Siqueira et al. (2026) Do our models capture the range of ENSO events?

Identification

• Journal: Climate Dynamics
• Year: 2026
• Date: 2026-03-27
• Authors: Leo Siqueira, Ben P. Kirtman
• DOI: 10.1007/s00382-026-08137-9

Research Groups

• Rosenstiel School for Marine and Atmospheric Science, University of Miami, Coral Gables, USA
• Institute for Data Science and Computing, University of Miami, Coral Gables, USA
• Cooperative Institute for Marine and Atmospheric Studies, University of Miami, Coral Gables, USA

Short Summary

This study employs a novel event-based space-time analog approach to evaluate whether state-of-the-art coupled climate models (CESM1 and E3SM1) capture the full range of observed ENSO variability. It reveals that low-resolution models generally outperform their high-resolution counterparts in reproducing observed ENSO events, and increasing resolution does not significantly reduce mean-state biases.

Objective

• To evaluate whether current Earth System Models (CESM1 and E3SM1) at different resolutions capture the full range of observed ENSO variability and specific extreme events using an event-by-event space-time analog approach.

Study Configuration

• Spatial Scale: Tropical Indo-Pacific region (20° S–20° N, 30° E–80° W) for analog selection; equatorial Pacific (5° S–5° N) for cross-sections. Model resolutions include ocean eddy-parameterized (~1° horizontal resolution) and ocean eddy-resolving (~0.1° horizontal resolution), with atmospheric/land components at ~1° (low-resolution) and ~0.25° (high-resolution). All data are interpolated to a 1° longitude by 1° latitude grid for comparison.
• Temporal Scale: Observed ENSO events from 1980–2016 (for classification) and specific strong events (e.g., 1997–98, 2002–03, 1988–89). Analog search and evaluation cover a 12-month space-time evolution. Model simulations are multi-century (CESM1.3: 500 years preindustrial; E3SMv1: 164 years historical, 156 years 1950 CO2 forcing). Observational and reanalysis data span 1979–2018 (OISST, ORAS5) and 1980–2018 (SODA3.3.2).

Methodology and Data

• Models used:
  • Community Earth System Model version 1.3 (CESM1.3):
    • CESM1-LR: Nominal 1° horizontal resolution in ocean/atmosphere, with 1/3° equatorial ocean refinement.
    • CESM1-HR: 0.1° horizontal resolution in the ocean (eddy-resolving), 0.25° in atmosphere/land.
  • Energy Exascale Earth System Model version 1 (E3SMv1):
    • E3SM1-LR: Ocean resolution varying from ~30 km at the equator to ~60 km at mid-latitudes; land/atmosphere at ~100 km.
    • E3SM1-HR: Land/atmosphere at ~25 km; multiresolution ocean from ~6 km at poles to ~18 km at the equator.
• Data sources:
  • Observations: NOAA Optimum Interpolation Sea Surface Temperature, version 2 (OISSTv2).
  • Reanalysis: ECMWF Ocean Reanalysis System 5 (ORAS5) for Sea Surface Height (SSH). Simple Ocean Data Assimilation version 3.3.2 (SODA3.3.2) for surface wind stress and subsurface ocean temperatures (forced by NASA MERRA-2 reanalysis).
  • Methodology: An event-based space-time analog approach is used, where model states are selected from multi-century simulations by minimizing the root-mean-square (RMS) difference between observed target states and library states over a 12-month space-time evolution. Analogs are defined separately for Sea Surface Temperature Anomalies (SSTA) and Sea Surface Height Anomalies (SSHA) in the tropical Indo-Pacific region. ENSO events are classified based on detrended Niño3.4 and Niño4 indices. Diagnostics include mean-state biases, SSTA skewness, subsurface temperature variance and skewness, and analysis of feedback parameters (e.g., zonal wind stress–SSTA tendency) from a heuristic SSTA equation.

Main Results

• Neither CESM1 nor E3SM1 models show significant improvements in large-scale mean-state biases (e.g., excessively cold tongue, thermocline depth errors) with increased resolution.
• Low-resolution models (E3SM1-LR and CESM1-LR) consistently outperform their high-resolution counterparts in capturing Eastern Pacific (EP) and Central Pacific (CP) warm events, as well as cold events, for both SST-based and SSH-based analogs.
• SSH-based analogs exhibit Root Mean Square Error (RMSE) values approximately twice as large as SST-based analogs, with errors increasing rapidly with analog ranking.
• The analog approach successfully captures the observed space-time evolution and peak characteristics of strong EP and CP warm and cold events.
• All models display excessive variability in the western-central Pacific during both warm and cold events, and they struggle to accurately reproduce the magnitude of anomalies near the eastern boundary, which is linked to an overly strong local SSTA tendency—zonal wind stress coupling.
• The stronger the observed ENSO event, the more effective the analogs are, with only modest differences attributed to resolution.
• Models generally underestimate the observed positive asymmetry in SSTA and subsurface temperature anomalies, and this deficiency is insensitive to resolution in E3SM1 and CESM1.
• CESM1-HR shows degraded SSTA–thermocline coupling in the central-eastern Pacific compared to observations and other model configurations.

Contributions

• Introduces and applies a novel event-based space-time analog approach to assess whether observed ENSO events are represented within the state space of climate models, offering a complementary perspective to traditional statistical comparisons.
• Provides a comprehensive intercomparison of ENSO simulation capabilities of two prominent Earth System Models (CESM1 and E3SM1) across varying ocean resolutions (eddy-parameterized vs. eddy-resolving).
• Demonstrates that increasing model resolution does not necessarily lead to fundamental improvements in simulating ENSO mean-state biases or the event-by-event evolution of ENSO, with low-resolution models often performing better.
• Identifies specific model deficiencies, such as excessive variability in the western-central Pacific, underestimation of eastern boundary anomaly magnitudes, and degraded SSTA-thermocline coupling in high-resolution CESM1.
• Offers insights into how background state errors and local feedback mechanisms (e.g., zonal wind stress–SSTA tendency) impact the simulated ENSO cycle, providing clues for future model improvements.

Funding

• NOAA (NA20OAR4320472, NA22OAR4310603, NA23OAR4590384 & NA23OAR4310457)
• NSF (AGS2241538 & AGS2223263)
• Frost Institute for Data Science and Computing (IDSC)
• William R. Middelthon Chair of Earth Sciences (for Ben P. Kirtman)

Citation

@article{Siqueira2026Do,
  author = {Siqueira, Leo and Kirtman, Ben P.},
  title = {Do our models capture the range of ENSO events?},
  journal = {Climate Dynamics},
  year = {2026},
  doi = {10.1007/s00382-026-08137-9},
  url = {https://doi.org/10.1007/s00382-026-08137-9}
}