This is the official repository for the paper Benchmarking Tropical Cyclone Rapid Intensification with Satellite Images and Attention-based Deep Models. You can find the code to reproduce all the experiment results, including 4 ablation study models for rapid intensification prediction.
Model | vanilla | CCA | SSA | CCA + SSA |
---|---|---|---|---|
PR-AUC | 0.951 | 0.963 | 0.951 | 0.961 |
Heidke skill score | 0.159 | 0.164 | 0.161 | 0.152 |
You can install the recommended environment as follows:
conda env create -f env.yml -n cyclone
The pretrained model weights needed to be combined as follows:
cd ./pretrained_models
chmod +x combine.sh
./combine.sh
The data needs to first be downloaded from here.
The data paths in config.yaml
needs to then be updated according to the path of the data.
The config files are located in the specific model directories in ./pretrained_models
.
To train the model, run
python train.py --exp [config.yaml]
The default config files can be found in the individual pretrained model directories in ./pretrained_models
.
To evaluate model performance, run
python predict.py --model_dir [pretrained_model_dir] --models [model_1,model_2,...,model_n]
For instance to evaluate the pretrained models, run
python predict.py --model_dir ./pretrained_models --models ConvLSTM,ConvLSTM_CCA,ConvLSTM_SSA,ConvLSTM_CCA_SSA
Please cite our work if you use this repo.
@inproceedings{bai2020tcri,
author = {Ching-Yuan Bai and Buo-Fu Chen and Hsuan-Tien Lin},
title = {Benchmarking Tropical Cyclone Rapid Intensification
with Satellite Images and Attention-based Deep
Models},
booktitle = {Proceedings of the European Conference on
Machine Learning and Principles and Practice of
Knowledge Discovery in Databases (ECML/PKDD)},
month = sep,
year = 2020,
data = {http://www.csie.ntu.edu.tw/~htlin/program/TCRISI},
pdf = {http://www.csie.ntu.edu.tw/~htlin/paper/doc/ecml20tcrisi.pdf},
preliminary = {A preliminary version appeared in the Workshop on
Machine Learning for Earth Observation @ ECML/PKDD
'19.}
}