reading-a-planet: rapid intensification before tropical-cyclone landfall, 1980–2024

Data for the paper Rapid Intensification Before Landfall Around the World, 1980–2024: How Often It Strikes, Where, and How Far the Best Track Can Predict It by Thanatchon Chongmankhong (read online · PDF). Every tropical-cyclone fix in six basins from IBTrACS v4 best tracks, with the predictors and the rapid intensification (RI, ≥30 kt in 24 h) outcome used in a two-stage pre-registered study, plus every first landfall, the sealed-test-season forecasts and the per-country results. Code: https://github.com/thanatchon36/reading-a-planet; archived on Zenodo: 10.5281/zenodo.24216355. Fitted models: thanatchon36/reading-a-planet.

All data are real. Nothing is synthetic or imputed beyond what each column's description says.

Independent personal research. Does not represent the views of any employer or institution.

Load

from datasets import load_dataset
fixes = load_dataset("thanatchon36/reading-a-planet", "fixes", split="train").to_pandas()

With pandas, read the CSV files with keep_default_na=False, na_values=[""]: the North Atlantic basin code NA is otherwise read as a missing value and the basin disappears. The Parquet files store it as text.

Files

path	content
parquet/fixes.parquet = data/global/points.csv.gz	Main table. 127,252 six-hourly fixes of 4,350 storms in six basins (NA, EP, WP, NI, SI, SP), 1980–2024, one row per fix with track, intensity, geography, predictors, outcome and split. eligible marks the 59,565 fixes that enter the forecast sample
parquet/storms.parquet = data/global/storms.csv.gz	one row per storm (4,350): season, lifetime maximum wind, basin where it first reached 35 kt (gen_basin; empty for the 463 storms that never did), whether it ever underwent RI
parquet/landfalls.parquet = data/global/landfalls.csv.gz	2,533 landfalls; first_in_country marks the 1,745 first landfalls per storm and country; RI_pre48 whether RI occurred within the 48 h before
parquet/test_predictions.parquet = results/part2/test_predictions.csv	the 3,627 forecasts of the sealed test seasons 2022–2024, with the climatology, basin model (M1, M1c) and pooled model (MGc) probabilities
parquet/landfall_by_country.parquet = results/part2/landfall_by_country.csv	share of first landfalls preceded by RI for the 20 country–basin pairs with at least 15 landfalls, with 95% bootstrap intervals
data/part1/	the development-basin (western North Pacific) tables of Part I: points, storms, landfalls, monthly and July–October Niño-3.4 (enso_*.csv) and the 1981–2010 ERSSTv5 climatology at storm positions (sst_clim_1981_2010.nc)
results/part1/, results/part2/	every result table and JSON behind the paper: model selection, validation and test skill, bootstrap intervals, permutation importance, partial dependence, landfall shares by decade and country, ENSO composites
logs/	test_opened_part{1,2}.log (when each set of test seasons was opened, once) and the data-preparation summaries
raw_sha256.txt	SHA-256 hashes of the raw inputs the tables were built from
SHA256SUMS	checksums of every file in this repository

Main columns (fixes)

column	meaning
track_id, season, basin, time	IBTrACS storm id, season, basin of the fix, time (UTC, synoptic hours)
lat, lon, lon360	position (degrees; longitude in −180..180 and 0..360)
status, wind, slp, sshs_cat	IBTrACS status, maximum sustained wind (kt, one-minute, US agencies), pressure (hPa), Saffir–Simpson category
V, V_m6, V_m12, V_p24	wind now, 6 h and 12 h before, 24 h after (kt)
dV6, dV12, dV24	intensity change over the past 6 and 12 h and the next 24 h (kt)
RI30	outcome: 1 if dV24 ≥ 30 kt
land, dist_land_km, water_0_24	centre over land; distance to the nearest coast (Natural Earth 1:50m); over water for the whole next 24 h
sst_clim	1981–2010 ERSSTv5 calendar-month climatological SST at the storm centre (°C)
speed_kmh, u_kmh, v_kmh, poleward_kmh	translation speed and its zonal, meridional and poleward components over the past 6 h
age_h, vmax_sofar, V_minus_vmax	hours since the first fix, lifetime maximum so far, current minus maximum so far
doy_sin, doy_cos, doyh_sin, doyh_cos, abs_lat	season phase (doyh_* shifted half a year in the Southern Hemisphere), absolute latitude
eligible	in the forecast sample: V ≥ 35 kt, V_p24 observed, over water from t to t+24 h, not extratropical or subtropical
split	train (seasons ≤ 2018), val (2019–2021), test (2022–2024)

Sources

• IBTrACS v4 (Knapp et al., 2010), US-agency subset (NHC for NA and EP, JTWC elsewhere), as distributed offline in huracanpy 1.5.0. Not verified against the NOAA NCEI master file.
• NOAA ERSSTv5 (Huang et al., 2017) monthly SST 1970–2021, via the pydata/xarray-data mirror.
• Natural Earth 1:50m land and admin-0 countries (public domain).

Tables are built by src/prepare.py and src/prepare_global.py in the code repository; make all regenerates every file here.

Licence

The derived tables and results are released under CC BY 4.0. The inputs remain under their providers' terms: IBTrACS and ERSSTv5 are NOAA products (please cite them as above); Natural Earth is public domain.

Citation

@misc{chongmankhong2026rapid,
  author = {Chongmankhong, Thanatchon},
  title  = {Rapid Intensification Before Landfall Around the World, 1980--2024: How Often It Strikes, Where, and How Far the Best Track Can Predict It},
  year   = {2026},
  note   = {Preprint},
  doi    = {10.5281/zenodo.24216355},
  url    = {https://thanatchon36.github.io/reading-a-planet/paper/}
}