pith. sign in

arxiv: 2606.09369 · v1 · pith:NKCPYXRMnew · submitted 2026-06-08 · 🧮 math.NA · cs.NA· physics.ao-ph

Residual Pseudospectra Reveal a Physics-Informed Koopman Backbone for Tropical Pacific Variability and ENSO Prediction

Paula Lorenzo-Sanchez , Matthew J. Colbrook , Antonio Navarra This is my paper

Pith reviewed 2026-06-27 15:32 UTC · model grok-4.3

classification 🧮 math.NA cs.NAphysics.ao-ph
keywords Koopman operatorENSO predictiontropical Pacific SSTresidual pseudospectradynamic mode decompositionclimate variabilityspectral analysisforecasting
0
0 comments X

The pith

Residual minimization isolates a 19-mode Koopman backbone from SST records that reconstructs Nino3.4 variance and improves ENSO forecasts at 8-18 month leads.

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

Tropical Pacific sea-surface-temperature variability involves multiple timescales with ENSO as the main interannual feature. Finite records yield dense, ambiguous spectra when applying Koopman operator learning. The paper combines kernel extended dynamic mode decomposition with residual minimization and pseudospectral analysis, treating the Koopman eigenvalue relation as a consistency test. This process identifies 19 robust residual-minimum frequencies whose spatial modes persist across datasets and sampling choices. These modes form a compact backbone that captures low-frequency to quasi-biennial variability, reconstructs substantial Nino3.4 variance, and delivers skillful out-of-sample predictions especially at longer leads.

Core claim

The residual landscape identifies 19 robust residual-minimum frequencies with coherent spatial modes that persist across products and sampling realizations. Together these modes define a compact Koopman backbone spanning low-frequency modulation through quasi-biennial components, including ENSO-band variability. The surrounding spectral cloud is structured by integer powers and nonlinear combinations of this backbone, forming a residual-ordered Koopman hierarchy. The backbone reconstructs substantial Nino3.4 variance and enables skillful out-of-sample forecasts, with greatest gains at 8-18-month leads.

What carries the argument

The residual-minimization criterion applied to kernel EDMD spectra, using the Koopman eigenvalue relation as a physics-informed consistency test to select and organize modes from finite noisy records.

If this is right

  • The backbone spans low-frequency modulation through quasi-biennial components including ENSO-band variability.
  • The surrounding spectral cloud is structured by integer powers and nonlinear combinations of the backbone modes.
  • The backbone reconstructs substantial Nino3.4 variance from the SST records.
  • Out-of-sample forecast skill improves with the largest gains occurring at 8-18 month leads.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same residual-minimization procedure could be applied to SST records from other ocean basins to test for analogous compact backbones.
  • Evolving the backbone modes in a dynamical model and comparing against independent observations beyond the training period would provide an external test of physical consistency.
  • The residual-ordered hierarchy might support construction of reduced-order predictive models focused only on the backbone frequencies.

Load-bearing premise

That the residual-minimization criterion applied to finite noisy SST records isolates dynamically meaningful operator eigenvalues rather than sampling artifacts or method-specific biases.

What would settle it

An independent check showing whether the 19 selected modes, when evolved outside the training window, satisfy the underlying fluid equations or conserve known invariants would confirm or refute the dynamical content of the backbone.

read the original abstract

Tropical Pacific sea-surface-temperature (SST) variability spans interacting timescales, with the ENSO as its dominant interannual expression. Yet the dynamical structure organizing this variability and underpinning extended-range predictability remains difficult to extract from high-dimensional observations. Koopman operator learning offers spectral coordinates for nonlinear dynamics, yet finite geophysical records often produce dense, sampling-sensitive spectra whose physical content is ambiguous. We show that this apparent redundancy reflects coherent operator-level structure. Combining kernel Extended Dynamic Mode Decomposition with residual minimization and pseudospectral analysis, we use the Koopman eigenvalue relation as a physics-informed consistency test to organize learned spectra. Applied to ERA5 and HadISST tropical Pacific SST anomalies, the residual landscape identifies 19 robust residual-minimum frequencies with coherent spatial modes that persist across products and sampling realizations. Together, these modes define a compact Koopman backbone spanning low-frequency modulation through quasi-biennial components, including ENSO-band variability. The surrounding spectral cloud is structured by integer powers and nonlinear combinations of this backbone, forming a residual-ordered Koopman hierarchy. The backbone reconstructs substantial Nino3.4 variance and enables skillful out-of-sample forecasts, with greatest gains at 8-18-month leads. By embedding dynamical consistency into physics-informed operator learning, the framework turns opaque spectra into robust, interpretable and predictive representations of 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 manuscript proposes combining kernel Extended Dynamic Mode Decomposition (EDMD) with residual minimization and pseudospectral analysis applied to tropical Pacific SST anomaly fields from ERA5 and HadISST. It identifies 19 residual-minimum frequencies whose associated spatial modes persist across the two products and multiple sampling realizations; these modes are asserted to constitute a compact 'Koopman backbone' spanning low-frequency to quasi-biennial scales (including ENSO-band variability). The surrounding spectrum is described as hierarchically organized by integer powers and nonlinear combinations of the backbone. The backbone is shown to reconstruct substantial Niño3.4 variance and to yield improved out-of-sample forecasts, with largest gains at 8–18-month leads.

Significance. If the central claim is substantiated, the work supplies a concrete, reproducible procedure for distilling interpretable, dynamically consistent spectral coordinates from high-dimensional, noisy geophysical records. The explicit use of the Koopman eigenvalue residual as an internal consistency filter, together with cross-product and cross-realization persistence, offers a template that could be applied to other climate variables where dense spectra currently obscure physical structure. Demonstrated forecast skill at sub-seasonal to interannual leads would be of direct operational interest.

major comments (3)
  1. [Abstract, §4] Abstract and §4 (results on residual landscape): the claim that the 19 residual-minimum frequencies constitute dynamically meaningful Koopman eigenvalues rests on persistence across two SST products and sampling realizations, yet no quantitative robustness metric (e.g., frequency overlap fraction, mode cosine similarity, or bootstrap-derived confidence intervals) is reported. Without such a metric it remains possible that the selected frequencies reflect shared spectral biases of the input products rather than operator eigenvalues of the underlying dynamics.
  2. [§3.2, §5] §3.2 (residual minimization procedure) and §5 (forecast evaluation): the residual-minimization step is presented as an independent physics-informed consistency test, but the manuscript does not demonstrate that the selected modes satisfy the Koopman eigenvalue equation (or any derived conservation law) when the learned operator is applied to data strictly outside the fitting window. Forecast skill on held-out periods is shown, yet this does not directly verify the eigenvalue relation itself on those periods.
  3. [§4.3] §4.3 (backbone reconstruction of Niño3.4): the reported reconstruction skill is obtained after the 19 modes have already been selected by the residual criterion on the same data; an ablation that withholds the residual-minimization step and compares against a random or uniformly spaced selection of 19 frequencies is not provided, leaving open the possibility that any compact 19-mode truncation would yield comparable skill.
minor comments (2)
  1. [§3.1] Notation for the residual threshold and kernel bandwidth (listed as free parameters in the axiom ledger) should be introduced with explicit symbols and default values in §3.1 so that the procedure is fully reproducible from the text alone.
  2. [Figure 3] Figure captions for the spatial mode plots should state the exact percentage of Niño3.4 variance captured by each individual mode (or the cumulative backbone) rather than qualitative descriptors.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and insightful comments, which help clarify the presentation of robustness, out-of-sample validation, and the necessity of ablation studies. We address each major comment below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract, §4] Abstract and §4 (results on residual landscape): the claim that the 19 residual-minimum frequencies constitute dynamically meaningful Koopman eigenvalues rests on persistence across two SST products and sampling realizations, yet no quantitative robustness metric (e.g., frequency overlap fraction, mode cosine similarity, or bootstrap-derived confidence intervals) is reported. Without such a metric it remains possible that the selected frequencies reflect shared spectral biases of the input products rather than operator eigenvalues of the underlying dynamics.

    Authors: We agree that quantitative metrics are needed to substantiate persistence. In the revised manuscript we will add frequency overlap fractions (computed as the fraction of selected frequencies within a small tolerance across ERA5/HadISST and across bootstrap realizations), mean cosine similarities of the associated spatial modes, and bootstrap-derived 95% confidence intervals on the residual-minimum frequencies. These will be reported in §4 with corresponding updates to the abstract. revision: yes

  2. Referee: [§3.2, §5] §3.2 (residual minimization procedure) and §5 (forecast evaluation): the residual-minimization step is presented as an independent physics-informed consistency test, but the manuscript does not demonstrate that the selected modes satisfy the Koopman eigenvalue equation (or any derived conservation law) when the learned operator is applied to data strictly outside the fitting window. Forecast skill on held-out periods is shown, yet this does not directly verify the eigenvalue relation itself on those periods.

    Authors: We acknowledge that explicit verification of the eigenvalue residual on data strictly outside the fitting window would provide a stronger test. While out-of-sample forecast skill offers indirect support, we will add in the revision a direct computation of the Koopman residual for the 19 modes on a separate validation window withheld from both mode selection and operator fitting. This additional check will be included in §5. revision: yes

  3. Referee: [§4.3] §4.3 (backbone reconstruction of Niño3.4): the reported reconstruction skill is obtained after the 19 modes have already been selected by the residual criterion on the same data; an ablation that withholds the residual-minimization step and compares against a random or uniformly spaced selection of 19 frequencies is not provided, leaving open the possibility that any compact 19-mode truncation would yield comparable skill.

    Authors: We agree that an ablation is required to isolate the contribution of residual-based selection. In the revised manuscript we will add to §4.3 a direct comparison of Niño3.4 reconstruction skill using the residual-selected 19 modes versus (i) 19 frequencies drawn uniformly at random from the same spectral range and (ii) 19 uniformly spaced frequencies. The ablation will quantify the improvement attributable to the physics-informed selection criterion. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation applies standard kernel EDMD + residual selection with independent out-of-sample validation

full rationale

The paper applies kernel EDMD, computes residuals against the Koopman eigenvalue relation, selects the 19 lowest-residual frequencies, and then demonstrates that the resulting backbone reconstructs Nino3.4 variance and yields out-of-sample forecast skill at 8-18 month leads. These performance metrics are measured on held-out data and across independent SST products, providing external checks that do not reduce to the fitting procedure by construction. No self-citation chain is invoked to establish uniqueness or to smuggle an ansatz; the residual-minimization step is an explicit algorithmic choice whose output is tested rather than presupposed. The derivation therefore remains self-contained against the stated benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the Koopman operator learned from finite SST records admits a sparse residual-minimum spectrum whose modes are dynamically meaningful; this is not derived from first principles but imposed via the residual test. No explicit free parameters are listed in the abstract, but kernel bandwidth, number of modes retained, and residual threshold are implicit choices that shape the backbone.

free parameters (2)
  • kernel bandwidth
    Controls the feature map in kernel EDMD; its value is not stated and must be chosen to produce the reported 19-mode structure.
  • residual threshold
    Defines which frequencies qualify as 'robust residual minima'; the abstract does not specify how this cutoff is set or validated.
axioms (1)
  • domain assumption The Koopman operator for the tropical Pacific SST system is well-approximated by a finite-rank kernel representation on the observed record.
    Invoked when the method is applied to ERA5/HadISST anomalies without reference to the underlying fluid equations.

pith-pipeline@v0.9.1-grok · 5781 in / 1641 out tokens · 18950 ms · 2026-06-27T15:32:51.967838+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

51 extracted references

  1. [1]

    Berry, D

    T. Berry, D. Giannakis, and J. Harlim. Nonparametric forecasting of low-dimensional dynamical systems.Physical Review E, 91(3), Mar. 2015.[2] S. Brunton, B. Brunton, J. Proctor, E. Kaiser, and J. Kutz. Chaos as an intermittently forced linear system.Nature Communications, 8(1):19, 2017.[3] S. Brunton, B. Brunton, J. Proctor, and J. Kutz. Koopman invari- a...

    2015

  2. [2]

    Brunton, M

    S. Brunton, M. Budiˇ si´ c, E. Kaiser, and J. Kutz. Modern koopman theory for dynamical systems.SIAM Review, 64(2):229–340, 2022

    2022

  3. [3]

    S. L. Brunton, B. R. Noack, and P. Koumoutsakos. Machine learning for fluid mechanics.Annual Review of Fluid Mechanics, 52:477–508, 2020.[6] M. Budiˇ si´ c, R. Mohr, and I. Mezi´ c. Applied koopmanism.Chaos, 22(4):047510, 2012

    2020

  4. [4]

    Capotondi, A

    A. Capotondi, A. T. Wittenberg, M. Newman, E. Di Lorenzo, et al. Mechanisms of tropical pacific decadal variability.Nature Reviews Earth & Environment, 2023

    2023

  5. [5]

    Capotondi, A

    A. Capotondi, A. T. Wittenberg, M. Newman, E. Di Lorenzo, J.-Y. Yu, P. Braconnot, J. Cole, B. Dewitte, B. Giese, E. Guilyardi, F.-F. Jin, K. Karnauskas, B. Kirtman, T. Lee, N. Schneider, Y. Xue, and S.-W. Yeh. Understanding ENSO diversity.Bulletin of the American Meteorological Society, 96(6):921–938, 2015

    2015

  6. [6]

    M. D. Chekroun, N. Zagli, and V. Lucarini. Kolmogorov modes and linear response of jump-diffusion models: Applications to stochastic excitation of the ENSO recharge oscillator.Reports on Progress in Physics, 88:127601, 2025

    2025

  7. [7]

    J. C. H. Chiang and D. J. Vimont. Analogous pacific and atlantic meridional modes of tropical atmosphere–ocean variability.Journal of Climate, 17(21):4143–4158, 2004

    2004

  8. [8]

    Colbrook

    M. Colbrook. The multiverse of dynamic mode decomposition algo- rithms. InHandbook of Numerical Analysis, pages 127–230. 2024

    2024

  9. [9]

    Colbrook, L

    M. Colbrook, L. Ayton, and M. Sz˝ oke. Residual dynamic mode decom- position: robust and verified koopmanism.Journal of Fluid Mechanics, 955, 2023

    2023

  10. [10]

    Colbrook, I

    M. Colbrook, I. Mezi´ c, and A. Stepanenko. Adversarial dynami- cal systems characterize when data-driven learning succeeds or fails. Manuscript under minor revision atNature Communications, 2026

    2026

  11. [11]

    M. J. Colbrook. Another look at residual dynamic mode decomposi- tion in the regime of fewer snapshots than dictionary size.Physica D: Nonlinear Phenomena, 469, 2024

    2024

  12. [12]

    M. J. Colbrook, Z. Drmaˇ c, and A. Horning. An introductory guide to koopman learning. In D. Alpay, F. Colombo, and I. Sabadini, editors, Operator Theory, pages 1–49. Springer, Basel, 2026. First online: 28 October 2025; accessed 24 May 2026

    2026

  13. [13]

    M. J. Colbrook and A. Townsend. Rigorous data-driven computation of spectral properties of koopman operators for dynamical systems. Communications on Pure and Applied Mathematics, 77(1):221–283, 2024.[17] Z. Drmaˇ c and I. Mezi´ c. A data driven koopman-schur decomposi- tion for computational analysis of nonlinear dynamics.arXiv preprint, abs/2312.15837, ...

    arXiv 2024

  14. [14]

    Drmaˇ c, I

    Z. Drmaˇ c, I. Mezi´ c, and R. Mohr. Data driven modal decompositions: analysis and enhancements.SIAM Journal on Scientific Computing, 40(4):A2253–A2285, 2018. 12

    2018

  15. [15]

    Froyland, D

    G. Froyland, D. Giannakis, B. R. Lintner, M. Pike, and J. Slawin- ska. Spectral analysis of climate dynamics with operator-theoretic approaches.Nature Communications, 12(1):6570, Nov. 2021

    2021

  16. [16]

    Froyland, D

    G. Froyland, D. Giannakis, E. Luna, and J. Slawinska. Revealing trends and persistent cycles of non-autonomous systems with autonomous operator-theoretic techniques.Nature Communications, 15(1):4268, 2024.[21] G. Froyland, G. A. Gottwald, and A. Hammerlindl. A computa- tional method to extract macroscopic variables and their dynamics in multiscale systems...

    2024

  17. [17]

    Froyland, S

    G. Froyland, S. Lloyd, and N. Santitissadeekorn. Coherent sets for nonautonomous dynamical systems.Physica D: Nonlinear Phenom- ena, 239(16):1527–1541, 2010

    2010

  18. [18]

    Ghil and V

    M. Ghil and V. Lucarini. The physics of climate variability and climate change.Reviews of Modern Physics, 92(3):035002, 2020

    2020

  19. [19]

    Giannakis

    D. Giannakis. Data-driven spectral decomposition and forecasting of ergodic dynamical systems.Applied and Computational Harmonic Analysis, 47(2):338–396, 2019

    2019

  20. [20]

    Giannakis and C

    D. Giannakis and C. Valva. Consistent spectral approximation of koopman operators using resolvent compactification.Nonlinearity, 37(7):075021, 2024

    2024

  21. [21]

    D. A. Haggerty, M. J. Banks, E. Kamenar, A. B. Cao, P. C. Cur- tis, I. Mezi´ c, and E. W. Hawkes. Control of soft robots with inertial dynamics.Science Robotics, 8(81):eadd6864, 2023

    2023

  22. [22]

    Hersbach, B

    H. Hersbach, B. Bell, P. Berrisford, S. Hirahara, A. Hor´ anyi, J. Mu˜ noz- Sabater, J. Nicolas, C. Peubey, R. Radu, D. Schepers, A. Simmons, C. Soci, S. Abdalla, X. Abellan, G. Balsamo, P. Bechtold, G. Biavati, J. Bidlot, M. Bonavita, G. De Chiara, P. Dahlgren, D. Dee, M. Dia- mantakis, R. Dragani, J. Flemming, R. Forbes, M. Fuentes, A. J. Geer, L. Haimb...

    2020

  23. [23]

    Jiang, J

    N. Jiang, J. D. Neelin, and M. Ghil. Quasi-quadrennial and quasi- biennial variability in the equatorial pacific.Climate Dynamics, 12:101–112, 1995

    1995

  24. [24]

    S. Klus, F. N¨ uske, P. Koltai, H. Wu, I. Kevrekidis, C. Sch¨ utte, and F. No´ e. Data-driven model reduction and transfer operator approxima- tion.Journal of Nonlinear Science, 28(3):985–1010, 2018

    2018

  25. [25]

    B. Koopman. Hamiltonian systems and transformation in hilbert space. Proc. Natl. Acad. Sci. USA, 10:315–318, 1931

    1931

  26. [26]

    Koopman and J

    B. Koopman and J. Neumann. Dynamical systems of continuous spectra.Proc. Natl. Acad. Sci. USA, 18:255–263, 1932

    1932

  27. [27]

    Lewin and E

    M. Lewin and E. Ser` e. Spectral pollution and how to avoid it. Proceedings of the London Mathematical Society, 100(3):864–900, 2010.[34] J. Lohmann, H. A. Dijkstra, M. Jochum, V. Lucarini, and P. D. Ditlevsen. Multistability and intermediate tipping of the atlantic ocean circulation.Science Advances, 10(12):eadi4253, 2024

    2010

  28. [28]

    Lorenzo-S´ anchez, M

    P. Lorenzo-S´ anchez, M. Newman, J. Albers, A. Subramanian, and A. Navarra. Koopman theory for enhanced tropical pacific sst forecast- ing.Artif. Intell. Earth Syst., 4, 2025

    2025

  29. [29]

    Lucarini and M

    V. Lucarini and M. D. Chekroun. Theoretical tools for understanding the climate crisis from Hasselmann’s programme and beyond.Nature Reviews Physics, 5(12):744–765, 2023

    2023

  30. [30]

    Lusch, J

    B. Lusch, J. Kutz, and S. Brunton. Deep learning for universal lin- ear embeddings of nonlinear dynamics.Nature Communications, 9(1), 2018.[38] M. e. a. L’Heureux. Enso prediction.Geophysical Monograph Series, pages 227–246, 2020

    2018

  31. [31]

    I. Mezi´ c. Analysis of fluid flows via spectral properties of the koopman operator.Annual Review of Fluid Mechanics, 45:357–378, 2013

    2013

  32. [32]

    I. Mezi´ c. Spectral properties of dynamical systems, model reduction and decompositions.Nonlinear Dynamics, 41, 2005

    2005

  33. [33]

    Mohr and I

    R. Mohr and I. Mezi’c. Construction of eigenfunctions for scalar- type operators via laplace averages with connections to the koopman operator.arXiv: Spectral Theory, 2014

    2014

  34. [34]

    NOAA confirms 4th global coral bleaching event, 2024

    National Oceanic and Atmospheric Administration. NOAA confirms 4th global coral bleaching event, 2024. Accessed 2026-05-18

    2024

  35. [35]

    Navarra, J

    A. Navarra, J. Tribbia, and S. Klus. Estimation of koopman transfer operators for the equatorial pacific sst.Journal of the Atmospheric Sciences, 78(4):1227–1244, 2021

    2021

  36. [36]

    Navarra, J

    A. Navarra, J. Tribbia, S. Klus, and P. Lorenzo-S´ anchez. Variability of sst through koopman modes.Journal of Climate, 37(16):4095–4114, 2024.[45] M. Newman, M. A. Alexander, and J. D. Scott. An empirical benchmark for decadal forecasts of global surface temperature anomalies.Clim Dyn, 37, 2011

    2024

  37. [37]

    Orvieto et al

    A. Orvieto et al. Resurrecting recurrent neural networks for long sequences.Proceedings of Machine Learning Research, 202:26670– 26698, 2023. Proceedings of the 40th International Conference on Machine Learning

    2023

  38. [38]

    W. Pan, C. W. Rowley, and D. M. Luchtenburg. Efficient extraction of dynamically relevant modes in large datasets.Physical Review E, 100(3):032209, 2019

    2019

  39. [39]

    Penland and P

    C. Penland and P. D. Sardeshmukh. The optimal growth of tropical sea surface temperature anomalies.Journal of Climate, 8(8):1999–2024, 1995.[49] J. Proctor and P. Eckhoff. Discovering dynamic patterns from infectious disease data using dynamic mode decomposition.International Health, 7(2):139–145, 2015

    1999

  40. [40]

    N. A. Rayner, D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexan- der, D. P. Rowell, E. C. Kent, and A. Kaplan. Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century.Journal of Geophysical Research: Atmospheres, 108(D14), 2003

    2003

  41. [41]

    C. W. Rowley, I. Mezi´ c, S. Bagheri, P. Schlatter, and D. S. Henning- son. Spectral analysis of nonlinear flows.Journal of Fluid Mechanics, 641:115–127, 2009

    2009

  42. [42]

    Sch¨ utte, P

    C. Sch¨ utte, P. Koltai, and S. Klus. On the numerical approximation of the perron-frobenius and koopman operator.Journal of Computational Dynamics, 3(1):1–12, Sep 2016

    2016

  43. [43]

    Slawinska and D

    J. Slawinska and D. Giannakis. Indo-pacific variability on seasonal to multidecadal time scales. part i: Intrinsic sst modes in models and observations.Journal of Climate, 30(14):5265–5294, 2017

    2017

  44. [44]

    L. N. Trefethen and M. Embree.Spectra and Pseudospectra: The Behavior of Nonnormal Matrices and Operators. Princeton University Press, Princeton, NJ, 2005

    2005

  45. [45]

    J. H. Tu, C. W. Rowley, D. M. Luchtenburg, S. B. Brunton, and J. N. Kutz. On dynamic mode decomposition: Theory and applications. Journal of Computational Dynamics, 1(2):391, 2014

    2014

  46. [46]

    Valva and E

    C. Valva and E. Gerber. The qbo, the annual cycle, and their inter- actions: Isolating periodic modes with koopman analysis.Journal of Climate, 38(14):3437–3452, 2025

    2025

  47. [47]

    D. J. Vimont, J. M. Wallace, and D. S. Battisti. The seasonal foot- printing mechanism in the pacific: Implications for ENSO.Journal of Climate, 16(16):2668–2675, 2003

    2003

  48. [48]

    X. Wang, J. Slawinska, and D. Giannakis. Extended-range statisti- cal enso prediction through operator-theoretic techniques for nonlinear dynamics.Scientific Reports, 10(1), 2020

    2020

  49. [49]

    M. O. Williams, I. G. Kevrekidis, and C. W. Rowley. A data–driven approximation of the koopman operator: Extending dynamic mode decomposition.Journal of Nonlinear Science, 25(6):1307–1346, Dec 2015.[60] M. O. Williams, C. W. Rowley, and I. G. Kevrekidis. A kernel- based method for data-driven koopman spectral analysis.Journal of Computational Dynamics, 2(...

    2015

  50. [50]

    Anticipating the impact of drought in South- ern African countries: Key findings from 2023/2024, 2025

    World Food Programme. Anticipating the impact of drought in South- ern African countries: Key findings from 2023/2024, 2025. Accessed 2026-05-18.[62] World Meteorological Organization. El Ni˜ no weakens but impacts continue, 2024. Accessed 2026-05-18

    2023

  51. [51]

    Zhang, C

    R.-H. Zhang, C. Gao, and L. Feng. Recent ENSO evolution and its real- time prediction challenges.National Science Review, 9(4), 03 2022. 13 Supplementary Material This document contains supplementary figures, tables, and methodological details supporting the main manuscript. The supplementary material provides additional evidence for the residual and pseu...

    2022