QiFeng-CYGNSS: A global kilometre-scale tropical cyclone inner-core vector wind field dataset from CYGNSS observations

QiFeng-CYGNSS is a global kilometre-scale tropical cyclone (TC) inner-core 10 m vector wind dataset reconstructed from CYGNSS satellite observations using a physics-guided score-based diffusion assimilation framework. The dataset is described in the companion data descriptor paper submitted to Earth System Science Data (Han et al., 2026a). The reconstruction methodology, validation, and ablation experiments are documented in the methodology preprint (Han et al., 2026b (https://arxiv.org/abs/2605.18477). Name origin The name QiFeng (Chinese: 栖风, qī fēng — literally "the wind at rest") evokes the Chinese poetic image of 使风栖定,令无形之风归于完整之形 — "letting the wind settle, so that the formless wind returns to a complete form". The name reflects the dataset's purpose: gathering sparse, direction-free CYGNSS scalar observations and letting them coalesce into a structured kilometre-scale vector wind field. Temporal coverage - January 2020 – September 2022. - Snapshots at every IBTrACS reporting time with available CYGNSS coverage (primarily 6-hourly at 00/06/12/18 UTC; some agencies provide 3-hourly reports). Spatial coverage - All six active global TC basins: North Atlantic (NA), Eastern Pacific (EP), Western Pacific (WP), North Indian (NI), South Indian (SI), South Pacific (SP). - Storm-relative Cartesian domain of 384 km × 384 km centred on each TC, on a 256 × 256 grid at 1.5 km grid spacing. Contents - 313 tropical cyclones identified by IBTrACS SID, 4955 reconstructed snapshots in total. - 1960 snapshots (39.6 %) pass the Observation Coverage Sufficiency (OCS) quality criterion (`meets_ocs`) and are recommended for quantitative analysis. - Pixel-level 16-member ensemble uncertainty for 138 major-hurricane snapshots. - Format: NetCDF-4 (CF-1.8 convention), one file per TC (e.g. `2022266N12294.nc`) plus a separate `ensemble_uncertainty.nc`. - Variables per file: `u10`, `v10` (10-m eastward / northward wind components, m s⁻¹), `time`, `sid`, `latitude`, `longitude`, `land_mask`, `center_lat`, `center_lon`, `ibt_vmax`, `n_obs`, `meets_ocs`, `meets_ocs_recomputed`, and CYGNSS observation wind speeds, acquisition coordinates and times, grid indices and observation errors. The archive README and `variables.csv` describe all fields. `ensemble_uncertainty.nc` stores `ws_spread(record, y, x)` with SID, time and a unique record ID. - Total volume: ~2.0 GB. Validation summary Independent validation against C-band SAR using the archived coverage flag (46 overpasses) yields a pooled pixel-level wind speed RMSE of 6.45 m s⁻¹ and a median per-overpass RMSE of 5.58 m s⁻¹. Airborne Tail Doppler Radar validation (23 paired snapshots) yields a pixel-level wind speed RMSE of 6.9 m s⁻¹, with V_max bias reduced by ~79 % versus ERA5 and ~75 % versus CCMP across the full sample. Files in this record - `QiFeng_output_dataset.zip` (~2.0 GB): the dataset itself, containing one NetCDF-4 file per TC (e.g. `2022266N12294.nc`) plus `ensemble_uncertainty.nc`. The archive includes a README, variable definitions, snapshot indices and reference-relative empirical error intervals with their calibration limits. - `visualize_dataset.ipynb`: a Jupyter notebook (Python 3, requires `xarray`, `numpy`, `matplotlib`) demonstrating how to load a per-TC NetCDF file, read the geographic grid coordinates, plot the wind speed field at peak intensity, and filter snapshots by the OCS quality flag. - `export_tiff.py`: a command-line script that batch-converts each NetCDF snapshot into a 3-band GeoTIFF (`u10`, `v10`, wind speed) projected to EPSG:4326 (WGS84). Suitable as forcing input for GIS workflows, storm-surge models, and wave models. Run `python export_tiff.py -i /path/to/dataset/ -o ./tiff_out/` (requires `rasterio`). Dataset tools are also available on GitHub at [github.com/Watanabeyouuu/QiFeng-CYGNSS-dataset-tools](https://github.com/Watanabeyouuu/QiFeng-CYGNSS-dataset-tools); this Zenodo record is the permanent archive. Licence Creative Commons Attribution 4.0 International (CC BY 4.0). Citation If you use this dataset, please cite both the data descriptor paper and this Zenodo record: Han, X., Li, X., Yang, J., Ni, H., Niu, Z., and Huang, W.: A global kilometre-scale tropical cyclone inner-core vector wind field dataset from CYGNSS observations, Earth System Science Data, in review, 2026. Han, X., Li, X., Yang, J., Niu, Z., Han, G., Wang, J., Huang, W., Zheng, Y., Ni, H., Wang, Y., Tao, W., Aouf, L., Peng, S., and Chen, D.: Global kilometre-scale tropical cyclone inner-core vector winds from sparse scalar CYGNSS observations, arXiv preprint, arXiv:2605.18477, https://doi.org/10.48550/arXiv.2605.18477, 2026. Han, X., Li, X., Yang, J., Ni, H., Niu, Z., and Huang, W.: QiFeng-CYGNSS: A Global Kilometre-Scale Tropical Cyclone Inner-Core Vector Wind Field Dataset [data set], Zenodo, https://doi.org/10.5281/zenodo.20046108, 2026. This release adds two-dimensional grid coordinates, land masks, CYGNSS observation locations and times, and empirical error intervals. Storm files are organized by IBTrACS SID, and ensemble spread can be selected by storm ID and time. The data-reading code has been updated.