Code for RGCN-PSA

Paper: Cross-Region Graph Convolutional Network with Periodicity Shift Adaptation for Wide-area SST Prediction

Authors:
Han Peng, Wengen Li*, Chang Jin, Yichao Zhang, Jihong Guan*, Hanchen Yang, and Shuigeng Zhou (*Corresponding authors)

Affiliation:
Department of Computer Science and Technology, Tongji University, China; Department of Computing, The Hong Kong Polytechnic University, China; Shanghai Key Lab of Intelligent Information Processing and School of Computer Science, Fudan University, China

---

Table of Contents

• Code for RGCN-PSA
  • Table of Contents
    • Quick Start
    • Dependencies
    • Data Preparation
    • Code Structure
    • Citation

---

Quick Start

# Clone the repository
git clone https://github.com/ADMIS-TONGJI/RGCN-PSA.git

# Install dependencies
pip install -r requirements.txt

# Run North Pacific dataset
python main_NP.py

# Run South Atlantic dataset
python main_SA.py

---

Dependencies

• Python 3.8.19
• PyTorch 2.3.1
• Key libraries: NumPy, SciPy, OpenCV, etc.
  See full requirements in requirements.txt.

---

Data Preparation

• Public Data (Raw SST Data)
  Download sea surface temperature (SST) data from NOAA's Optimum Interpolation SST dataset:
  🔗 https://www.ncei.noaa.gov/products/optimum-interpolation-sst

• Preprocessed Data (Ready-to-Use)
  Download our processed datasets from Google Drive:
  🔗 https://drive.google.com/drive/folders/1RO91kEj2geNtqnF9KtPRODsNjoR9Qus_

---

Code Structure

.
├── RGCN_PSA_NORTH_PACIFIC_1.conf
├── RGCN_PSA_SOUTH_ATLANTIC_1.conf
├── data
│   ├── NORTH_PACIFIC_1
│   │   └── NORTH_PACIFIC_1.npz
│   └── SOUTH_ATLANTIC_1
│       └── SOUTH_ATLANTIC_1.npz
├── lib
│   ├── dataloader.py
│   ├── logger.py
│   └── metrics.py
├── main_NP.py
├── main_SA.py
├── model
│   ├── GCN.py
│   ├── GCRAN_TS.py
│   ├── RAA.py
│   ├── SelfAttention.py
│   ├── TSAttention.py
│   ├── TimeEncoder.py
│   └── train.py
├── readme.md
└── requirements.txt

Citation

If this code aids your research, please cite our paper:

@article{peng2025cross,
  title={Cross-Region Graph Convolutional Network with Periodicity Shift Adaptation for Wide-area SST Prediction},
  author={Peng, Han and Li, Wengen and Jin, Chang and Zhang, Yichao and Guan, Jihong and Yang, Hanchen and Zhou, Shuigeng},
  journal={ACM Transactions on Intelligent Systems and Technology},
  year={2025},
  publisher={ACM New York, NY}
}