Li et al. (2026) Improving seasonal prediction of global mean surface temperature by incorporating dynamic ENSO realistic forecasts

Identification

• Journal: npj Climate and Atmospheric Science
• Year: 2026
• Date: 2026-03-23
• Authors: Kexin Li, Fei Zheng
• DOI: 10.1038/s41612-026-01386-9

Research Groups

• State Key Laboratory of Earth System Numerical Modeling and Application, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
• University of Chinese Academy of Sciences, Beijing, China

Short Summary

This study identifies an underrepresented ENSO-driven pantropical coupling mechanism as a major source of error in autumn-initialized global mean surface temperature (GMST) predictions. By incorporating skillful ENSO realistic forecasts into a new dynamic-statistical framework, the reliable GMST prediction lead-time is extended from two to four months, reducing hindcast errors by an average of 41% during 1980–2024.

Objective

• To investigate the key factors contributing to autumn-initialized global mean surface temperature (GMST) prediction errors and improve forecast performance earlier.
• To develop a dynamic-statistical framework that incorporates skillful ENSO realistic forecasts to enhance the timeliness and accuracy of seasonal GMST predictions by better representing ENSO-driven pantropical coupling mechanisms.

Study Configuration

• Spatial Scale: Global, with a focus on pantropical regions including the tropical central-eastern Pacific (Niño3.4 region), tropical western-central Indian Ocean, and tropical North Atlantic (TNA).
• Temporal Scale: Seasonal predictions, specifically autumn-initialized forecasts (September, October) with lead times extended from two to four months. Hindcast period is 1980–2024, with observational data spanning 1850–2024 and climate indices from 1950–2024.

Methodology and Data

• Models used:
  • IAP GMST SEPS (Statistical Ensemble Prediction System at the Institute of Atmospheric Physics) as the baseline.
  • Hybrid dynamic-statistical prediction system (new framework) incorporating ENSO forecasts.
  • Multivariate linear regression model (for initial bias correction testing with Niño3.4, TNA, and IODMI).
  • Univariate linear regression model (for ENSO-only bias correction).
  • IAP ENSO ensemble prediction system (EPS) for Niño3.4 index forecasts.
  • North American Multimodel Ensemble (NMME) dynamical models (CanCM4i, CMC1-CanCM3, CMC2-CanCM4, GEM-NEMO, GFDL-SPEAR, NASA-GEOSS2S, NCAR-CESM1, NCEP-CFSv2) for comparison.
• Data sources:
  • Global surface temperature data: Hadley Center and Climate Research Unit analysis version 5 (HadCRUT5), NOAA Global Surface Temperature Version 5, and Berkeley Global Temperature Data.
  • Climate indices: Niño3.4 index (from U.S. National Weather Service Climate Prediction Center), Tropical North Atlantic (TNA) index (from NOAA), Indian Ocean Dipole Mode Index (IODMI) (from NOAA).
  • Niño3.4 index forecasts: IAP ENSO ensemble prediction system (EPS).
  • NMME 2-metre temperature data: IRI/LDEO Climate Data Library.

Main Results

• An underrepresented ENSO-driven pantropical coupling mechanism, linking ENSO evolution to tropical ocean–atmosphere interactions and a coherent autumn–winter global temperature response, was identified as a key contributor to autumn-initialized GMST prediction errors.
• Autumn-initiated GMST prediction errors showed significant positive correlations with tropical central-eastern Pacific sea surface temperature (SST) (r > 0.7) and the Niño3.4 index (r ~ 0.6).
• Prediction errors also exhibited correlations with the Indian Ocean Dipole Mode Index (IODMI) (r ~ 0.4–0.5) and the Tropical North Atlantic (TNA) index (r ~ 0.5–0.6).
• The newly developed hybrid dynamic-statistical framework, incorporating skillful ENSO realistic forecasts, extended reliable GMST prediction lead-time from two months to four months.
• The framework reduced hindcast errors by an average of 41% in 64% of years during 1980–2024, with particularly large improvements during El Niño years (85% reduction).
• For September-started forecasts, the Root Mean Square Error (RMSE) decreased to below 0.03 °C, meeting the threshold for high predictive accuracy.
• The ENSO-corrected IAP GMST SEPS achieved predictive skill comparable to the leading tier of the North American Multimodel Ensemble (NMME) dynamical models.

Contributions

• Identified a critical, previously underrepresented ENSO-driven pantropical coupling mechanism as a primary source of error in autumn-initialized GMST predictions.
• Developed a novel hybrid dynamic-statistical prediction system that effectively integrates dynamic ENSO forecasts with a statistical GMST prediction model.
• Significantly extended the reliable lead-time for seasonal GMST predictions from two months to four months.
• Achieved substantial reductions in GMST hindcast errors, particularly during ENSO-active periods, enhancing the practical value of seasonal climate forecasts.
• Demonstrated a computationally efficient and operationally flexible approach that maintains competitive predictive skill compared to complex fully dynamical climate models.
• Introduced an adaptive design using a 30-year sliding regression to account for decadal modulations in the ENSO–GMST relationship, improving robustness under evolving climate conditions.

Funding

• Jing-Jin-Ji Regional Integrated Environmental Improvement- National Science and Technology Major Project (Grant No. 2025ZD1201700)
• National Natural Science Foundation of China (Grant No. 42430114 and 42505145)
• China Postdoctoral Science Foundation (Certificate No. 2025M770300)
• Fundamental Research Funds of Institute of Atmospheric Physics, Chinese Academy of Sciences

Citation

@article{Li2026Improving,
  author = {Li, Kexin and Zheng, Fei},
  title = {Improving seasonal prediction of global mean surface temperature by incorporating dynamic ENSO realistic forecasts},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  doi = {10.1038/s41612-026-01386-9},
  url = {https://doi.org/10.1038/s41612-026-01386-9}
}