Dynamics for El Niño-La Niña asymmetry constrain equatorial-Pacific warming pattern

ENSO arises from coupled ocean–atmosphere feedbacks in the equatorial Pacific and exerts nonlinear rectification effects on the mean state, particularly over the cold tongue. Despite advances in theory and modeling, reproducing ENSO’s complex spatial–temporal behavior and the correct balance of feedbacks is challenging. State-of-the-art models often achieve realistic ENSO amplitude via compensating errors between overly weak positive dynamical and overly strong negative thermodynamic feedbacks. A critical shortcoming is the poor simulation of ENSO nonlinearity, notably the observed positive skewness of SST and subsurface temperature anomalies in the eastern equatorial Pacific, which limits realistic occurrence of extreme El Niño events and reduces nonlinear rectification onto the mean state. Prior work has implicated mean-state SST biases that shift the Walker circulation westward, feedback errors, and state-dependent stochastic forcing, yet the dominant nonlinear dynamical process driving the asymmetry has remained unclear. Observations identify subsurface nonlinear dynamical heating (NDH), associated with weakening/halt of the Equatorial Undercurrent during strong El Niño, as a key mechanism enhancing asymmetry by reducing cooling tendencies during transitions to La Niña. The research question addressed here is which dynamical processes control inter-model diversity in ENSO asymmetry and how these processes affect projected tropical Pacific warming patterns under greenhouse forcing.

The study builds on extensive literature on ENSO dynamics and diversity, error compensation of feedbacks in coupled models, and asymmetry (skewness) biases in CMIP models. Prior studies have linked weak asymmetry to mean-state SST and Walker circulation biases affecting atmospheric feedbacks, state-dependent noise, and have highlighted the importance of nonlinear processes, including surface-layer NDH and subsurface dynamics tied to the Equatorial Undercurrent. Work has also debated future changes in ENSO amplitude and the tropical Pacific warming pattern (El Niño-like vs La Niña-like), with mechanisms including changes in upwelling, circulation weakening, and off-equatorial forcing. Analytical and modeling studies suggested that surface NDH alone may not guarantee positive SST skewness, pointing toward subsurface processes as crucial. This study evaluates these concepts across reanalyses and CMIP ensembles to identify controlling dynamics for ENSO asymmetry.

Data: Historical simulations from CMIP5 (25 models, 1850–2005) and CMIP6 (26 models, 1850–2014) were analyzed for monthly SST (tos), zonal wind stress (tauu), near-surface air temperature (tas), surface heat flux components (rsus, rsds, rlus, rlds, hfss, hfls), and precipitation (pr). Ocean potential temperature (thetao) and 3D currents (uo, vo, wo or wmo) were available for all 25 CMIP5 models and 18 of 26 CMIP6 models. For models lacking wo, vertical velocity was derived from vertical mass transport (wmo) using gridcell area (areacello) and reference density (1035 kg m−3). MPI model currents were corrected for curved equatorial axes. Surface fields were interpolated to 1°×1°; ocean fields to 1°×0.5°. Future scenarios analyzed include CMIP5 RCP8.5 (2006–2100) and CMIP6 SSP5-8.5 (2015–2100); linear trends for composites primarily use CMIP5 RCP8.5 due to limited CMIP6 availability. Anomalies were linearly detrended for each period. Reanalyses: Four ocean reanalyses (ORAS3, ORAS5, SODA3.3.1, GODAS) provided subsurface temperature, currents, wind stress, and surface flux; SST was taken as the nearest-surface potential temperature. Three atmospheric reanalyses and observational products (ERA-Interim, ERA5, TropFlux) plus radiative flux datasets (OAFlux, CERES EBAF, GEWEX SRB, ISCCP-FH) and GPCP precipitation were used. NOAA ERSSTv5 provided SST for regressions. Indices and diagnostics: ENSO amplitude (σENSO) and asymmetry (γENSO) were defined as the standard deviation and skewness (third normalized moment) of detrended Niño-3 SST anomalies (150°–90°W, 5°S–5°N). Shortwave (SW) and longwave surface heat flux anomalies were averaged over Niño regions; the SW zonal contrast ASW was defined as western (140°–170°E) minus eastern (140°–170°W) averages (5°S–5°N). Feedbacks were computed as regression coefficients of anomalies onto Niño-3 SST; feedback asymmetry was the difference between regressions for positive vs negative Niño-3 SST anomalies. Central Pacific (CP) zonal wind stress anomalies were averaged over 150°E–120°W, 5°S–5°N. Nonlinear dynamical heating (NDH): NDH was computed from nonlinear temperature advection terms using monthly anomalies and climatologies of velocity (u, v, w) and temperature (T); residual terms include thermodynamic and subgrid processes but advective terms dominate subsurface temperature tendency in reanalyses. Subsurface NDH was averaged over 100°W–180°, 1°S–1°N, 50–150 m. NDH variability amplitude is the monthly standard deviation (K month−1). NDH efficiency is defined as σNDHsub/σENSO (month−1). Surface NDH was also examined above 50 m. Model grouping: Models were classified by NDH efficiency: Group H (high NDH efficiency; ≥ multi-model mean 0.16 month−1; 10 CMIP5, 4 CMIP6), Group HH (subset of H with efficiency >0.20 month−1; 6 CMIP5, 1 CMIP6), and Group L (low NDH efficiency; 29 models). Additional subgroups within H for future projections: H-sEN (projected ENSO strengthening; 7 models) and H-wEN (projected ENSO weakening; 6 models). Future metrics: ENSO amplitude change and mean NDH change were computed as differences between historical and 2051–2100 scenario periods. El Niño-likeness was defined as the spatial correlation between the SST linear trend pattern (scenario) and the historical ENSO-regressed SST anomaly pattern over 90°E–60°W, 20°S–20°N (positive indicates more eastern Pacific warming). Statistical analysis: Two-tailed Student’s t-tests assessed confidence for composites and correlations; Welch’s t-test assessed differences between CMIP groups. Confidence levels are provided in figure descriptions.

- ENSO SST amplitude (σENSO) in CMIP models is on average close to observations, but ENSO SST skewness (γENSO) is severely underestimated, with multi-model mean not distinguishable from zero; no improvement from CMIP5 to CMIP6. - Subsurface NDH mean and variability are too weak in CMIP models relative to reanalyses, indicating deficiencies in subsurface ocean nonlinear dynamics. - γENSO strongly increases with subsurface NDH efficiency (σNDHsub/σENSO) across models: r = 0.78 (p < 1e−5), explaining ~60% of inter-model variance. Observed NDH efficiency ≈ 0.30 month−1; few models approach this value. Surface NDH above 50 m does not explain γENSO spread. - Atmospheric feedback nonlinearities contribute less: γENSO vs SW feedback asymmetry (ASW) shows a moderate correlation r = 0.37 (p = 0.008); γENSO vs CP wind stress feedback asymmetry r = 0.13 (weak). Reanalyses show near-zero dynamic feedback asymmetry; many models simulate positive asymmetry, yet this does not fix skewness. - Grouping by NDH efficiency reveals dynamics: Group H and especially HH better reproduce positive SST skewness in the east and negative in the west, and show positive mean subsurface NDH (rectified warming) near the thermocline, similar to reanalyses. Group L (≈70% of models) fails to reproduce these nonlinear features. - Linear coupling biases in Group L: too-weak CP zonal wind stress response to Niño-3 SST (SST-to-wind coupling) and a weak anomalous westward EUC response to westerly wind anomalies (wind-to-EUC coupling). These weaken wind-induced subsurface NDH and thus ENSO asymmetry. Group H/HH show improved linear couplings and reduced cold-tongue bias. - Error compensation: Many models balance weak positive dynamical feedbacks with strong negative thermodynamic feedbacks to yield realistic ENSO amplitude, but this compensation fails for ENSO asymmetry because it depends on dynamical coupling via subsurface NDH. - Future projections conditioned on dynamics: In Group H, change in mean subsurface NDH scales with ENSO amplitude change (Fig. 6a): r = 0.97 (Group H), r = 0.99 (Group HH), r = 0.65 (Group L). El Niño-likeness of tropical warming increases with ENSO amplitude change in Group H (r = 0.57, p = 0.042) and is even stronger in HH (r = 0.89, p = 0.0073). Group L shows no relationship (r = 0.00) and generally projects El Niño-like warming regardless of ENSO change. - CMIP5 RCP8.5 composites: H-sEN vs H-wEN show significantly different SST warming patterns—enhanced eastern equatorial Pacific warming and precipitation in H-sEN; reduced in H-wEN. Corresponding subsurface NDH trends show east–west contrasts consistent with rectified warming changes. Group L sEN vs wEN differences are statistically indistinguishable and ENSO-related trends are muted. - Implication: Only models that realistically capture nonlinear ENSO dynamics (subsurface NDH) translate ENSO amplitude changes into distinct equatorial Pacific warming patterns under greenhouse forcing.

The findings identify subsurface NDH as a key dynamical control on ENSO asymmetry and on how ENSO changes modulate future equatorial Pacific warming patterns. Weak NDH variability in most CMIP models arises from biases in linear ocean–atmosphere coupling—specifically, a too-weak SST-to-wind response and an unrealistic wind-to-EUC response. While compensating errors between dynamical and thermodynamic feedbacks can yield realistic ENSO amplitude, this masks fundamental deficiencies exposed by low ENSO skewness. Conditioning the model ensemble on realistic ENSO nonlinear dynamics reveals a strong linear link between ENSO amplitude change and both subsurface NDH trends and El Niño-likeness of surface warming. Consequently, models with realistic NDH project that strengthening ENSO enhances eastern Pacific warming via increased rectified subsurface warming and a deepened thermocline, whereas weakening ENSO reduces this effect, leading to less El Niño-like warming. Without realistic NDH, models default to El Niño-like warming driven by ENSO-independent mean-state processes (e.g., weakening Walker/Hadley circulations), obscuring the ENSO contribution. These results underscore that accurate simulation of both linear and nonlinear ENSO dynamics is essential for credible projections of tropical and even global climate, given potential feedbacks on large-scale circulation and climate sensitivity.

This study demonstrates that inter-model differences in ENSO asymmetry are largely controlled by subsurface nonlinear dynamical heating along the equatorial thermocline. Most CMIP models underestimate NDH and possess weak linear dynamical coupling, leading to poor SST skewness despite reasonable ENSO amplitude via error compensation. A subset of models with relatively realistic NDH efficiency not only better reproduces observed asymmetry but also shows a robust linkage between ENSO amplitude changes and the equatorial Pacific warming pattern under greenhouse forcing. Improving representation of the SST-to-wind and wind-to-EUC couplings and subsurface NDH is therefore critical to reduce uncertainty in projections of tropical Pacific warming and its global impacts. Future research should target diagnosing and correcting biases in atmospheric convection and boundary-layer coupling that weaken wind responses to SST, improving ocean model responses of the EUC to wind forcing, enhancing vertical and horizontal advection representations controlling NDH, and expanding availability of required subsurface diagnostics in CMIP archives to better constrain ENSO nonlinear dynamics.

- Data availability limited NDH evaluation to 25 CMIP5 and 18 CMIP6 models due to missing 3D ocean fields (uo, vo, wo/wmo), reducing sample size for some analyses and especially for CMIP6 scenarios. - Future trend composites primarily used CMIP5 RCP8.5 because only three CMIP6 models in Group H had SSP5-8.5 data; differences between RCP8.5 and SSP5-8.5 pathways may introduce scenario-specific nuances. - NDH diagnostics rely on advective tendency dominance documented in reanalyses; residual thermodynamic and subgrid processes are not explicitly resolved. - Reanalysis products have inherent uncertainties; multi-product consistency mitigates but does not eliminate this. - Relationships are based on correlations and conditioning by model fidelity; while dynamically motivated, causal attributions within individual models may still be influenced by other model-specific biases.