pdf.
Download required data and checkpoints from GoogleDrive. (When applying for permission, could you please briefly explain your occupation such as student, corporate employee, research institution staff, etc., and the purpose of using the model? Thank you for your cooperation!)
Store files according to the following structure:
. ├── 2017-12-31T18:00:00.npy ├── 2018-01-01T00:00:00.npy ├── 2018-01-01T06:00:00.npy ├── BoostEns.py ├── climatology-2018-01-01T06:00:00.npy ├── data_mean.npy ├── data_std.npy ├── diffusion_max.npy ├── diffusion_min.npy ├── max_logvar.npy ├── min_logvar.npy ├── model │ ├── attend.py │ └── denoising_diffusion_pytorch.py ├── model_d.onnx ├── model_g.pth ├── pic.py ├── run.py └── utils.py
Run run.py on GPU device.
For the input and output of the model, their dimensions are [B, C, H, W], where C=69, each channel corresponds to a weather variable, and their correspondence is shown in the variable-order.
If it runs correctly, you will see the following picture in the current directory.
@article{xu2024extremecast,
title={ExtremeCast: Boosting Extreme Value Prediction for Global Weather Forecast},
author={Xu, Wanghan and Chen, Kang and Han, Tao and Chen, Hao and Ouyang, Wanli and Bai, Lei},
journal={arXiv preprint arXiv:2402.01295},
year={2024}
}