meheru/setcd-tropical-cyclone

SETCD-WP — Satellite–ERA5 Tropical Cyclone Dataset, Western Pacific subset

Storm-centred windows pairing GRIDSAT-B1 infrared brightness temperature with four co-located ERA5 reanalysis fields, indexed by IBTrACS best-track identity. This is the Western Pacific cross-basin subset (1978, 2019, 2020) used for the generalisation experiment in the thesis below. The full multi-year corpus is on IEEE DataPort.

Contents

File	What
setcd_wp_1978.tar, setcd_wp_2019.tar, setcd_wp_2020.tar	storm windows, one tar per year
era5_wp.tar	matching raw ERA5 extractions, before regridding
latents_mc_vae.tar	pre-computed MC-VAE latents for the corpus
metadata/	coordinate lookups, basin inventory, per-channel statistics
build_scripts/	the code that built this from the public sources

Layout inside the tars

setcd_wp/<year>/<storm SID>/<YYYY-MM-DD HH_MM_SS>/
    GRIDSAT_data.npy    (1, 256, 256) float32  -- IR brightness temperature, K
    ERA5_data.npy       (4, 256, 256) float32  -- u wind, v wind, temperature, pressure

Both arrays are already regridded to 256 x 256, so the five channels stack directly with np.concatenate; no resizing step is needed at load time. The native resolutions (GRIDSAT 286², ERA5 80²) survive only in era5_wp.tar and in the source archives. Note the very different per-channel scales -- brightness temperature is ~220-300 K while pressure is ~10⁵ Pa -- so normalise per channel using metadata/data_stats.txt. Cadence is 3-hourly, set by GRIDSAT-B1.

Usage

from huggingface_hub import hf_hub_download
import tarfile
p = hf_hub_download("meheru/setcd-tropical-cyclone", "setcd_wp_2020.tar",
                    repo_type="dataset")
tarfile.open(p).extractall("data", filter="data")

Or use download_setcd.py from the thesis archive, which handles checksums and places everything where the training code expects it.

Sources and licence

Derived from three openly licensed sources, all permitting redistribution:

• GRIDSAT-B1 — NOAA NCEI, public domain
• ERA5 — ECMWF / Copernicus Climate Change Service
• IBTrACS v04r01 — NOAA NCEI, public domain

Contains modified Copernicus Climate Change Service information (ERA5). Neither the European Commission nor ECMWF is responsible for any use of this information.

Released as CC-BY-4.0. Please cite the three sources above alongside this repo.

Related

• Model weights: https://huggingface.co/meheru/tropical_cyclone_forecasting
• Demo: https://huggingface.co/spaces/meheru/tropical_cyclone_forecasting

Undergraduate thesis, Department of CSE, Khulna University of Engineering & Technology. Meheru Zannat, supervised by Prof. Dr. Sk. Md. Masudul Ahsan.