Dataset Card for TCBench

Dataset Details

Dataset Description

TCBench leverages existing datasets to create an integrated, value-added dataset spanning 1980-Present. Additionally, it provides tools to seamlessly integrate new data sources and pre-process them in a consistent matter across studies. It targets the development and refinement of data-driven models for tropical cyclone prediction, including precipitation downscaling and intensity predictions. Tropical cyclones are some of the highest-impact weather events in affected regions, and we thus strive to support the development of a homogenized, publicly available, database of high-impact events. Our efforts include the optimized use of multiple datasets by fusing reanalysis and observational datasets into a comprehensive benchmark dataset in an expandable framework for experimental flexibility.

We plan to include state-of-the art uncertainty quantification methods designed for AI models of geospatial data and, following best practices, to contain open-access, AI-ready datasets that follow FAIR principles and will provide clear evaluation protocols for AI models. Finally, tropical cyclone datasets often require complex pre-processing, hindering their use by AI experts without domain knowledge. By providing homogenized data and well-defined evaluation protocols, TCBench will advance the use of AI for Earth sciences. It will facilitate the application and evaluation of sophisticated AI frameworks for tropical cyclone predictions.

In summary, TCBench aims to provide opportunities to study the predictability of tropical cyclones (and changes in behavior associated with changing climate), as well as developing a dataset and evaluation tools that can be used freely by the scientific community.

• License: MIT

Dataset Sources [optional]

The data for TCBench is compiled from a variety of sources. The International Best Track Archive for Climate Stewardship (IBTrACS, Knapp et. al. (2010), Gahtan et. al. (2024)) is the most complete observational archive of global tropical cyclones, and serves as the ``ground truth'' for model evaluation. ERA5 reanalysis (Hersbach et al. (2020)) provides dozens of high-resolution meteorological variables used to initialize physical and neural weather models. ERA5 reanalysis is available through the European Centre for Medium-range Weather Forecasting (ECMWF). Hindcasts from physical weather models are obtained from The International Grand Global Ensemble (TIGGE) product (Bougeault et. al. ), which provides ensemble reforecasts of many physical weather prediction models, including the Global Ensemble Forecasting System (GEFS) and the International Forecast System (IFS). Finally, neural weather model forecasts were obtained from various sources, including NVIDIA's FourCastNetv2 (Bonev et. al. (2023)), Huawei's Pangu-Weather (Bi et al. (2023)), Google DeepMind's GenCast (Price et al. (2025)), and the ECMWF's AIFS v1.0 (Lang et. al. (2024)).

• Bi, K. L. Xie, H. Zhang, X. Chen, X. Gu, and Q. Tian, 2023: Accurate medium-range global weather forecasting with 3D neural networks. Nature, 619, 533-538. doi:10.1038/s41586-023-06185-3.
• Bonev, B., T. Kurth, C. Hundt, J. Pathak, M. Baust, K. Kashinak, and A. Anandkumar, 2023: Spherical Fourier Neural Operators: Learning Stable Dynamics on the Sphere. arXiv preprint: https://arxiv.org/abs/2306.03838.
• Bougeault, P., and coauthors, 2010: The THORPEX Interactive Grand Global Ensemble. Bull. Amer. Meteorol. Soc., 91, 1059-1072. doi:10.1175/2010BAMS2853.1.
• Gahtan, J., K.R. Knapp, C.J. III Schreck, H.J. Diamond, J.P. Kossin, and M.C. Kruk, 2024: International Best Track Archive for Climate Stewardship (IBTrACS) Project, Version 4.01. NOAA National Centers for Environmental Information, https://doi.org/10.25921/82ty-9e16.
• Hersbach, H., B. Bell, P. Berrisford, S. Hirahara, A. Hor{'a}nyi, J. Mu{~n}oz-Sabater, J. Nicolas, C. Peubey, R. Radu, D. Schepers, and coauthors, 2020: The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146, 1999-2049.
• Knapp, K. R., M. C. Kruk, D. H. Levinson, H. J. Diamond, and C. J. Neumann, 2010: The International Best Track Archive for Climate Stewardship (IBTrACS): Unifying tropical cyclone best track data. Bull. Amer. Meteor. Soc., 91, 363-376. doi:10.1175/2009BAMS2755.1
• Lang, S., M. Alexe, M. Chantry, J. Dramsch, F. Pinault, B. Raoult, M. Clare, C. Lessig, M. Maier-Gerber, L. Magnusson, and coauthors, 2024: AIFS--ECMWF's data-driven forecasting system. arXiv preprint arXiv:2406.01465.
• Price, I., A. Sanchez-Gonzalez, F. Alet, T.R. Andersson, A. El-Kadi, D. Masters, T. Ewalds, J. Stott, S. Mohamed, P. Battaglia, and coauthors, 2025: Probabilistic weather forecasting with machine learning. Nature, 637, 84-90.

Out-of-Scope Use

TCBench should not be used for general atmospheric prediction tasks, extra-tropical cyclones, El Niño prediction, nor should it be used as a production capability. TCBench is made to measure the performance of prediction models for tropical cyclones only. The impossibility of creating a single dataset that fits all purposes necessitates trade-offs, such as neglecting some teleconnections or using datasets only available for the past approximately 15 years.

Dataset Structure

matched_tracks
|-- (Tracks obtained from the neural weather model).csv
|-- 2023_aifs.csv
|-- ...

neural_weather_models
|-- weather_model_1/
|---- (Raw weather model outputs for when a storm was observed).nc
|---- ...
|-- ..

Dataset Creation

Data Collection and Processing

[More Information Needed]

Bias, Risks, and Limitations

TCBench represents a subset of all possible tropical cylone tracks and materialization conditions. It has global coverage, but not every region that can see a tropical cyclone has a materalized cyclone in the dataset. Therefore, as an evaluation, TCBench may be missing certain geographical areas or settings under which tropical cyclones could materialize.

Citation

@article{gomez2025tcbench,
  title={TCBench: A Benchmark for Tropical Cyclone Track and Intensity Forecasting at the Global Scale},
  author={...},
  journal={TBD},
  year={2025}
}

Dataset Card Authors

• Ritwik Gupta
• Milton Gomez
• Marie McGraw
• Ilia Azizi

Dataset Card Contact

Ritwik Gupta firstlast@berkeley.edu