FourCastNet Hurricane Intensity Estimation

This repository contains code and resources for estimating hurricane intensity using FourCastNet, a deep learning model trained on atmospheric reanalysis data. The project supports both ERA5 and MERRA reanalysis datasets.

Project Structure

.
├── example/                      # Example code and notebooks
│   ├── run_inference.ipynb      # Jupyter notebook for running inference
│   └── src.py                   # Source code for the model
├── files/                       # Sample data files
│   ├── ERA_*.nc                # ERA5 reanalysis data
│   ├── MERRA_*.nc              # MERRA reanalysis data
│   ├── *_ERA.csv               # ERA5 intensity data
│   ├── *_MERRA.csv             # MERRA intensity data
│   └── *_HURDAT.csv            # HURDAT2 intensity data
├── ERA_scaling/                 # ERA5 data scaling parameters
├── MERRA_scaling/              # MERRA data scaling parameters
├── bias_correction_checkpoints_ERA/    # ERA5 model checkpoints
└── bias_correction_checkpoints_MERRA/  # MERRA model checkpoints

Features

• Hurricane intensity estimation using FourCastNet
• Support for both ERA5 and MERRA reanalysis datasets
• Bias correction for improved accuracy
• Example notebooks for running inference
• Pre-trained model checkpoints

Requirements

The project requires Python 3.8+ and the following key dependencies:

Core Scientific Computing

• NumPy (>=1.21.0): For numerical computations and array operations
• SciPy (>=1.7.0): For scientific computing and signal processing
• Pandas (>=1.3.0): For data manipulation and analysis

Deep Learning

• PyTorch (>=1.9.0): Deep learning framework
• PyTorch Geometric (>=2.0.0): For graph neural networks

Data Processing

• Xarray (>=0.19.0): For handling labeled multi-dimensional arrays
• netCDF4 (>=1.5.7): For reading/writing NetCDF files

Machine Learning

• scikit-learn (>=0.24.0): For machine learning utilities and metrics

Visualization

• Matplotlib (>=3.4.0): For plotting and visualization

Development Tools

• tqdm (>=4.62.0): For progress bars
• glob2 (>=0.7): For file path operations
• Jupyter (>=1.0.0): For interactive development
• Notebook (>=6.4.0): For running Jupyter notebooks
• ipykernel (>=6.0.0): For Jupyter kernel support

You can install all dependencies using:

pip install -r requirements.txt

Getting Started

1. Clone the repository:

git clone [repository-url]
cd FourCastNet_Hurricane_Intensity_Estimation

2. Install dependencies (requirements.txt to be added)

3. Run the example notebook:

jupyter notebook example/run_inference.ipynb

Data

The project uses two main reanalysis datasets:

• ERA5: European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis
• MERRA: Modern-Era Retrospective Analysis for Research and Applications

Sample data files are provided in the files/ directory for testing and demonstration purposes.

Model Checkpoints

Pre-trained model checkpoints are available in:

• bias_correction_checkpoints_ERA/ for ERA5 models
• bias_correction_checkpoints_MERRA/ for MERRA models

License

This project is licensed under the terms of the included LICENSE file.

Citation

If you use this code in your research, please cite:

Ankur, K. (2025). FourCastNet Hurricane Intensity Estimation (v1.0) [Software]. Zenodo. https://doi.org/10.5281/zenodo.12345678

Contact

Ankur Kumar ankurk017@gmail.com ankur.lumar@nasa.gov