Extended Recharge Oscillator (XRO) Framework

• The Extended Recharge Oscillator (XRO) framework is a physically interpretable model that captures ENSO variability using minimal deterministic and stochastic equations.
• It introduces multiplicative state-dependent noise and intermittent jump forcing to generate stochastic chaos and realistic ENSO event statistics.
• Embedding the XRO framework in data-driven forecasting enhances predictability diagnostics and climate sensitivity assessments through robust modal analysis.

The Extended Recharge Oscillator (XRO) framework is a mathematically rigorous and physically interpretable model family describing El Niño–Southern Oscillation (ENSO) dynamics and their response to stochastic, nonlinear, and non-Gaussian perturbations. XRO builds on the classical Jin recharge oscillator by introducing both multiplicative state-dependent noise and intermittent jump forcing, resulting in a stochastic chaos regime with rich variability, enhanced predictability diagnostics, and robust characterization of climate sensitivity. The framework encompasses minimal, yet realistic dynamical equations and provides a foundation for physically guided machine learning approaches in climate prediction.

1. Mathematical Formulation and Extensions

The minimal XRO in Han et al. (Han et al., 12 Jun 2025) retains two state variables:

• $T \equiv T_E(t)$: eastern-Pacific SST anomaly (e.g., Niño 3)
• $h \equiv h_w(t)$: western-Pacific thermocline depth anomaly

The governing equations integrate deterministic and stochastic nonlinearities: $$\begin{aligned} \frac{dT}{dt} &= (R_0 - R_a \cos(\omega_a t - \phi))\, T + F_1\, h + b_T\, T^2 - c_T\, T^3 - b_h\, T^2 + \sigma_T\, (1 + B\, H(T)\, T) \, \xi_T(t) \ \frac{dh}{dt} &= -\frac{1}{\tau_h}\, h - F_2\, T - b_h\, T^2 + \sigma_h\, \xi_h(t) \end{aligned}$$ where $H(T)$ is the Heaviside function, and $\xi_T,\xi_h$ are independent Gaussian white‑noise processes.

In the jump-diffusion generalization (Chekroun et al., 2024), the system is augmented by discrete state-dependent Poisson jumps representing westerly wind bursts: $$\dot{X}_t = \mathbf{F}(X_t) + \sigma\, \dot{W}_t + D\, \mathbf{B}(X_t)\, f(t)$$ with $\mathbf{B}$ tailored to mimic nonlinear feedback mechanisms, and $f(t)$ a random Bernoulli pulse train indicating jump occurrences.

Recent large-scale multivariate XRO systems (Zhang et al., 18 Jan 2026) embed ten physically interpretable climate modes:

• $$X(t) = [T_{ENSO}, h_{WWV}, T_{NPMM}, \ldots, T_{SASD}]^T$$ with

$\frac{dX}{dt} = L \cdot X + N_{XRO}(X)$

where $L$ encodes linear cross-basin teleconnections and $N_{XRO}$ includes nonlinear ENSO and Indian Ocean Dipole quadratic couplings.

2. Physical Interpretation and Parameter Choices

Physically, XRO captures key features of observed ENSO statistics through its parameterization:

• Strong damping ($\tau_h \sim 7.7$ months) ensures rapid decay of disturbances and suppresses spurious bimodal histograms.
• Quadratic and cubic terms model amplitude skew (El Niño events are stronger and shorter), finite saturation, and the persistence of La Niña.
• State-dependent multiplicative noise ($B\, H(T)\, T$) represents WWB-SST feedback, crucial for realistic transition probabilities and spectral properties.
• Seasonal modulation ($R_a$) imposes phase locking to boreal winter and generates combination tones.

Parameter sweep experiments (Han et al., 12 Jun 2025) identify optimal dimensional values (see Table below):

Parameter	Value	Physical Meaning
$R_0$	$0.03$ month$^{-1}$	Steady Bjerknes growth rate
$R_a$	$0.16$ month$^{-1}$	Seasonal modulation amplitude
$\tau_h$	$7.7$ months	Ocean damping timescale
$F_1$	$0.015$ K·m$^{-1}$·month$^{-1}$	SST→thermocline coupling
$F_2$	$1.45$ m·K$^{-1}$·month$^{-1}$	Recharge→SST coupling
$b_T$	$0.023$ K$^{-1}$·month$^{-1}$	Amplitude asymmetry
$c_T$	$0.001$ K$^{-2}$·month$^{-1}$	Cubic damping
$b_h$	$0.31$ m·K$^{-2}$·month$^{-1}$	La Niña persistence
$B$	$0.45$ K$^{-1}$	Multiplicative noise coef.
$\sigma_T$, $\sigma_h$	$0.18$, $1.6$ units	Noise amplitudes

A plausible implication is that these minimal components suffice to reproduce observed ENSO variance, autocorrelation, skew, transition processes, and kurtosis.

3. Stochastic Chaos and Kolmogorov Modes

The introduction of state-dependent jump and diffusive noise (Chekroun et al., 2024) leads the XRO to stochastic chaos—a regime marked by heavy-tailed event statistics, enhanced spectral gap, and stretched–folded invariant densities. The Kolmogorov operator $\mathcal{L}_K$ associated with the jump-diffusion SDE governs the time evolution of observables and density:

$$\mathcal{L}_K \psi = \mathbf{F}(x) \cdot \nabla\psi + \frac{\sigma^2}{2}\, \Delta\psi + f_r \left[\psi(x + D \mathbf{B}(x)) - \psi(x)\right]$$

Eigen-decomposition yields Kolmogorov modes ($\phi_k$) and resonances ($\lambda_k$), forming a modal basis for decomposing correlations and forced responses. Modes are interpretable as principal directions––and decay rates––of phase-space variability.

4. Fluctuation–Dissipation Theory and Linear Response

XRO generalizes linear response (FDT) theory to mixed jump-diffusion systems. The system’s response to small deterministic drift perturbations $\epsilon\, g(t)\, \mathbf{G}(x)$ is encoded in Green’s functions $G_{\Psi, G}(t)$, which admit modal expansions:

$$\delta^{(1)}[\Psi](t) = \epsilon \int_{-\infty}^t G_{\Psi,\,G}(t-s)\, g(s)\, ds$$

with

$$G_{\Psi, G}(t) = \Theta(t) \int_{\mathbb{R}^2} (e^{t\mathcal{L}_K} \Psi)(x) \left[ - \nabla\cdot (\mathbf{G} \rho_{eq}) \right](x) dx$$

This decomposition enables accurate prediction of system mean response (e.g., to climate sensitivity or anthropogenic forcings) even under stochastic chaos induced by jumps.

5. Application to ENSO Prediction and Skill Horizons

XRO variables provide a foundation for skillful, physically guided data-driven forecasting (Zhang et al., 18 Jan 2026). Physics-guided Deep Echo State Networks (DESN) use XRO climate modes as input, embedding their cross-basin couplings and nonlinear recharge processes within reservoir computing architectures. Mechanistic experiments demonstrate that:

• Extended predictability (16–20 month Niño3.4 skill) arises from nonlinear coupling of subsurface heat content (WWV) to remote basin modes.
• Finite predictability horizons (30–34 months) reflect the intrinsic nonlinear error growth regimes captured by XRO dynamics.
• Sparse nonlinear XRO extensions (SN-XRO) replicate DESN skill decay, confirming the key mechanistic role of higher-order WWV–inter-basin couplings.

6. Modal Diagnostics and Climate Sensitivity

By decomposing Green’s function responses into Kolmogorov modes, XRO offers a robust approach to diagnosing climate sensitivity and attribution. Modal “fingerprints” identify dominant directions of system response and error growth, informing risk assessment for climate tipping points and high-impact extremes. The Ulam method enables direct extraction of Kolmogorov modes from data, yielding explicit bases for natural and forced variability.

7. Conceptual and Computational Significance

The XRO framework synthesizes minimal physical ingredients necessary for realistic ENSO simulation:

• Strong ocean damping with stochastic WWB noise and weak nonlinearities favor a damped, stochastically forced regime matching observed spectral peaks, transition asymmetries, and kurtosis.
• Jump-diffusion extensions bridge conceptual oscillators and stochastic climate modeling, supporting rigorous fluctuation–dissipation analysis and robust statistical prediction.
• Embedding XRO into reservoir computing structures generalizes its interpretability and extends skillful lead-times with low computational cost.

This synthesis advances Hasselmann’s stochastic climate paradigm by supplying a comprehensive, mode-resolved theory for ENSO’s complexity, variability, and predictability (Chekroun et al., 2024, Han et al., 12 Jun 2025, Zhang et al., 18 Jan 2026).