Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 39 min ago
Peer-reviewedCC BYJournal of Geophysical Research: Machine Learning and Computation

Downscaling of Tropical Cyclone Surface Wind Fields With a Hybrid Attention Transformer

  • Chunhua Wang1
  • Dazhi Xi2
  • Na He3
  • Shuai Wang4
  • Kekuan Chu1
4 affiliations
  1. 1State Key Laboratory of Severe Weather Meteorological Science and Technology Nanjing University Nanjing China
  2. 2Department of Earth and Planetary Sciences The University of Hong Kong Hong Kong China
  3. 3Laboratory for Research and Application of Aviation Weather Forecasting and Service Technologies CAAC East China Regional Administration Shanghai China
  4. 4Department of Geography and Spatial Sciences University of Delaware Newark DE USA
Read full textPDFDOICite

Abstract

This study employs a Hybrid Attention Transformer‐based super‐resolution model to reconstruct tropical cyclone (TC) surface wind fields. The model downscales coarse‐resolution ERA5 reanalysis data to high‐resolution HWind analyses (1998–2013), increasing the horizontal resolution by a factor of five. First, this study assesses the impact of data set partitioning strategies and different meteorological input variables on model performance. The reconstructed TC surface wind fields and the radial profiles of azimuthally averaged surface winds agree well with the HWind analysis. Key TC metrics, including the maximum sustained wind speed (MSW) and the radius of 17 m s−1 winds (R17), show high correlations with HWind (0.81 and 0.92, respectively). When independently validated against IBTrACS records over the North Atlantic (1990–2020), the model maintains robust correlations, confirming its out‐of‐sample reliability across different periods and cases. The reconstructed past TC intensity and size metrics achieved by the developed model provide new opportunities for research, demonstrating the model's future potential in TC climatology and hazard assessment studies.

Related

  1. Added value of dynamical downscaling in sub-seasonal tropical cyclone forecast
    Taehyung Kim, Eunji Kim, Haerin Park +2
    arXivPreprintCC BY-NC-ND 4.0Full textAtmospheric science & meteorologyarXiv
  2. A Physics-Constrained Implicit Profile Network for Continuous Reconstruction of Tropical Cyclone Near-Surface Wind Profiles
    Jian Ma, Yilin Yang, Robert Rogers +2
    arXivPreprintCC BY-NC-SA 4.0Full textAtmospheric science & meteorologyarXiv
  3. Skillful Seasonal Tropical Cyclone Prediction with AI Weather--Climate Models Forced by Dynamically Predicted SSTs
    Hiroyuki Murakami, Baoqiang Xiang
    EarthArXivPreprintPublisher termsFull textAtmospheric science & meteorologyEarthArXiv
  4. Tibetan Plateau uplift increased western North Pacific tropical cyclone frequency
    Tingyu Zhang, Tianjun Zhou, Linqiang He +3
    Nature CommunicationsPeer-reviewedCC BY-NC-ND 4.0Full textClimate scienceCrossref
  5. Probabilistic assessment of tropical cyclone recovery burden across US Atlantic and Gulf Coast counties
    Simona Meiler, Nikola Blagojevic, Meredith Lochhead +2
    EGUspherePreprintCC BY 4.0Full textGeneral earth scienceEGUsphere

Cite

BibTeXRIS
BibTeX
@article{wang2026downscaling,
  title = {{Downscaling of Tropical Cyclone Surface Wind Fields With a Hybrid Attention Transformer}},
  author = {Chunhua Wang and Dazhi Xi and Na He and Shuai Wang and Kekuan Chu},
  journal = {Journal of Geophysical Research: Machine Learning and Computation},
  year = {2026},
  month = oct,
  doi = {10.1029/2025jh001170},
  url = {https://doi.org/10.1029/2025jh001170},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - Downscaling of Tropical Cyclone Surface Wind Fields With a Hybrid Attention Transformer
AU  - Chunhua Wang
AU  - Dazhi Xi
AU  - Na He
AU  - Shuai Wang
AU  - Kekuan Chu
PY  - 2026
DA  - 2026/10/28
JO  - Journal of Geophysical Research: Machine Learning and Computation
DO  - 10.1029/2025jh001170
UR  - https://doi.org/10.1029/2025jh001170
AB  - This study employs a Hybrid Attention Transformer‐based super‐resolution model to reconstruct tropical cyclone (TC) surface wind fields. The model downscales coarse‐resolution ERA5 reanalysis data to high‐resolution HWind analyses (1998–2013), increasing the horizontal resolution by a factor of five. First, this study assesses the impact of data set partitioning strategies and different meteorological input variables on model performance. The reconstructed TC surface wind fields and the radial profiles of azimuthally averaged surface winds agree well with the HWind analysis. Key TC metrics, including the maximum sustained wind speed (MSW) and the radius of 17 m s−1 winds (R17), show high correlations with HWind (0.81 and 0.92, respectively). When independently validated against IBTrACS records over the North Atlantic (1990–2020), the model maintains robust correlations, confirming its out‐of‐sample reliability across different periods and cases. The reconstructed past TC intensity and size metrics achieved by the developed model provide new opportunities for research, demonstrating the model's future potential in TC climatology and hazard assessment studies.
SN  - 2993-5210
ER  -