Liu et al. (2026) Deep learning trained on high-fidelity climate model simulations extends the skillful prediction of Northern Tropical Atlantic SST anomalies

Identification

• Journal: npj Climate and Atmospheric Science
• Year: 2026
• Date: 2026-09-16
• Authors: Ao Liu, Jinqing Zuo, Hui Gao, Steven C. Hardiman, Lijuan Chen, Junhu Zhao, Jiacan Yuan
• DOI: 10.1038/s41612-026-01546-x

Research Groups

• Chinese Academy of Meteorological Sciences
• State Key Laboratory of Climate System Prediction and Risk Management/China Meteorological Administration Key Laboratory for Climate Prediction Studies, National Climate Centre
• Department of Atmospheric and Oceanic Sciences and Institute of Atmospheric Sciences, Fudan University
• Met Office Hadley Centre

Short Summary

This study introduces a deep learning framework based on convolutional neural networks (CNNs) to predict Northern Tropical Atlantic SST anomalies. The CNN is trained on CMIP6 models that realistically capture the ENSO–NTA teleconnection, and it extends effective predictions to 8 lead months.

Objective

• To develop a physics-informed deep learning framework for predicting NTA SST anomalies.
• To investigate the impact of training dataset selection on the performance of the CNN.

Study Configuration

• Spatial Scale: Global climate model simulations (CMIP6) and regional analysis over the Northern Tropical Atlantic.
• Temporal Scale: Monthly time resolution, with a focus on seasonal to interannual timescales.

Methodology and Data

• Models used: CMIP6 models, convolutional neural networks (CNNs)
• Data sources: Historical simulations from CMIP6, observational data for validation

Main Results

• The CNN trained on selected CMIP6 models outperforms the baseline CNN and dynamical models in predicting NTA SST anomalies.
• The strong-teleconnection CNN extends effective predictions to 8 lead months, compared to 7 months for the MME.
• The CNN captures physically consistent precursors, with predictability governed by the interplay between ENSO forcing and local ocean–atmosphere variability.

Contributions

• This study provides a novel approach to improving seasonal prediction skills using physics-informed deep learning.
• The results highlight the importance of selecting CMIP6 models that realistically capture the ENSO–NTA teleconnection for training the CNN.

Funding

• National Natural Science Foundation of China (Grant No. 41875054)
• Chinese Academy of Meteorological Sciences Research Fund (Grant No. 2020KYYQ02)

Citation

@article{Liu2026Deep,
  author = {Liu, Ao and Zuo, Jinqing and Gao, Hui and Hardiman, Steven C. and Chen, Lijuan and Zhao, Junhu and Yuan, Jiacan},
  title = {Deep learning trained on high-fidelity climate model simulations extends the skillful prediction of Northern Tropical Atlantic SST anomalies},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  doi = {10.1038/s41612-026-01546-x},
  url = {https://doi.org/10.1038/s41612-026-01546-x}
}