https://github.com/HFAiLab/OpenCastKit

The open-source solutions of FourCastNet and GraphCast
https://github.com/HFAiLab/OpenCastKit

foucastnet weather-forecast

Last synced: 11 days ago
JSON representation

The open-source solutions of FourCastNet and GraphCast

• Host: GitHub
• URL: https://github.com/HFAiLab/OpenCastKit
• Owner: HFAiLab
• License: mit
• Created: 2022-06-09T11:21:34.000Z (over 2 years ago)
• Default Branch: master
• Last Pushed: 2023-04-18T02:57:00.000Z (over 1 year ago)
• Last Synced: 2024-05-20T07:48:08.847Z (6 months ago)
• Topics: foucastnet, weather-forecast
• Language: Python
• Homepage:
• Size: 3.38 MB
• Stars: 274
• Watchers: 6
• Forks: 71
• Open Issues: 24
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-WeatherAI - OpenCastKit - High-Flyer
• open-sustainable-technology - OpenCastKit - Open-source solutions of global data-driven high-resolution weather forecasting. (Atmosphere / Meteorological Observation and Forecast)

README

        
# OpenCastKit: an open-source solutions of global data-driven high-resolution weather forecasting

简体中文 | [English](README_en.md)

本项目是由幻方AI团队复现优化，并开源的全球AI气象预报模型工具库。基于 [FourCastNet](https://arxiv.org/abs/2202.11214) 和 [GraphCast](https://arxiv.org/abs/2212.12794) 的论文，我们构建了一个新的全球AI气象预报项目——**OpenCastKit**，它能够与欧洲中期天气预报中心（ECMWF）的传统物理模型——高分辨率综合预测系统（IFS），进行直接比较。

我们将基于1979年1月到2022年12月的ERA5数据训练出来的模型参数开源到 [Hugging Face 仓库](https://huggingface.co/hf-ai/OpenCastKit)中，并上线了一个每日更新的 [HF-Earth](https://www.high-flyer.cn/hf-earth/)，展示模型的预测效果。

下面是一些预测案例：

![台风路径预测与真实路径比较](./img/wind_small.gif)

![汽水浓度预测与真实情况比较](./img/precipitation_small.gif)

## 依赖

- [hfai](https://doc.hfai.high-flyer.cn/index.html) >= 7.9.5

- torch >=1.8

## 训练

原始数据来自欧洲中期天气预报中心（ECMWF）提供的一个公开可用的综合数据集 [ERA5](https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5) ，需要通过 `data_factory/convert_ear5_hourly.py` 脚本提出数据特征，转化为[高性能训练样本格式 FFRecord](https://www.high-flyer.cn/blog/ffrecord/) 下的样本数据。

### 训练 FourCastNet

本地运行：

```shell

   python train_fourcastnet.py --pretrain-epochs 100 --fintune-epochs 40 --batch-size 4

```

也可以提交任务至幻方萤火集群，使用96张A100进行数据并行训练

```shell

   hfai python train_fourcastnet.py --pretrain-epochs 100 --fintune-epochs 40 --batch-size 4 -- -n 12 --name train_fourcastnet

```

### 训练 GraphCast

本地运行：

```shell

   python train_graphcast.py --epochs 200 --batch-size 2

```

也可以提交任务至幻方萤火集群，使用256张A100进行流水线并行训练

```shell

   hfai python train_graphcast.py --epochs 200 --batch-size 2 -- -n 32 --name train_graphcast

```

## 引用

```bibtex

@article{pathak2022fourcastnet,

  title={Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators},

  author={Pathak, Jaideep and Subramanian, Shashank and Harrington, Peter and Raja, Sanjeev and Chattopadhyay, Ashesh and Mardani, Morteza and Kurth, Thorsten and Hall, David and Li, Zongyi and Azizzadenesheli, Kamyar and others},

  journal={arXiv preprint arXiv:2202.11214},

  year={2022}

}

```

```bibtex

@article{remi2022graphcast,

  title={GraphCast: Learning skillful medium-range global weather forecasting},

  author={Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Alexander Pritzel, Suman Ravuri, Timo Ewalds, Ferran Alet, Zach Eaton-Rosen, Weihua Hu, Alexander Merose, Stephan Hoyer, George Holland, Jacklynn Stott, Oriol Vinyals, Shakir Mohamed, Peter Battaglia},

  journal={arXiv preprint arXiv:2212.12794},

  year={2022}

}

```