalized pandas DataFrame. Records with missing intensity observations flagged and forward-filled only where gaps are ≤1 step. IBTrACS v4 loaded as cross-validation reference.

02

RI flagging

For each 6-hour observation, compute the 24-hour forward wind-speed delta. Observations where Δv ≥ 30 kt tagged as RI onset. Consecutive RI flags collapsed to a single event to avoid double-counting sustained rapid deepening. Post-landfall points excluded.

03

Annual frequency & trend

RI event counts aggregated by season (June–November). Ordinary least-squares trend fit with 95% bootstrap confidence intervals. Mann–Kendall non-parametric test applied to confirm monotone direction without distributional assumptions.

04

Reproducibility

Full pipeline in a single Jupyter notebook: raw download → cleaned parquet → figures. Environment pinned via conda-lock. Figures generated with matplotlib; all seeds fixed. Any researcher can rerun end-to-end from the public HURDAT2 URL.

What the data shows

+47%

increase in annual RI events

From the 1980s decadal mean to the 2015–2024 period. The trend passes both OLS and Mann–Kendall tests at p < 0.05.

≈ 8 events/season

2020–2024 average

The five most recent seasons average nearly double the 1980–1989 baseline of 4.2 events per season.

Gulf of Mexico

highest RI concentration

Warm loop current waters and frequent low-shear synoptic patterns make the Gulf the most RI-prone sub-basin by onset density.

18–36 h

median forecast lead time

At which the NHC intensity guidance currently shows measurable RI skill. Before 18 hours, onset is largely indistinguishable from climatology.

Annual RI events · 1980–2024

Annual count
Linear trend
Recent seasons

Data: illustrative representation of HURDAT2-derived counts.

MODIS satellite image of Hurricane Ian approaching Florida on 28 September 2022
Ian · Terra/Aqua MODIS · 28 Sep 2022 Western basin, approaching the eastern Gulf · NASA

Ian on approach to the eastern Gulf — the same sub-basin where this record shows the densest RI onset. A closed eye and a symmetric canopy at this range is the satellite signature of a storm that has already taken the jump.

RI volume

Isometric · decade × basin

Mean RI events per season · three-dimensional

Left / right faces
Earlier decades
2020s top faces

Each column is a decade × sub-basin cell. Height is mean RI events per season. The Gulf stack is the tallest in every decade; the 2020s lift is visible as amber caps. TanStack Charts, isometric projection of the same illustrative counts.

Why it matters

Insurance & reinsurance

Pricing models calibrated on historical frequency assume a stationary distribution. If RI events are trending upward, expected loss curves understate tail risk. Re-calibration of exceedance probability tables is overdue.

Emergency management

Evacuation orders timed to Category 2 landfalls become dangerously conservative when a storm can become Category 4 in 24 hours. Improved RI predictability shortens the window between warning issuance and impact.

Infrastructure investment

Building codes and hardening standards often reference the 50- or 100-year storm. A rising RI baseline compresses the return period for extreme-intensity events at any given location.

Forecast model improvement

Identifying synoptic and oceanic precursors that reliably precede RI onset gives numerical weather prediction a target: resolve those precursors, and intensity forecast skill improves where it matters most.

Stack

Python 3.11 Runtime
pandas 2.x Data processing
scipy Statistics
matplotlib Figures
TanStack Charts Site figures
Jupyter Notebook

Data sources

HURDAT2 Primary
IBTrACS v4 Cross-validation
ERA5 SST Environmental
NCEP/NCAR R1 Reanalysis

Reproducibility

Pipeline is a single notebook from raw HURDAT2 download to final figures. No manual preprocessing steps.

Environment locked with conda-lock. All random seeds fixed. Re-runnable by any researcher with the public data URL.

johnathan-radojevich/atms517 →

Next project

Catching epidemic rapid growth →
View on GitHub