Cui et al. (2026) ENSO‐CausalNet: Integrating Causal Inference Into Deep Learning for Robust ENSO Prediction

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Geophysical Research Letters
• Year: 2026
• Date: 2026-01-05
• Authors: Yuehan Cui, Bin Mu, Shijin Yuan, Bo Qin
• DOI: 10.1029/2025gl118701

Research Groups

Not available in the provided abstract.

Short Summary

This study proposes a paradigm integrating causal inference into data-driven modeling, developing ENSO-CausalNet to achieve skillful El Niño-Southern Oscillation (ENSO) prediction up to 22 months ahead, revealing varying dominant physical processes and causal pathways driving ENSO variability.

Objective

• To integrate causal inference into data-driven modeling to enable climate predictions based on genuine causal relationships, specifically for El Niño-Southern Oscillation (ENSO), and to elucidate the distinct causal pathways and varying dominant physical processes affecting ENSO variability with lead time.

Study Configuration

• Spatial Scale: Global to basin-scale, focusing on interactions within the extratropical Pacific, Atlantic, and Indian Oceans relevant to ENSO.
• Temporal Scale: Prediction lead times up to 22 months; analysis of ENSO variability (interannual to decadal).

Methodology and Data

• Models used: ENSO-CausalNet, a deep learning model incorporating causal inference.
• Data sources: Not explicitly detailed in abstract, but implied to be historical climate observations and/or reanalysis data related to ocean-atmosphere interactions.

Main Results

• ENSO-CausalNet achieves skillful ENSO prediction of the Niño 3.4 index up to 22 months ahead.
• Dominant physical processes affecting ENSO are found to vary significantly with lead time.
• Distinct causal pathways are elucidated, showing how Bjerknes feedback and extratropical Pacific, Atlantic, and Indian Ocean air-sea interactions drive ENSO variability.
• Increasing input dimensionality in the model may lead to the learning of incomplete causal relationships, resulting in degraded prediction skill.
• Forecast capability is critically dependent on a comprehensive causal understanding, thereby confirming the model's physical validity.

Contributions

• Proposes a novel paradigm for climate prediction that integrates causal inference into data-driven modeling, moving beyond traditional correlation-based approaches.
• Develops ENSO-CausalNet, a specific implementation demonstrating skillful long-lead (up to 22 months) ENSO prediction based on genuine causal relationships.
• Provides mechanistic insights into ENSO dynamics by identifying varying dominant physical processes and distinct causal pathways (e.g., Bjerknes feedback, extratropical ocean air-sea interactions) at different lead times.
• Highlights the critical importance of comprehensive causal understanding for robust forecast capability and physical validity in climate models.
• Offers a robust prediction paradigm extensible to other climate systems, facilitating mechanistic analysis and scientific discovery.

Funding

Not available in the provided abstract.

Citation

@article{Cui2026ENSOCausalNet,
  author = {Cui, Yuehan and Mu, Bin and Yuan, Shijin and Qin, Bo},
  title = {ENSO‐CausalNet: Integrating Causal Inference Into Deep Learning for Robust ENSO Prediction},
  journal = {Geophysical Research Letters},
  year = {2026},
  doi = {10.1029/2025gl118701},
  url = {https://doi.org/10.1029/2025gl118701}
}