TCDiff

This is the code of TCDiff, a Generative AI with Physics-Driven Meteorological Factor Constraint for improving Tropical Cyclone Forecasting.

Requirements

• python 3.8.8
• Pytorch 1.11.0 (GPU)

Data Preparation

First, we need to download all the data we used in TCDiff.

• TCDiff's processed datasets part1 and part2
• TCDiff's checkpoint

After completing the downloading, move these file to correct file path.

• Move TCDiff's processed datasets to /TCDiff, change the data_dir in TCDiff/configs/baseline.yaml to TCDiff/process_data_1D-ENV-ERA5-ALL-WE-NEED-vocen-vodis
• Move TCDiff's checkpoint to /TCDiff/experiment/TCDiff_ori

If you want to process the data yourself, the processing code of TC tilt metrics can be found in TCDiff/process_data_1D-ENV-ERA5-ALL-WE-NEED-vocen-vodis.py.

Train

## change the eval_mode in TCDiff/configs/baseline.yaml to False ##
cd TCDiff
python main.py

Test

## change the eval_mode in TCDiff/configs/baseline.yaml to True ##
cd TCDiff
python main.py

Acknowledgement

Part of our code is borrowed from MID. We thank the authors for releasing their code and models.