pith. machine review for the scientific record. sign in

arxiv: 2604.07137 · v2 · submitted 2026-04-08 · ⚛️ physics.ao-ph

Recognition: unknown

What's in the latent space? Exploring coupled tropical Pacific variability within a Multi-branch β-Variational Autoencoder

Emily F. Wisinski , Maria J. Molina , Kyle J. C. Hall , Hannah Bao , Salil Mahajan , Nan Rosenbloom , John Fasullo
Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:00 UTC · model grok-4.3

classification ⚛️ physics.ao-ph
keywords tropical Pacific variabilityEl Niñovariational autoencoderlatent spaceclimate fieldssea surface temperatureocean heat contentoutgoing longwave radiation
0
0 comments X

The pith

A multi-branch β-VAE produces a reduced latent representation of tropical Pacific climate that aligns with known El Niño and La Niña modes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors train a multi-branch β-variational autoencoder on three coupled fields from the tropical Pacific: sea surface temperature, ocean heat content, and outgoing longwave radiation. They show that the resulting latent space reconstructs the main spatial patterns well and that individual dimensions correspond to specific types of variability, such as eastern or central Pacific El Niño events. A reader might care because this offers a way to simplify the study of complex climate interactions while keeping physical meaning. The work demonstrates that variability is distributed unevenly across the latent dimensions, with temperature more concentrated than the other fields.

Core claim

The multi-branch β-variational autoencoder yields a skillful and physically informative reduced representation of coupled tropical Pacific variability. Latent dimensions align with conventional El Niño and La Niña diagnostics, as well as decadal-scale coupled ocean-atmosphere variability, identified through sensitivity experiments and latent traversals.

What carries the argument

The multi-branch β-VAE, a neural network that learns a low-dimensional latent space from multiple input fields with a regularization term controlled by β to promote disentangled representations.

Load-bearing premise

That the alignments between latent dimensions and climate diagnostics represent true physical couplings instead of being produced by the model's architecture, training data, or the way the results were interpreted after training.

What would settle it

Observing whether the same latent dimensions align with El Niño and La Niña patterns when the model is retrained or evaluated on independent observational datasets of the tropical Pacific.

read the original abstract

What is encoded in the latent space of a multi-branch $\beta$-variational autoencoder ($\beta$-VAE) trained on coupled tropical Pacific climate fields? To answer this question, we assess the reconstruction skill and physical interpretability of the latent space of a multi-branch $\beta$-VAE trained on sea surface temperature, ocean heat content, and outgoing longwave radiation across the tropical Pacific from a 500-year preindustrial control simulation. The model generalizes well, with only modest degradation from training to test performance, and preserves the dominant basin-scale structure of all three fields. Latent-space diagnostics show that variability is organized unevenly across dimensions: sea surface temperature is concentrated in a smaller subset of latent dimensions, whereas ocean heat content and outgoing longwave radiation are more broadly distributed across multiple dimensions. Comparisons with conventional tropical Pacific diagnostics further show that several latent dimensions align with known El Ni\~no and La Ni\~na variability, while others capture related coupled ocean-atmosphere variability on decadal or longer timescales. Sensitivity experiments and latent traversals identify dimensions associated with eastern-Pacific-like, central-Pacific-like, coastal, subsurface-dominant, and atmosphere-dominant variability. Together, these results show that the multi-branch $\beta$-variational autoencoder yields a skillful and physically informative reduced representation of coupled tropical Pacific variability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper trains a multi-branch β-variational autoencoder on sea surface temperature, ocean heat content, and outgoing longwave radiation fields over the tropical Pacific from a 500-year preindustrial control simulation. It evaluates reconstruction skill and generalization, reports that variability is unevenly distributed across latent dimensions, and uses comparisons to conventional ENSO/decadal diagnostics, latent traversals, and sensitivity experiments to argue that several dimensions capture physically meaningful coupled ocean-atmosphere modes.

Significance. If the reported alignments prove robust, the work shows that a multi-branch β-VAE can deliver a skillful, interpretable reduced representation of coupled tropical Pacific variability that preserves basin-scale structures and aligns with known modes. The long control run enables examination of decadal timescales, and the multi-field input plus interpretability diagnostics represent a constructive application of ML to climate dynamics.

major comments (3)
  1. [Methods] Methods section: the specific value of the β hyperparameter and the chosen latent dimensionality are free parameters whose selection criteria and sensitivity are not quantified; these choices directly affect how variability is partitioned across dimensions and must be documented with ablation results.
  2. [Results] Results on generalization: the abstract asserts 'only modest degradation from training to test performance' and 'preserves the dominant basin-scale structure,' yet no numerical reconstruction metrics (RMSE, anomaly correlation, or field-specific error tables) are referenced, preventing assessment of whether the skill is competitive with linear baselines such as EOF analysis.
  3. [§4-5] Latent-space diagnostics: the claim that dimensions align with eastern-Pacific-like, central-Pacific-like, and decadal variability rests on qualitative traversals and sensitivity tests; without reported correlation coefficients or explained-variance fractions between individual latent coordinates and standard Niño indices or PDO-like indices, the physical correspondence remains vulnerable to post-hoc interpretation.
minor comments (2)
  1. [Abstract] The abstract would benefit from stating the exact latent dimensionality and the number of branches used.
  2. [Figures] Figure captions for traversals should include physical units and reference climatological ranges to aid reader interpretation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments and positive overall assessment. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Methods] Methods section: the specific value of the β hyperparameter and the chosen latent dimensionality are free parameters whose selection criteria and sensitivity are not quantified; these choices directly affect how variability is partitioned across dimensions and must be documented with ablation results.

    Authors: We agree that explicit documentation of hyperparameter choices and their sensitivity is necessary. The revised Methods section will state the selected β value and latent dimensionality, describe the validation-based selection criteria, and include ablation results showing reconstruction performance and variability partitioning across a range of β values and latent dimensions. revision: yes

  2. Referee: [Results] Results on generalization: the abstract asserts 'only modest degradation from training to test performance' and 'preserves the dominant basin-scale structure,' yet no numerical reconstruction metrics (RMSE, anomaly correlation, or field-specific error tables) are referenced, preventing assessment of whether the skill is competitive with linear baselines such as EOF analysis.

    Authors: The current manuscript provides only qualitative statements on generalization. We will add a table in the revised Results section reporting RMSE and anomaly correlation values for each field on training and test sets, together with a direct comparison against EOF analysis to quantify competitiveness in preserving basin-scale structures. revision: yes

  3. Referee: [§4-5] Latent-space diagnostics: the claim that dimensions align with eastern-Pacific-like, central-Pacific-like, and decadal variability rests on qualitative traversals and sensitivity tests; without reported correlation coefficients or explained-variance fractions between individual latent coordinates and standard Niño indices or PDO-like indices, the physical correspondence remains vulnerable to post-hoc interpretation.

    Authors: We acknowledge that quantitative metrics would strengthen the physical interpretation. The revised Sections 4 and 5 will include correlation coefficients between individual latent dimensions and standard Niño indices as well as PDO-like indices, plus the fraction of variance in each physical index captured by the latent coordinates. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper reports an empirical ML workflow: a multi-branch β-VAE is trained on three coupled fields from a single control simulation, reconstruction fidelity is measured on held-out data, and latent dimensions are inspected via traversals and sensitivity tests then aligned with independent conventional ENSO and decadal diagnostics. No equation or claim reduces by construction to a quantity defined solely in terms of the model's own fitted parameters; the central claim of a skillful and physically informative representation follows directly from the reported metrics and external comparisons. No self-citation chain, uniqueness theorem, or ansatz smuggling is invoked to justify the core result.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The claim rests on standard variational autoencoder assumptions plus the representativeness of the preindustrial simulation; no new physical entities are introduced.

free parameters (2)
  • β hyperparameter
    Controls the balance between reconstruction fidelity and latent-space regularization; its value is chosen to produce interpretable dimensions.
  • latent dimensionality
    Number of latent dimensions is a modeling choice that affects how variability is distributed across dimensions.
axioms (2)
  • domain assumption The 500-year preindustrial control simulation adequately samples natural coupled tropical Pacific variability.
    Used to justify training and generalization claims.
  • domain assumption Latent dimensions learned by the β-VAE can be meaningfully aligned with physical climate modes via traversals and sensitivity tests.
    Underpins the physical interpretability conclusions.

pith-pipeline@v0.9.0 · 5566 in / 1466 out tokens · 84012 ms · 2026-05-10T17:00:30.938996+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

67 extracted references · 60 canonical work pages · 2 internal anchors

  1. [1]

    Ashok, K., S. K. Behera, S. A. Rao, H. Weng, and T. Yamagata, 2007: El Ni ˜no Modoki and its possible teleconnection.Journal of Geophysical Research: Oceans,112 (C11), https://doi.org/ 10.1029/2006JC003798

    work page doi:10.1029/2006jc003798 2007

  2. [2]

    Bader, D. C., R. Leung, M. Taylor, and R. B. McCoy, 2022: E3SM-Project E3SM2.0 model output prepared for CMIP6 CMIP piControl. Earth System Grid Federation, URL https://doi.org/10. 22033/ESGF/CMIP6.16941, [Data set]. Version YYYYMMDD. Accessed on DD-MON-YYYY, https://doi.org/10.22033/ESGF/CMIP6.16941

    work page doi:10.22033/esgf/cmip6.16941 2022

  3. [3]

    Bamston, A. G., M. Chelliah, and S. B. Goldenberg, 1997: Documentation of a highly ENSO- related SST region in the equatorial Pacific: Research note.Atmosphere-Ocean,35 (3), 367–383, https://doi.org/10.1080/07055900.1997.9649597

    work page doi:10.1080/07055900.1997.9649597 1997

  4. [4]

    Bretherton, C. S., M. Widmann, V. P. Dymnikov, J. M. Wallace, and I. Blad´e, 1999: The effective number of spatial degrees of freedom of a time-varying field.Journal of Climate,12 (7), 1990–2009, https://doi.org/10.1175/1520-0442(1999)012⟨1990:TENOSD⟩2.0.CO;2

    work page doi:10.1175/1520-0442(1999)012 1999

  5. [5]

    Capotondi, A., and Coauthors, 2015: Understanding ENSO Diversity.Bulletin of the American Meteorological Society,96 (6), 921–938, https://doi.org/10.1175/BAMS-D-13-00117.1

    work page doi:10.1175/bams-d-13-00117.1 2015

  6. [6]

    B., and M

    Chelton, D. B., and M. G. Schlax, 1996: Global observations of oceanic Rossby waves.Science, 272 (5259), 234–238, https://doi.org/10.1126/science.272.5259.234

    work page doi:10.1126/science.272.5259.234 1996

  7. [7]

    Chen, X., B. Qiu, Y. Du, S. Chen, and Y. Qi, 2016: Interannual and interdecadal variability of the North Equatorial Countercurrent in the Western Pacific.Journal of Geophysical Research: Oceans,121 (10), 7743–7758, https://doi.org/10.1002/2016JC012190

    work page doi:10.1002/2016jc012190 2016

  8. [8]

    M., and D

    Chiodi, A. M., and D. E. Harrison, 2013: El Ni˜no impacts on seasonal US atmospheric circulation, temperature, and precipitation anomalies: The OLR-event perspective.Journal of Climate, 26 (3), 822–837, https://doi.org/10.1175/JCLI-D-12-00097.1

    work page doi:10.1175/jcli-d-12-00097.1 2013

  9. [9]

    Deser, C., M. A. Alexander, S.-P. Xie, and A. S. Phillips, 2010: Sea Surface Temperature Variabil- ity: Patterns and Mechanisms.Annual Review of Marine Science,2 (1), 115–143, https://doi.org/ 10.1146/annurev-marine-120408-151453. 45

    work page doi:10.1146/annurev-marine-120408-151453 2010

  10. [10]

    Bayr, and C

    Dommenget, D., T. Bayr, and C. Frauen, 2013: Analysis of the non-linearity in the pattern and time evolution of El Ni ˜no Southern Oscillation.Climate Dynamics,40 (11), 2825–2847, https://doi.org/0.1007/s00382-012-1475-0

    2013

  11. [11]

    Lamb, 2025: Unsupervised Classification of Absorbing Aerosols with the SP2 via a Variational Autoencoder (V AE).EGUsphere,2025, 1–20, https://doi.org/10.5194/ egusphere-2025-3210

    Doshi, A., and K. Lamb, 2025: Unsupervised Classification of Absorbing Aerosols with the SP2 via a Variational Autoencoder (V AE).EGUsphere,2025, 1–20, https://doi.org/10.5194/ egusphere-2025-3210

    2025

  12. [12]

    Fasullo, J., and Coauthors, 2023: An overview of the E3SM version 2 large ensemble and com- parison to other E3SM and CESM large ensembles.EGUsphere,2023, 1–32, https://doi.org/ 10.5194/esd-15-367-2024

    work page doi:10.5194/esd-15-367-2024 2023

  13. [13]

    T., and Coauthors, 2024: Modes of Variability in E3SM and CESM Large Ensembles

    Fasullo, J. T., and Coauthors, 2024: Modes of Variability in E3SM and CESM Large Ensembles. Journal of Climate,37 (8), 2629–2653, https://doi.org/10.1175/JCLI-D-23-0454.1

    work page doi:10.1175/jcli-d-23-0454.1 2024

  14. [14]

    C., and Coauthors, 2025: Taking the Garbage Out of Data-Driven Prediction Across Cli- mate Timescales.arXiv preprint arXiv:2508.07062, https://doi.org/10.48550/arXiv.2508.07062

    Furtado, J. C., and Coauthors, 2025: Taking the Garbage Out of Data-Driven Prediction Across Cli- mate Timescales.arXiv preprint arXiv:2508.07062, https://doi.org/10.48550/arXiv.2508.07062

    work page doi:10.48550/arxiv.2508.07062 2025

  15. [15]

    Geng, T., W. Cai, L. Wu, and Y. Yang, 2019: Atmospheric convection dominates genesis of ENSO asymmetry.Geophysical Research Letters,46 (14), 8387–8396, https://doi.org/10.1029/ 2019GL083213

    2019

  16. [16]

    Golaz, J.-C., and Coauthors, 2022: The DOE E3SM model Version 2: Overview of the Physical Model and Initial Model Evaluation.Journal of Advances in Modeling Earth Systems,14 (12), e2022MS003 156, https://doi.org/10.1029/2022MS003156

    work page doi:10.1029/2022ms003156 2022

  17. [17]

    Bengio, and A

    Goodfellow, I., Y. Bengio, and A. Courville, 2016:Deep Learning. MIT Press, http://www. deeplearningbook.org

    2016

  18. [18]

    Hall, K. J., M. J. Molina, E. F. Wisinski, G. A. Meehl, and A. Capotondi, 2025: Knowledge-guided machine learning for disentangling Pacific sea surface temperature variability across timescales. arXiv preprint arXiv:2508.08490, https://doi.org/10.48550/arXiv.2508.08490

    work page doi:10.48550/arxiv.2508.08490 2025

  19. [19]

    Kim, and J.-J

    Ham, Y.-G., J.-H. Kim, and J.-J. Luo, 2019: Deep learning for multi-year ENSO forecasts.Nature, 573 (7775), 568–572, https://doi.org/10.1038/s41586-019-1559-7. 46

    work page doi:10.1038/s41586-019-1559-7 2019

  20. [20]

    Han, T., Z. Chen, S. Guo, W. Xu, W. Ouyang, and L. Bai, 2025: Climate science data can be compressed efficiently by dual-stage extreme compression with a variational auto- encoder transformer.Communications Earth & Environment,6 (1), 955, https://doi.org/ 10.1038/s43247-025-02903-z

    work page doi:10.1038/s43247-025-02903-z 2025

  21. [21]

    Matthey, A

    Higgins, I., L. Matthey, A. Pal, C. Burgess, X. Glorot, M. Botvinick, S. Mohamed, and A. Lerchner, 2017: beta-V AE: Learning Basic Visual Concepts with a Constrained Variational Framework. International Conference on Learning Representations, URL https://openreview.net/forum?id= Sy2fzU9gl

    2017

  22. [22]

    C., and M

    Johnson, G. C., and M. J. McPhaden, 1999: Interior pycnocline flow from the subtropical to the equatorial Pacific Ocean.Journal of Physical Oceanography,29 (12), 3073–3089, https://doi.org/10.1175/1520-0485(1999)029⟨3073:IPFFTS⟩2.0.CO;2

    work page doi:10.1175/1520-0485(1999)029 1999

  23. [23]

    Kadow, C., D. M. Hall, and U. Ulbrich, 2020: Artificial intelligence reconstructs missing climate information.Nature Geoscience,13 (6), 408–413, https://doi.org/10.1038/s41561-020-0582-5

    work page doi:10.1038/s41561-020-0582-5 2020

  24. [24]

    Yu, 2009: Contrasting Eastern-Pacific and Central-Pacific Types of ENSO

    Kao, H.-Y., and J.-Y. Yu, 2009: Contrasting Eastern-Pacific and Central-Pacific Types of ENSO. Journal of Climate,22 (3), 615–632, https://doi.org/10.1175/2008JCLI2309.1

    work page doi:10.1175/2008jcli2309.1 2009

  25. [25]

    Kiladis, G. N., M. C. Wheeler, P. T. Haertel, K. H. Straub, and P. E. Roundy, 2009: Con- vectively coupled equatorial waves.Reviews of Geophysics,47 (2), https://doi.org/10.1029/ 2008RG000266

    2009

  26. [26]

    Disentangling by Factorising

    Kim, H., and A. Mnih, 2018: Disentangling by factorising.International conference on machine learning, PMLR, 2649–2658, https://doi.org/10.48550/arXiv.1802.05983

    work page Pith review doi:10.48550/arxiv.1802.05983 2018

  27. [27]

    Pettersen, D

    King, F., C. Pettersen, D. Posselt, S. Ringerud, and Y. Xie, 2025: Leveraging Sparse Au- toencoders to Reveal Interpretable Features in Geophysical Models.Journal of Geophysi- cal Research: Machine Learning and Computation,2 (4), e2025JH000 769, https://doi.org/ 10.1029/2025JH000769

    work page doi:10.1029/2025jh000769 2025

  28. [28]

    Auto-Encoding Variational Bayes

    Kingma, D. P., 2013: Auto-Encoding Variational Bayes.arXiv preprint arXiv:1312.6114, https://doi.org/10.48550/arXiv.1312.6114

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1312.6114 2013

  29. [29]

    Adam: A Method for Stochastic Optimization

    Kingma, D. P., and J. Ba, 2014: Adam: A Method for Stochastic Optimization.arXiv preprint arXiv:1412.6980, https://doi.org/10.48550/arXiv.1412.6980. 47

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980 2014

  30. [30]

    P., and P

    Kirtman, B. P., and P. S. Schopf, 1998: Decadal Variability in ENSO Predictability and Prediction. Journal of Climate,11 (11), 2804–2822, https://doi.org/10.1175/1520-0442(1998)011⟨2804: DVIEPA⟩2.0.CO;2

    work page doi:10.1175/1520-0442(1998)011 1998

  31. [31]

    Jin, and S.-I

    Kug, J.-S., F.-F. Jin, and S.-I. An, 2009: Two Types of El Ni ˜no Events: Cold Tongue El Ni ˜no and Warm Pool El Ni ˜no.Journal of Climate,22 (6), 1499–1515, https://doi.org/ 10.1175/2008JCLI2624.1

    work page doi:10.1175/2008jcli2624.1 2009

  32. [32]

    E., and K

    Lemmon, D. E., and K. B. Karnauskas, 2019: A metric for quantifying El Ni ˜no pattern diversity with implications for ENSO–mean state interaction.Climate Dynamics,52 (12), 7511–7523, https://doi.org/10.1007/s00382-018-4194-3

    work page doi:10.1007/s00382-018-4194-3 2019

  33. [33]

    Zhang, and C

    Ma, X., L. Zhang, and C. K. Wikle, 2025: Modeling Spatio-temporal Extremes via Condi- tional Variational Autoencoders.arXiv preprint arXiv:2512.06348, https://doi.org/10.48550/ arXiv.2512.06348

    work page arXiv 2025

  34. [34]

    MacMillan, T., and N. T. Ouellette, 2025: Towards mechanistic understanding in a data-driven weather model: internal activations reveal interpretable physical features.arXiv preprint arXiv:2512.24440, https://doi.org/10.48550/arXiv.2512.24440

    work page doi:10.48550/arxiv.2512.24440 2025

  35. [36]

    Mayer, K. J., K. Dagon, and M. J. Molina, 2025: Can Transfer Learning Be Used to Identify Tropical State-Dependent Bias Relevant to Midlatitude Subseasonal Predictability?Artificial Intelligence for the Earth Systems,4 (4), 240 091, https://doi.org/10.1175/AIES-D-24-0091.1

    work page doi:10.1175/aies-d-24-0091.1 2025

  36. [37]

    McPhaden, M. J., S. E. Zebiak, and M. H. Glantz, 2006: ENSO as an integrating concept in Earth science.Science,314 (5806), 1740–1745, https://doi.org/10.1126/science.1132588

    work page doi:10.1126/science.1132588 2006

  37. [38]

    Journal of Climate , author =

    Meinen, C. S., and M. J. McPhaden, 2000: Observations of warm water volume changes in the equatorial Pacific and their relationship to El Ni ˜no and La Ni ˜na.Journal of Climate,13 (20), 3551–3559, https://doi.org/10.1175/1520-0442(2000)013⟨3551:OOWWVC⟩2.0.CO;2. 48

    work page doi:10.1175/1520-0442(2000)013 2000

  38. [39]

    Molina, M. J., and Coauthors, 2023: A Review of Recent and Emerging Machine Learning Applications for Climate Variability and Weather Phenomena.Artificial Intelligence for the Earth Systems,2 (4), 220 086, https://doi.org/10.1175/AIES-D-22-0086.1

    work page doi:10.1175/aies-d-22-0086.1 2023

  39. [40]

    Neelin, J. D., D. S. Battisti, A. C. Hirst, F.-F. Jin, Y. Wakata, T. Yamagata, and S. E. Zebiak, 1998: ENSO theory.Journal of Geophysical Research: Oceans,103 (C7), 14 261–14 290, https://doi.org/10.1029/97JC03424

    work page doi:10.1029/97jc03424 1998

  40. [41]

    Newman, M., and Coauthors, 2016: The Pacific Decadal Oscillation, Revisited.Journal of Climate, 29 (12), 4399–4427, https://doi.org/10.1175/JCLI-D-15-0508.1

    work page doi:10.1175/jcli-d-15-0508.1 2016

  41. [42]

    arXiv preprint arXiv:1907.08956, https://doi.org/10.48550/arXiv.1907.08956

    Odaibo, S., 2019: Tutorial: Deriving the Standard Variational Autoencoder (V AE) Loss Function. arXiv preprint arXiv:1907.08956, https://doi.org/10.48550/arXiv.1907.08956. Pac ¸al, A., B. Hassler, K. Weigel, M.-´A. Fern´andez-Torres, G. Camps-Valls, and V. Eyring, 2025: Understanding European Heatwaves with Variational Autoencoders.EGUsphere,2025, 1–35, h...

    work page doi:10.48550/arxiv.1907.08956 2019

  42. [43]

    Li, 2025: Diversity of La Ni˜na onset.npj Climate and Atmospheric Science,8 (1), 265, https://doi.org/10.1038/s41612-025-01141-6

    Pan, X., and T. Li, 2025: Diversity of La Ni˜na onset.npj Climate and Atmospheric Science,8 (1), 265, https://doi.org/10.1038/s41612-025-01141-6

    work page doi:10.1038/s41612-025-01141-6 2025

  43. [44]

    S., and S

    Passarella, L. S., and S. Mahajan, 2023: Assessing Tropical Pacific–Induced Predictability of Southern California Precipitation Using a Novel Multi-Input Multioutput Autoencoder.Artificial Intelligence for the Earth Systems,2 (4), e230 003, https://doi.org/10.1175/AIES-D-23-0003.1

    work page doi:10.1175/aies-d-23-0003.1 2023

  44. [45]

    Power, S., and Coauthors, 2021: Decadal climate variability in the tropical Pacific: Characteristics, causes, predictability, and prospects.Science,374 (6563), eaay9165, https://doi.org/10.1126/ science.aay9165

    2021

  45. [46]

    Yu, 2014: ENSO indices from sea surface salinity observed by Aquarius and Argo.Journal of Oceanography,70 (4), 367–375, https://doi.org/10.1007/s10872-014-0238-4

    Qu, T., and J.-Y. Yu, 2014: ENSO indices from sea surface salinity observed by Aquarius and Argo.Journal of Oceanography,70 (4), 367–375, https://doi.org/10.1007/s10872-014-0238-4

    work page doi:10.1007/s10872-014-0238-4 2014

  46. [47]

    M., and T

    Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in Tropical Sea Surface Temperature and Surface Wind Fields Associated with the Southern Oscillation/El Ni˜no.Monthly Weather Review, 110 (5), 354–384, https://doi.org/10.1175/1520-0493(1982)110⟨0354:VITSST⟩2.0.CO;2. 49

    work page doi:10.1175/1520-0493(1982)110 1982

  47. [48]

    Camps-Valls, B

    Reichstein, M., G. Camps-Valls, B. Stevens, M. Jung, J. Denzler, N. Carvalhais, and F. Prabhat, 2019: Deep learning and process understanding for data-driven Earth system science.Nature, 566 (7743), 195–204, https://doi.org/10.1038/s41586-019-0912-1

    work page doi:10.1038/s41586-019-0912-1 2019

  48. [49]

    Rodgers, K. B., P. Friederichs, and M. Latif, 2004: Tropical Pacific Decadal Variability and Its Re- lation to Decadal Modulations of ENSO.Journal of Climate,17 (19), 3761–3774, https://doi.org/ 10.1175/1520-0442(2004)017⟨3761:TPDV AI⟩2.0.CO;2

    work page doi:10.1175/1520-0442(2004)017 2004

  49. [50]

    Sakoe, H., and S. Chiba, 2003: Dynamic programming algorithm optimization for spoken word recognition.IEEE Transactions on Acoustics, Speech, and Signal Processing,26 (1), 43–49, https://doi.org/10.1109/TASSP.1978.1163055

    work page doi:10.1109/tassp.1978.1163055 2003

  50. [51]

    Delcroix, and S

    Singh, A., T. Delcroix, and S. Cravatte, 2011: Contrasting the flavors of El Ni ˜no-Southern Oscillation using sea surface salinity observations.Journal of Geophysical Research: Oceans, 116 (C6), https://doi.org/10.1029/2010JC006862

    work page doi:10.1029/2010jc006862 2011

  51. [52]

    Vialard, F

    Srinivas, G., J. Vialard, F. Liu, A. Voldoire, T. Izumo, E. Guilyardi, and M. Lengaigne, 2024: Dominant contribution of atmospheric nonlinearities to ENSO asymmetry and extreme El Ni˜no events.Scientific Reports,14 (1), 8122, https://doi.org/10.1038/s41598-024-58803-3

    work page doi:10.1038/s41598-024-58803-3 2024

  52. [53]

    Montecinos, K

    Takahashi, K., A. Montecinos, K. Goubanova, and B. Dewitte, 2011: ENSO regimes: Reinterpret- ing the canonical and Modoki El Ni ˜no.Geophysical Research Letters,38 (10), https://doi.org/ 10.1029/2011GL047364

    work page doi:10.1029/2011gl047364 2011

  53. [54]

    Tang, Y., and W. W. Hsieh, 2003: Nonlinear modes of decadal and interannual variability of the subsurface thermal structure in the Pacific Ocean.Journal of Geophysical Research: Oceans, 108 (C3), https://doi.org/10.1029/2001JC001236DigitalObjectIdentifier(DOI)

    work page doi:10.1029/2001jc001236digitalobjectidentifier(doi 2003

  54. [55]

    Timmermann, A., and Coauthors, 2018: El Ni ˜no–Southern Oscillation Complexity.Nature, 559 (7715), 535–545, https://doi.org/10.1038/s41586-018-0252-6

    work page doi:10.1038/s41586-018-0252-6 2018

  55. [56]

    E., 1997: The definition of El Ni˜no.Bulletin of the American Meteorological Society, 78 (12), 2771–2778, https://doi.org/10.1175/1520-0477(1997)078⟨2771:TDOENO⟩2.0.CO;2

    Trenberth, K. E., 1997: The definition of El Ni˜no.Bulletin of the American Meteorological Society, 78 (12), 2771–2778, https://doi.org/10.1175/1520-0477(1997)078⟨2771:TDOENO⟩2.0.CO;2

    work page doi:10.1175/1520-0477(1997)078 1997

  56. [57]

    Waliser, D. E., N. E. Graham, and C. Gautier, 1993: Comparison of the highly reflective cloud and outgoing longwave radiation datasets for use in estimating tropical deep convection.Journal 50 of Climate,6 (2), 331–353, https://doi.org/10.1175/1520-0442(1993)006⟨0331:COTHRC⟩2.0. CO;2

    work page doi:10.1175/1520-0442(1993)006 1993

  57. [58]

    Wang, Y., D. M. Blei, and J. P. Cunningham, 2023: Posterior Collapse and Latent Variable Non- Identifiability.arXiv preprint arXiv:2301.00537, https://doi.org/10.48550/arXiv.2301.00537

    work page doi:10.48550/arxiv.2301.00537 2023

  58. [59]

    Dufresne, Y

    Watanabe, M., J.-L. Dufresne, Y. Kosaka, T. Mauritsen, and H. Tatebe, 2021: Enhanced warming constrained by past trends in equatorial Pacific sea surface temperature gradient.Nature Climate Change,11 (1), 33–37, https://doi.org/10.1038/s41558-020-00933-3

    work page doi:10.1038/s41558-020-00933-3 2021

  59. [60]

    Welch, P. D., 1967: The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms.IEEE Transactions on Audio and Electroacoustics,15 (2), 70–73, https://doi.org/10.1109/TAU.1967.1161901

    work page doi:10.1109/tau.1967.1161901 1967

  60. [61]

    N., and C

    Williams, I. N., and C. M. Patricola, 2018: Diversity of ENSO Events Unified by Convective Threshold Sea Surface Temperature: A Nonlinear ENSO Index.Geophysical Research Letters, 45 (17), 9236–9244, https://doi.org/10.1029/2018GL079203

    work page doi:10.1029/2018gl079203 2018

  61. [62]

    Li, J.-Y

    Yang, S., Z. Li, J.-Y. Yu, X. Hu, W. Dong, and S. He, 2018: El Ni ˜no–Southern Oscillation and its impact in the changing climate.National Science Review,5 (6), 840–857, https://doi.org/ 10.1093/nsr/nwy046

    work page doi:10.1093/nsr/nwy046 2018

  62. [63]

    Yeh, S.-W., J.-S. Kug, B. Dewitte, M.-H. Kwon, B. P. Kirtman, and F.-F. Jin, 2009: El Ni ˜no in a changing climate.Nature,461 (7263), 511–514, https://doi.org/10.1038/nature08316

    work page doi:10.1038/nature08316 2009

  63. [64]

    Yu, J.-Y., H.-Y. Kao, T. Lee, and S. T. Kim, 2011: Subsurface ocean temperature indices for Central-Pacific and Eastern-Pacific types of El Ni˜no and La Ni˜na events.Theoretical and Applied Climatology,103 (3), 337–344, https://doi.org/10.1007/s00704-010-0307-6

    work page doi:10.1007/s00704-010-0307-6 2011

  64. [65]

    Yu, J.-Y., Y. Zou, S. T. Kim, and T. Lee, 2012: The changing impact of El Ni ˜no on US winter temperatures.Geophysical Research Letters,39 (15), https://doi.org/10.1029/2012GL052483

    work page doi:10.1029/2012gl052483 2012

  65. [66]

    Murphy and Robert L

    Zebiak, S. E., and M. A. Cane, 1987: A model El Ni ˜no–southern oscillation.Monthly Weather Review,115 (10), 2262–2278, https://doi.org/10.1175/1520-0493(1987)115⟨2262:AMENO⟩2. 0.CO;2. 51

    work page doi:10.1175/1520-0493(1987)115 1987

  66. [67]

    Zhang, C., 2005: Madden-Julian Oscillation.Reviews of Geophysics,43 (2), https://doi.org/ 10.1029/2004RG000158

    work page doi:10.1029/2004rg000158 2005

  67. [68]

    Zhang, Y., J. M. Wallace, and D. S. Battisti, 1997: ENSO-like Interdecadal Variability: 1900– 93.Journal of Climate,10 (5), 1004–1020, https://doi.org/10.1175/1520-0442(1997)010⟨1004: ELIV⟩2.0.CO;2. 52

    work page doi:10.1175/1520-0442(1997)010 1997