Jo et al. (2026) Explainable Deep Learning Reveals Future Changes in East Asian Extreme Winter Predictability

Identification

• Journal: npj Climate and Atmospheric Science
• Year: 2026
• Date: 2026-10-05
• Authors: Seo‐Young Jo, Jeong-Hwan Kim, Daehyun Kang, Jae-Heung Park, Seung‐Ki Min, Minju Kim, Mi‐Kyung Sung
• DOI: 10.1038/s41612-026-01564-9

Research Groups

• Center for Climate and Carbon Cycle Research, Korea Institute of Science and Technology (KIST)
• Division of Environmental Science and Engineering, Pohang University of Science and Technology (POSTECH)
• School of Earth and Environmental Sciences, Seoul National University (SNU)

Short Summary

This study applies a deep learning-based framework to investigate the seasonal predictability of extreme winters in East Asia. The results show that the model exhibits reliable and competitive prediction skills relative to physics-based and few-shot learning frameworks.

Objective

• Assess the ability of a convolutional neural network (CNN) to skillfully predict both mean and extreme temperatures of East Asian winter climate.
• Investigate how prediction skill may differ between current and future climates.
• Understand the drivers of these changes using explainable artificial intelligence (XAI) approaches.

Study Configuration

• Spatial Scale: Regional, focusing on East Asia (30°–50°N and 100°–140°E).
• Temporal Scale: Seasonal, predicting December–January–February (DJF) mean surface air temperature.

Methodology and Data

• Models used: Convolutional Neural Network (CNN)
• Data sources:
  • CESM2 Large Ensemble (CESM2-LE) simulations for historical and SSP3-7.0 scenarios.
  • Amplitude Focal Loss (AFL) optimization method.

Main Results

• The CNN model attains reliable and competitive prediction skills relative to physics-based SMYLE prediction system and the DL-based few-shot learning framework of Oh and Ham18.
• Prediction skill increases in the future climate, with warm winter extremes becoming more predictable while cold extremes show little change.
• The nonlinear and asymmetric behavior between tropical Pacific SST anomalies and their contributions to East Asian winter temperature prediction is consistent with the state-dependent response of tropical convection to SST anomalies.

Contributions

• This study provides a useful tool for exploring how seasonal predictability mechanisms evolve under climate change.
• The results highlight the importance of understanding the physical basis of DL model predictions and whether the learned sources of predictability remain valid under changing climate conditions.

Funding

• Not specified in the provided text.

Citation

@article{Jo2026Explainable,
  author = {Jo, Seo‐Young and Kim, Jeong-Hwan and Kang, Daehyun and Park, Jae-Heung and Min, Seung‐Ki and Kim, Minju and Sung, Mi‐Kyung},
  title = {Explainable Deep Learning Reveals Future Changes in East Asian Extreme Winter Predictability},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  doi = {10.1038/s41612-026-01564-9},
  url = {https://doi.org/10.1038/s41612-026-01564-9}
}