TropiCycloneNet-Dataset

Overview

This project introduces the details of the TropiCycloneNet Dataset ($TCN_D$), a comprehensive dataset for studying tropical cyclones (TCs). It includes $Data_{1d}$, $Data_{3d}$, and environmental data (Env-Data) collected for various tropical cyclones in the major oceanic regions from 1950 to 2023. The dataset is aimed at providing valuable insights for tropical cyclone research and predictive modeling.

Download

We offer two download options for the TropiCycloneNet Dataset ($TCN_D$):

• Full Dataset: Contains tropical cyclone data from 1950 to 2023, across six major oceans. The data size is approximately 25.7 GB.

  • Full Dataset

• Test Subset: A smaller subset of data from 2017 to 2023, intended for testing purposes. The data size is approximately 3.34 GB.

  • Test Subset

More download options will be added in the future, such as downloading by ocean region, by year, etc.

Check Data

We provide code to read and visualize $Data_{1d}$, $Data_{3d}$, and Env-Data. Researchers can flexibly use our dataset and visualize different types of data using the provided scripts.

Steps to Use the Dataset

1. Download and Extract the Dataset:

   • Download and extract the $TCN_D$ dataset to your desired path (dataset_path).

2. Set Up the Environment:

   • Install Python 3.7 and the necessary dependencies:

     pip install netCDF4==1.5.8
     pip install matplotlib==3.5.3
     pip install pandas==1.1.1
     pip install numpy==1.19.0

3. Run the Code:

   • After setting up the environment, run the read_TCND.py script:

     python read_TCND.py dataset_path TC_name TC_date area train_val_test

     Here:

     • dataset_path refers to the path where the dataset is located.
     • TC_name is the name of the tropical cyclone you wish to examine.
     • TC_date is the specific date and time of the cyclone in YYYYMMDDHH format.
     • area specifies the ocean region where the cyclone occurred (e.g., WP for Western Pacific, EP for Eastern Pacific, etc.).
     • train_val_test indicates whether the queried typhoon belongs to training, validation, or test set (train or val or test).

   • After running the script, you will find visualized images of Data1D, Data3D, and Env-Data in the current directory. The images will be named Data1D.png, Data3D.png, and Env-Data.png.

Visualization Example

Visualizing All Data

1. Get Details for $Data_{3d}$:

   The 3D data covers the 20° x 20° region around the tropical cyclone's center. The spatial resolution is 0.25°, and the temporal resolution is 6 hours. We collect Geopotential Height (GPH), U-component of wind, and V-component of wind at 200 hPa, 500 hPa, 850 hPa, and 925 hPa pressure levels. Sea Surface Temperature (SST) data is also included in the Data3D set.

2. Example of $Data_{1d}$, $Data_{3d}$, and Env-Data: The following command visualizes the tropical cyclone data for a specific time (2001101418 for Haiyan in the Western Pacific region):

   python read_TCND.py dataset_path Haiyan 2001101418 WP train

   After running the script, you will see the corresponding cyclone Data1D, Data3D, and Env-Data visualizations.

   Examples of $Data_{3d}$:

   

   We crop the data covering a 20° x 20° region around the TC center. The spatial resolution is 0.25°, and the time resolution is 6 hours. We collect Geopotential Height (GPH), U-component of wind, and V-component of wind at 200 hPa, 500 hPa, 850 hPa, and 925 hPa pressure levels. Sea Surface Temperature (SST) data is also included in the Data3D set.

   Examples of $Data_{1d}$:

   

   The bolded content in the first row of the figure represents some information about the typhoon Haiyan at 2001101418 that we want to examine.

   • ID: Time step of the TC.
   • LONG: Longitude of the TC center (with a precision of 0.1°E).
   • LAT: Latitude of the TC center (with a precision of 0.1°N).
   • PRES: Minimum pressure (hPa) near the TC center.
   • WND: Two-minute mean maximum sustained wind (MSW; m/s) near the TC center.
   • YYYYMMDDHH: Date and time (UTC) of the TC event.
   • Name: Name of the TC.

   The Data1D is normalized using specific rules to make it suitable for deep learning (DL) methods to extract useful information.

   Examples of Env-Data:

   

   The Env-Data includes the following attributes:

   • Movement Velocity: The movement velocity of the tropical cyclone.
   • Month: Month of occurrence.
   • Location Longitude and Latitude: The relative location on Earth.
   • 24-hour History of Direction: The movement direction of the cyclone in the past 24 hours.
   • 24-hour History of Intensity Change: The intensity change of the cyclone in the past 24 hours.
   • Subtropical High Region: Extracted from 500 hPa Geopotential Height (GPH) data. This variable is processed to make the data more suitable for input to DL models.

Additional Experiments

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

We would like to acknowledge the support of the research community and the institutions that contributed to the development of this dataset. The TropiCycloneNet Dataset has been designed for academic and research purposes in tropical cyclone studies.

---

Feel free to reach out with any questions or comments regarding the TropiCycloneNet Dataset or how to use this project.

Citing TropiCycloneNet

@article{TropiCycloneNet_under_review,
author = {Huang, Cheng and Mu, Pan and Zhang, Jinglin and Chan, Sixian and Zhang Shiqi and Yan, Hanting and Chen, Shengyong and Bai, Cong},
title = {TropiCycloneNet: A Benchmark Dataset and A Deep Learning Method for Global Tropical Cyclone Forecasting},
journal = {Nature Communications},
volume = {under_review},
number = {under_review},
pages = {under_review},
doi = {under_review},
url = {under_review},
year = {under_review}
}