Atlantic hurricane best tracks

Available since v0.12.0. The manifest requests the complete Atlantic HURDAT2 best-track file through noaa:hurdat2. The National Hurricane Center publishes one fixed-format text file per basin and revises it after each season, so the manifest names a basin and nothing else: dates and geographic filters are rejected rather than silently ignored.

In an activated Python 3.11+ virtual environment, install the package and save the manifest as dataset.yaml in a working directory. Run the commands from that directory:

python -m pip install "usdata[pandas]"
usdata pull dataset.yaml
usdata verify dataset.yaml

The Atlantic file is about 7 MB of text; the Pacific file (basin: pacific) is about 4 MB. Open it locally with the HURDAT2 reader, which item.open() selects without arguments. Run this with python in the same directory and environment:

from pathlib import Path

from usdata import pull, verify

manifest = Path("dataset.yaml")
result = pull(manifest)
(item,) = result.fetched
tracks = item.open()
landfalls = tracks[(tracks.record_identifier == "L") & (tracks.status == "HU")]
print(len(tracks), "track points;", len(landfalls), "hurricane-intensity landfall records")
print(tracks.loc[tracks.max_wind_kt.idxmax(), ["storm_id", "name", "time", "max_wind_kt"]])
assert verify(manifest) == []

For a source installation, run from examples/hurdat2/ and use uv run usdata and uv run python.

The reader returns one row per best-track point: storm_id, name, UTC time, record_identifier, status, signed latitude/longitude, max_wind_kt, min_pressure_mb, twelve wind-radii columns, and max_wind_radius_nm. The documented missing sentinels become NaN (-999, and -99 where a maximum wind was left unassigned on a non-developing depression). Wind radii are absent before 2004, and the radius of maximum wind is mostly NaN before 2021: the Atlantic file back-fills it for a few hundred earlier points, chiefly reanalyzed landfalls, while the Pacific file has none. A landfall row is not one landfall event per storm, and a track point is not an observation: HURDAT2 is a post-season reanalysis whose early decades undercount storms and underestimate intensities. Compare eras with that in mind.

A repeat pull uses the lockfile and checksummed cache. Keep the manifest, lockfile, and cached bytes for reproducibility: the NHC replaces this file with a newly named revision each season, and an old checksum cannot recreate a file that is no longer published. Use pull(manifest, force=True) when deliberately moving to the current revision. See NOAA access notes for revision selection, basin naming, and format limits.