Rainformer

The rainformer is a pytorch-based encoder-forecaster model for precipitation nowcasting.

For more information or paper, please refer to Rainformer.

The short introduction of files

train_seq.npy & test_seq.npy: the files of defining the order of data.

tool.py: This file contains some preprocessing function, such as data transfer function, evaluate function, show picture function, etc.

Rainformer/Attention.py: This file implements the channel-attention module and spatial-attention module.

Rainformer/Rainformer.py: This file is the kernel file, it builds the whole model, contain the local-attention module and global-attention moudle and gate fusion module.

Rainfromer/SwinTransformer.py: This file implements the Swin-Transformer.

Rainformer/test.py & train.py: The former contains the test process of the model. The train.py contains the train process of the model.

Train

Firstly you should apply for the KNMI dataset, you can apply for the dataset by KNMI.

Then, you can use Rainformer/Rainformer/train.py to train your new model or load the pre-trained model.

Test

You can use Rainformer/Rainformer/test.py to test your model.

Environment

Python 3.6+, Pytorch 1.0 and Ubuntu.

Citation

@ARTICLE{9743916,
 author={Bai, Cong and Sun, Feng and Zhang, Jinglin and Song, Yi and Chen, Shengyong},
 journal={IEEE Geoscience and Remote Sensing Letters}, 
 title={Rainformer: Features Extraction Balanced Network for Radar-Based Precipitation Nowcasting}, 
 year={2022},
 volume={19},
 number={},
 pages={1-5},
 doi={10.1109/LGRS.2022.3162882}}