Groom et al. (2025) Entropic Learning Enables Skilful Forecasts of ENSO Phase at up to 2 Years Lead Time

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

Identification

• Journal: Journal of Advances in Modeling Earth Systems
• Year: 2025
• Date: 2025-12-29
• Authors: Michael Groom, Davide Bassetti, Illia Horenko, Terence J. O'Kane
• DOI: 10.1029/2025ms005128

Research Groups

Not explicitly stated in the abstract, but the work builds on previous research by Groom et al. and benchmarks against the International Research Institute for Climate and Society (IRI).

Short Summary

This paper extends the entropy-optimal Sparse Probabilistic Approximation (eSPA) algorithm with an ensemble meta-learning strategy to predict ENSO phase using only satellite-era observational data. The enhanced eSPA model achieves probabilistic forecast skill comparable to the IRI plume, extends accurate lead times up to 24 months, and operates at a significantly lower computational cost.

Objective

• To extend the application of the entropy-optimal Sparse Probabilistic Approximation (eSPA) algorithm, incorporating an ensemble meta-learning strategy, to predict ENSO phase, defined by thresholding the Niño3.4 index.

Study Configuration

• Spatial Scale: Global sea surface temperature, tropical Pacific wind stresses, and the Niño3.4 region (5°N–5°S, 170°W–120°W).
• Temporal Scale: Satellite-era observational data for training and validation; retrospective forecasts from 2012 to 2022; lead times up to 24 months.

Methodology and Data

• Models used: Entropy-optimal Sparse Probabilistic Approximation (eSPA) algorithm, an ensemble approach aggregating multiple eSPA models via a novel meta-learning strategy.
• Data sources: Satellite-era observational data sets. Features include:
  • Leading principal components from a delay-embedded EOF analysis of global sea surface temperature.
  • Vertical temperature gradient (as a thermocline proxy).
  • Tropical Pacific wind stresses.
  • Data processed to prevent information leakage from the future.

Main Results

• The eSPA model successfully avoids overfitting despite a limited number of training instances.
• It produces probabilistic ENSO forecasts with skill comparable to the International Research Institute for Climate and Society (IRI) ENSO prediction plume.
• Beyond IRI's typical lead times, eSPA maintains skill out to 20 months for the ranked probability skill score and 24 months for accuracy and area under the ROC curve.
• The computational cost is a fraction of that required by fully coupled dynamical models.
• Successfully forecasted the 2015/2016 and 2018/2019 El Niño events at 24 months lead.
• Successfully forecasted the 2016/2017, 2017/2018, and 2020/2021 La Niña events at 24 months lead.
• Successfully forecasted the 2021/2022 and 2022/2023 La Niña events at 12 and 8 months lead, respectively.

Contributions

• Extension of the eSPA algorithm for ENSO phase prediction, building upon previous work.
• Introduction of a novel ensemble approach for eSPA, utilizing a meta-learning strategy to aggregate multiple models.
• Demonstration of high-skill probabilistic ENSO forecasts (comparable to IRI, with extended lead times up to 24 months) using only satellite-era observational data.
• Validation of the computational efficiency of the eSPA approach compared to fully coupled dynamical models.
• Rigorous data processing methodology to prevent information leakage, ensuring realistic real-time forecasting conditions.

Funding

Not explicitly stated in the abstract.

Citation

@article{Groom2025Entropic,
  author = {Groom, Michael and Bassetti, Davide and Horenko, Illia and O'Kane, Terence J.},
  title = {Entropic Learning Enables Skilful Forecasts of ENSO Phase at up to 2 Years Lead Time},
  journal = {Journal of Advances in Modeling Earth Systems},
  year = {2025},
  doi = {10.1029/2025ms005128},
  url = {https://doi.org/10.1029/2025ms005128}
}