Sun et al. (2025) Understanding the Decreased ENSO Predictability since the Early 2000s Based on Data-Driven and Dynamical Models

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

Identification

• Journal: Journal of Climate
• Year: 2025
• Date: 2025-12-18
• Authors: Qiming Sun, Jiye Wu, Jing‐Jia Luo, Fenghua Ling, Zijie Guo, Shunwu Zhou
• DOI: 10.1175/jcli-d-25-0086.1

Research Groups

Not explicitly detailed in the abstract.

Short Summary

This study evaluates the interdecadal change in El Niño–Southern Oscillation (ENSO) prediction skill over the past four decades using dynamical and deep learning models, revealing a significant decline since the early 2000s primarily due to worse prediction of ENSO phase transitions, weakened subsurface precursors, and increased bias in zonal advection.

Objective

• To investigate the change in ENSO prediction skill over the past four decades and identify the underlying possible causes using both dynamical and deep learning models.

Study Configuration

• Spatial Scale: Global, with a focus on the equatorial Pacific (western equatorial Pacific).
• Temporal Scale: Past four decades (seasonal–interannual), with specific periods of analysis including since the early 2000s and after approximately 2005, and specific months (June–August, May–July).

Methodology and Data

• Models used: Two dynamical models (with different data assimilation methods) and three deep learning (DL) models (with different architectures).
• Data sources: Five sets of hindcasts with maximum lead times exceeding 20 months. Evaluation implicitly uses observational or reanalysis data for comparison, though not explicitly listed as a data source in the abstract.

Main Results

• Deep learning models, despite high performance, exhibit interdecadal changes in ENSO forecast skill similar to dynamical models.
• ENSO prediction skill significantly decreased since the early 2000s, particularly during late spring and early summer when ENSO phase transitions occur.
• The primary cause of the overall decline in ENSO forecast skill in recent decades is the worse prediction of ENSO transitions.
• After approximately 2005, subsurface signals in the western equatorial Pacific during June–August of the preceding year are no longer effective precursors for predicting ENSO onset during May–July in both dynamical and DL models.
• Models display severe errors in predicting subsurface signals and their influence on surface temperature.
• The weakened influence of the subsurface temperature anomaly and model errors in its prediction jointly pose a significant challenge to ENSO forecasts after approximately 2005.
• Diagnostic analysis using the mixed-layer heat budget equation reveals that increased forecasting bias in the anomalous zonal advection of the mean sea surface temperature term in models after approximately 2005 is a key factor contributing to the decline in ENSO prediction skill.

Contributions

• Provides a comprehensive evaluation of interdecadal changes in ENSO prediction skill using a diverse set of dynamical and deep learning models.
• Identifies specific periods (since early 2000s, after ~2005) and mechanisms (ENSO transitions, weakened subsurface precursors, model errors in subsurface prediction, increased bias in zonal advection) contributing to the decline in ENSO predictability.
• Highlights new challenges for ENSO forecasts in the twenty-first century, offering insights that may help improve understanding and future prediction efforts.

Funding

Not explicitly detailed in the abstract.

Citation

@article{Sun2025Understanding,
  author = {Sun, Qiming and Wu, Jiye and Luo, Jing‐Jia and Ling, Fenghua and Guo, Zijie and Zhou, Shunwu},
  title = {Understanding the Decreased ENSO Predictability since the Early 2000s Based on Data-Driven and Dynamical Models},
  journal = {Journal of Climate},
  year = {2025},
  doi = {10.1175/jcli-d-25-0086.1},
  url = {https://doi.org/10.1175/jcli-d-25-0086.1}
}