Home

Awesome

XiHe-GlobalOceanForecasting

This is the official repository for the Xihe papers.

XiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting, arXiv preprint arXiv:2402.02995, 2024.

by Xiang Wang, Renzhi Wang, Ningzi Hu, Pinqiang Wang, Peng Huo, Guihua Wang, Huizan Wang, Senzhang Wang, Junxing Zhu, Jianbo Xu, Jun Yin, Senliang Bao, Ciqiang Luo, Ziqing Zu, Yi Han, Weimin Zhang, Kaijun Ren, Kefeng Deng, Junqiang Song

Resources including pre-trained models, and inference code are released here.

Installation

The downloaded files shall be organized as the following hierarchy:

├── root
│   ├── input_data
│   │   ├── input_surface_data
│   │   │	├── input_surface_20190101.npy
│   │   ├── input_deep_data
│   │   │	├── input_deep_20190101.npy
│   ├── output_data
│   ├── models
│   |   ├── xihe_1to22_1day.onnx
│   |   ├── ...
│   |   ├── xihe_1to22_10day.onnx
│   |   ├── xihe_23to33_1day.onnx
│   |   ├── ...
│   |   ├── xihe_23to33_10day.onnx
│   ├── src
│   |   ├── data.yaml
│   |   ├── normalize_mean_50.npz
│   |   ├── normalize_std_50.npz
│   |   ├── mask_surface.npy
│   |   ├── mask_deep.npy
│   |   ├── mercator_lat.npy
│   |   ├── mercator_lon.npy
│   |   ├── data_process.py
│   |   ├── inference.py

First install the packaged virtual environment pycdo（Baidu netdisk or Google netdisk）and activate the virtual environment.

News and Updates

Global ocean forecasting (inference) using the trained models

Downloading trained models

Please download the layers 1to22 and 23to33 pre-trained models for 1 to 10 days from Baidu netdisk or Google netdisk.

These models are stored using the ONNX format, and thus can be used via different languages such as Python, C++, C#, Java, etc.

Input data

We support GLORYS12 reanalysis as initial fields, please prepare the input data using numpy and transformer it into a .npy file using the netCDF4 package.

There are two input data files that shall be put under the input_data/input_surface_data and input_data/input_deep_data, which stores the input data for 1 to 22 layers and 23 to 33 layers respectively. The specific details of layers and variables can be found in the following text.

We provide an example of preprocessed input data, input_surface_20190101.npy and input_deep_20190101.npy, which correspond to the daily means of GLORYS12 reanalysis data at 2019/01/01. Please download them from Baidu netdisk or Google netdisk.

Inference

Running the following command, one can get the 7-day ocean forecast in the output_data folder.

python inference.py --lead_day 7 --save_path output_data

Data description

Data sources

We use three data to train Xihe:

The main training data of Xihe is the GLORYS12 reanalysis，which is from 1993 to 2019. We also support other ocean data as initial fields, but the actual depth of the input data layers needs to correspond to the above selected depth.

Input data

In both cases, the dimensions of 2041 and 4320 represent the size along the latitude and longitude, where the numerical range is [-80,90] degree and [-180,180] degree, respectively, and the spacing is 1/12 degrees. For each 2041x4320 slice, the data format is exactly the same as the .nc file download from the GLORYS12 reanalysis official website.

Note that the NumPy arrays should be in single precision (.astype(np.float32)).

Output data

The model predicts 6 ocean variables. There are 23 layers including: ocean temperature, salinity, zonal and meridional components of ocean current (i.e., layer 1: 0.49m, layer 3: 2.65m, layer 5: 5.08m, layer 7: 7.93m, layer 9: 11.41m, layer 11: 15.81m, layer 13: 21.60m, layer 15: 29.44m, layer 17: 40.34m, layer 19: 55.76m, layer 21: 77.85m, layer 22: 92.32m, layer 23: 109.73m, layer 23: 109.73m. The 24th layer: 130.67m, the 25th layer: 155.85m, the 26th layer: 186.13m, the 27th layer: 222.48m, the 28th layer: 266.04m, the 29th layer: 318.31m, the 30th layer: 380.21m, the 31st layer: 453.94m, the 32nd layer: 541.09m and the 33rd layer: 643.57m)，sea surface height above geoid and sea surface temperature.

References

If you use the resource in your research, please cite our paper:

@article{wang2024xihe,
  title={XiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting},
  author={Wang, Xiang and Wang, Renzhi and Hu, Ningzi and Wang, Pinqiang and Huo, Peng and Wang, Guihua and Wang, Huizan and Wang, Sengzhang and Zhu, Junxing and Xu, Jianbo and others},
  journal={arXiv preprint arXiv:2402.02995},
  year={2024}
}