Stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "Stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles".
Mira: Here's the gist of the paper. Stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles models the stochastic behavior of paleoclimate time series to evaluate intervariable coupling during Pleistocene glacial cycles.
Kai: The central challenge of understanding the pacing of the of Earth’s ice ages during the last 800,000 years is to discern the relative importance of external forcing associated with orbital eccentricity, obliquity and precession, and internal dynamics associated with the processes of the atmosphere, biosphere, cryosphere and oceans.
Mira: This study provides a comprehensive overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in shaping Earth’s climate evolution.
Kai: Stochastic Characterization of Noise Structure The authors first analyze the structure of the noise in temperature, ice volume, CO2, CH4, and N2O using a method that provides colored-noise structure on multiple timescales.
Paper summary: Kai: We just covered how this paper uses stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles to model the stochastic behavior of paleoclimate time series.
Mira: The thesis here is that understanding the pacing of Earth’s ice ages requires discerning the relative importance between external forcing from orbital eccentricity, obliquity, and precession versus internal dynamics involving the atmosphere, biosphere, cryosphere, and oceans.
Lev: It sets out to evaluate this coupling during those Pleistocene glacial cycles by using a comprehensive overview of glacial-interglacial dynamics and highlighting the role of cryosphere-climate feedbacks in shaping Earth’s climate evolution.
Kai: To do this, they start by analyzing the structure of the noise in temperature, ice volume, carbon dioxide, methane, and nitrogen dioxide using a method that provides colored-noise structure on multiple timescales.
Mira: Specifically, they employ multifractal time-weighted detrended fluctuation analysis or MFTWDFA to quantify fluctuations around the mean behavior across various timescales present in the data.
Lev: This MFTWDFA involves four steps, including constructing a nonstationary profile Y as Y equals X i minus X i k equal one X k minus X i, where X is the mean of the time series.
Kai: The dependence of Fq on the timescale s encodes the scaling properties of the time series, with Fq s proportional to s h q where h denotes generalized Hurst exponents.
Mira: They then formulate Ornstein–Uhlenbeck processes to model paleoclimate time series, treating them as periodic nonautonomous Langevin equations that explicitly account for both the deterministic behavior and stochastic variability of the data.
Lev: The one-variable model is described by d eta i dt equals a i t eta i t plus N i t xi i t, where a I t drives the mean-reverting drift and N I t represents temporal modulation of the stochastic forcing.
Kai: For the five-variable model, they construct it as d eta i dt equals a I t eta i t plus N I t xi i t plus X j not equal i b i j (t) h eta j (t) minus eta i (t), where b I j t is a linearized diffusive coupling term.
Mira: The time-dependent coefficients a I and N I are determined using a modified version of the methodology introduced by Moon and Wettlaufer.
Lev: They then use this five-variable model to extract dynamical coherence by requiring their evolution to be described consistently within a common system of coupled stochastic equations.
Kai: The coupling coefficients b i j t represent the influence exerted by variable j on variable i, and their sign determines the direction of influence.
Mira: They observe that the coupling coefficients associated with delta T to ice volume and carbon dioxide to ice volume are negative, meaning those two things facilitate the growth of ice volume.
Lev: Furthermore, they observe oscillatory dynamics in the response functions arising from opposite-sign cross-couplings between ice volume and delta T, and between ice volume and carbon dioxide.
Kai: This study provides a comprehensive overview of glacial-interglacial dynamics by showing how these variables interact through these coupled stochastic equations.
Mira: It highlights the crucial role of cryosphere-climate feedbacks in shaping Earth’s climate evolution during this critical time period.
Conclusion: Kai: So, looking at the paper "Stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles" by Patra, Giorgini, and Wettlaufer, we see that this approach successfully models the stochastic behavior of those paleoclimate time series.
Mira: The authors are essentially providing a systematic stochastic modeling framework for quantifying glacial-interglacial climate dynamics based on these coupled systems.
Lev: In simple terms, what they’re telling us is that ice volume functions as a slow integrator and regulator of climate forcing, while temperature and greenhouse gases act as faster amplifying feedback mechanisms that shape the amplitude and timing of those cycles.
Kai: This framework lets us systematically constrain tests how these variables interact over geological timescales by looking at the noise characteristics in the data.
Mira: The implication is that this method can be used to test complex climate models across a wide range of past climate states, provided we keep in mind their limitations for now.
Lev: Their limitation is that they are focusing on stochastic variability and don't incorporate nonlinear intervariable coupling or multiplicative or correlated noise which we think are important for the inherently nonlinear nature of the system.
Kai: It’s a promising way to constrain those models, even if we have to keep those more complex terms in mind for future work.
Nordita · Massachusetts Institute of Technology · Yale University
physics.ao-ph
Submitted: 2026-03-27
Updated: 2026-08-28
Comments: 23 pages, 8 figures
Journal ref: Chaos vol. 36, 113115 (2026)
DOI: 10.1063/5.0336001
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 91/100
The gist: Stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles models the stochastic behavior of paleoclimate time series to evaluate intervariable coupling during…
Key concepts
- Stochastic Coupling
- This method treats climate variables as interacting random systems. It models their combined behavior using coupled stochastic equations to understand how fluctuations in one variable influence another over time, capturing the complex, noisy nature of climate change.
- Ornstein–Uhlenbeck Processes
- These are mathematical models used to describe processes that tend to revert to a mean value. The authors use them as Langevin equations for paleoclimate data, allowing them to model both the predictable, slow drift and the random, noisy variability observed in temperature and ice volume records.
- Intervariable Coupling
- This refers to the direct influence one climate variable has on another. The study found that temperature and CO2 negatively couple with ice volume, meaning they facilitate its growth. This coupling also creates oscillatory dynamics between variables like ice volume and temperature.
Terminology
Summary
Stochastic coupling of climate variables and ice volume over the Late Pleistocene glacial cycles models the stochastic behavior of paleoclimate time series to evaluate intervariable coupling during Pleistocene glacial cycles. p. 0 Researchers have reconstructed consistent time
series for ice volume, temperature, and carbon dioxide by applying inverse forward modeling to
benthic oxygen isotope records. In this paper, we model the stochastic behavior of paleoclimate
time series to evaluate the coupling between climate variables during the Pleistocene glacial cycles. This study provides a comprehensive
overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in
shaping Earth’s climate evolution. The central challenge of understanding the pacing of the of Earth’s ice ages during the last
800,000 years is to discern the relative importance of external forcing associated with orbital
eccentricity, obliquity and precession, and internal dynamics associated with the processes of the
atmosphere, biosphere, cryosphere and oceans. To test our model, we compute
response functions for each pair of variables and compare them with empirical data, confirming
our predictions regarding intervariable stability and coupling. This study provides a comprehensive
overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in
shaping Earth’s climate evolution.Where the paper says it
Where the paper says it
Where the paper says it
Stochastic Characterization of Noise Structure
The authors first analyze the structure of the noise in temperature, ice volume, CO2, CH4, and N2O using a method that provides colored-noise structure on multiple timescales. They treat their co-variations as a coupled non-autonomous stochastic dynamical system. p. 0 Here we first analyze
the structure of the noise in temperature, ice
volume, CO2, CH4, and N2O using a method
that provides colored-noise structure on multiple
timescales. We then treat their co-variations as
a coupled non-autonomous stochastic dynamical
system.Where the paper says it
-
They employ multifractal time-weighted detrended fluctuation analysis (MFTWDFA) to quantify the fluctuations around the mean behavior across various timescales present in the data. p. 4
Where the paper says it
Background We employ multifractal time-weighted detrended fluctuation analysis (MFTWDFA) [42] to analyze the scaling dynamics and fluctuation structure in the paleoclimate time series. Using this method, we quantify the fluctuations around the mean behavior across various timescales present in the data through a fluctuation function, introduced later in this section.
-
The MFTWDFA algorithm involves four steps, including constructing a nonstationary profile Y (i) as Y (i) ≡ X i k=1 Xk − X i, where X is the mean of the time series. p. 5
Where the paper says it
Algorithm MFTWDFA algorithm involves four steps, as described below: 1. Given the original time series Xi , we first construct a nonstationary profile Y (i) as Y (i) ≡ X i k=1 Xk − X , i = 1, 2, .
-
The dependence of Fq(s) on the timescale s encodes the scaling properties of the time series, with Fq(s) ∝ s h(q), where h(q) denotes the generalized Hurst exponents. p. 5
Where the paper says it
Results: Data Analysis The dependence of Fq(s) on the timescale s encodes the scaling properties of the time series. For a given statistical moment q, the generalized fluctuation function exhibits power-law scaling with s, such that Fq(s) ∝ s h(q) , where h(q) denotes the generalized Hurst exponents.
Modeling Stochastic Dynamics
The authors formulate Ornstein–Uhlenbeck processes to model paleoclimate time series, treating them as periodic, nonautonomous Langevin equations that explicitly account for both the deterministic behavior and stochastic variability of the data.
-
The one-variable model is described by dηi(t) dt = ai(t)ηi(t) + Ni(t)ξi(t), where a i (t) drives the mean-reverting drift and N i (t) represents temporal modulation of the stochastic forcing. p. 7
Where the paper says it
The evolution of the ith variable, denoted as ηi(t), is described by; dηi(t) dt = ai(t)ηi(t) + Ni(t)ξi(t). (7) The time-periodic deterministic term, ai(t), drives the mean-reverting drift and characterizes the stability of the system.
-
The five-variable model is constructed as dηi(t) dt = ai(t)ηi(t) + Ni(t)ξi(t) + X j≠i b i j (t) h ηj (t)-ηi (t) i, where b i j (t) is a linearized diffusive coupling term. p. 9
Where the paper says it
10 where bij (t) is a linearized diffusive coupling term that captures the influence of ηj (t) on ηi(t). This form of coupling, expressed as the difference between interacting variables, has been employed in a broad range of systems (see for example Othmer and Scriven [60], Levin [61], Kopell and Howard [62], and Krause et al.
-
The time-dependent coefficients ai(t) and Ni(t) are determined using a modified version of the methodology introduced by Moon and Wettlaufer [58]. p. 8
Where the paper says it
The framework provides a coherent means of analyzing the observed multi-scale temporal variability in the filtered paleoclimate records. We determine the time-dependent coefficients ai(t) and Ni(t) using a modified version of the methodology introduced by Moon and Wettlaufer [58].
Intervariable Coupling and Feedbacks
The five-variable model allows for the extraction of dynamical coherence by requiring their evolution to be described consistently within a common system of coupled stochastic equations.
-
The coupling coefficients bij (t) represent the influence exerted by variable j on variable i, and their sign determines the direction of influence. p. 9
Where the paper says it
By construction, the coupling coefficient bij (t) (bji(t)), which connects variable j to variable i, or symbolically j → i, represents the influence exerted by variable j on variable i (variable i on variable j). Thus, |bij (t)| measures the interaction strength, and the sign of bij (t) determines the direction of the influence.
-
The coupling coefficients associated with ∆T → IV and CO2 → IV are negative, meaning temperature and carbon dioxide facilitate the growth of ice volume. p. 9
Where the paper says it
In both models, methane and nitrous oxide exhibit greater stability compared to carbon dioxide, temperature and ice volume (IV), implying that their deterministic drift exerts a stronger influence, driving them toward the long-term mean behavior. Coupling coefficients Figure 4(a) shows the time-dependent coupling coefficients for the five-variable model.
-
The authors observe oscillatory dynamics in the response functions arising from opposite-sign cross-couplings between IV and ∆T, and IV and CO2. p. 0
Where the paper says it
We then treat their co-variations as a coupled non-autonomous stochastic dynamical system. Important results are oscillatory dynamics associated with the couplings between ice volume, temperature and CO2, with a strong causal relationship between the latter driving the former two, which is the signature of the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions.
Model Fidelity and Interpretation
The model coefficients are used to generate synthetic time series forward in time using a standard Euler scheme to assess how effectively the simulations capture the various features of the observed data.
-
Figure 5(a) presents a comparison of the periodic standard deviations of the data and those produced by the models, showing good overall agreement in amplitude and periodic structure. p. 7
Where the paper says it
This periodicity allows us to determine the model coefficients from the periodic statistics of the data. The period of the coefficients based on (a) the power spectra of the time series, (b) the periodicities associated with Milankovitch cycles, and (c) the timescale-separated noise structure revealed by MFTWDFA.
-
The response-function analysis indicates that atmospheric CO2 exerts a more pronounced influence on ice volume (IV) than does temperature. p. 9
Where the paper says it
However, the five-variable model shows slightly more negative values for all five variables, suggesting that inter-variable couplings enhance overall stability. In both models, methane and nitrous oxide exhibit greater stability compared to carbon dioxide, temperature and ice volume (IV), implying that their deterministic drift exerts a stronger influence, driving them toward the long-term mean behavior.
-
The explicit representation of cryosphere–climate feedbacks allows the model to capture delayed cryosphere-climate feedbacks and permit damped oscillatory response modes that were not reproduced in earlier models. p. 0
Where the paper says it
To test our model, we compute response functions for each pair of variables and compare them with empirical data, confirming our predictions regarding intervariable stability and coupling. This study provides a comprehensive overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in shaping Earth’s climate evolution.
Conclusion on Climate Dynamics
The analysis supports the perspective that ice volume functions as a slow integrator and regulator of climate forcing, while temperature and greenhouse gases serve as faster amplifying feedback mechanisms that shape the amplitude, persistence, and timing of Late Pleistocene glacial cycles. The authors provide a systematic stochastic modeling framework for quantifying glacial–interglacial climate dynamics. p. 7 The results obtained from MFTWDFA support the applicability of a stochastic modeling framework for analyzing the climate variables and ice volume time series
during the Late Pleistocene. For all five time series, the
contrast between short- and long-timescale dynamics reveals a pronounced separation of timescales between subglacial and super-glacial periods. In contrast, Zhang et al. [8] propose that the nonlinear interaction between eccentricity-modulated precession and the
internal coupling between ice volume and the carbon cycle explain the 100-kyr cycle, and Hobart et al. For example, Shao and Ditlevsen [34] applied
a multifractal approach to Antarctic and Greenland icecore data, demonstrating that the Holocene climate is
predominantly monofractal, while the glacial climate exhibits multifractal properties, and concluded that the
glacial climate has a longer persistence time and stronger
nonlinearities. Using a similar multifractal method, we
have recently assessed the scaling dynamics of temperature and three greenhouse gas time series extracted from
ice cores drilled at Dome C in Antarctica by the European Project for Ice Coring in Antarctica (EPICA)
project [35].Where the paper says it
Where the paper says it
Where the paper says it
Appendix A: Multi-scale Analysis of Figure 3
-
On short timescales, the coefficients of the nonautonomous Ornstein–Uhlenbeck process dX = a(t)X dt + N(t) dWt (A1) change very little throughout a window of duration s < P. p. 7
Where the paper says it
Importantly, the Hurst exponent for timescales shorter than the smoothing window—approximately 3 kyr—remains unchanged, since the high-pass filter does not suppress variability below the smoothing window. These results underlie our use of a multidimensional nonautonomous Ornstein–Uhlenbeck process to model the data across a range of timescales.
-
The red-noise response to white noise forcing is reflected in the power spectrum S(f) ∼ f −β, where β = 2, yielding h(q = 2) = 1 + β/2 = 3/2. p. 6
Where the paper says it
a power spectrum of the form S(f) ∝ f −β , where f is frequency and β is the spectral decay exponent, the relationship h(2) = (1 + β)/2 holds [56]. Consequently, for white noise, characterized by β = 0, it yields h(2) = 1/2; in the case of pink noise, where β = 1, giving h(2) = 1; for red noise, with β = 2, this results in h(2) = 3/2, and so forth.
-
On long timescales, the flattening observed in Fig. p. 7
3(b) is produced by the explicit suppression of low-frequency power introduced by the high-pass filter. p. 7Where the paper says it
The disappearance of the long-timescale scaling after filtering is therefore expected and confirms that the slowly varying deterministic component has been successfully removed. As the length of the smoothing window is reduced, higher-frequency components are preferentially removed, causing the observed crossover to shift toward shorter timescales, as shown in Figure 3(b).
Where the paper says it
As the length of the smoothing window is reduced, higher-frequency components are preferentially removed, causing the observed crossover to shift toward shorter timescales, as shown in Figure 3(b). Importantly, the Hurst exponent for timescales shorter than the smoothing window—approximately 3 kyr—remains unchanged, since the high-pass filter does not suppress variability below the smoothing window.
Appendix B: Weak Coupling Example
-
In Eq (B1), dηi dt = ai − bi2 ηi(t) + bi2 η2(t) + Ni(t)ξi(t), the ice volume dynamics is an Ornstein-Uhlenbeck process. p. 7
Where the paper says it
One-variable model Following our previous study [35], we treat the dynamics of an individual time series with a one-dimensional non-autonomous Ornstein–Uhlenbeck model. The evolution of the ith variable, denoted as ηi(t), is described by; dηi(t) dt = ai(t)ηi(t) + Ni(t)ξi(t).
-
Using ⟨η2(t)⟩ in Eq (B1) with ηi(0) = 1 we obtain ηi(t) = e −(ai−bi2)t / ai − a2 − bi2 e (ai−a2)t / ai − a2- bi2. p. 8
Where the paper says it
We also employ a similar periodic representation for the noise amplitude coefficient Ni(t). We determine the discrete set of coefficients ai(k) ∀ k ∈ [1, T] using the analytic solution of the stochastic model, yielding ai(k) ≈ − 1 ∆t Si(k) − Ai(k) Si(k) , (8) where Si(k) and Ai(k) denote the approximate periodic variance and autocorrelation of the observed time series Xi , respectively, and are defined as Si(k) ≡ 1 M − 1 M X−1 j=0 X jT +k i X jT +k i ≈ ⟨(ηi(k))2 ⟩, and (9) Ai(k) ≡ 1 M − 1 M X−1 j=0 X jT +k i X jT +k+1 i ≈ ⟨ηi(k)ηi(k + 1)⟩.
-
For the most extreme–and never realized–ice volume coupling constant the stochastic process of the ice volume is mean reverting to the origin. p. 7
Where the paper says it
A Hurst exponent of zero implies an absence of long-range scaling and thus indicates mean-reverting behavior of the time series. This behavior arises because the filtered data no longer exhibit a slowly varying background and instead represent a stochastic process that relaxes toward a constant mean.
Appendix C: Model Coefficients using Gap-filled N2O Data
-
Application of this strategy to the gap-filled N2O data yields model coefficients that are nearly indistinguishable from those obtained using the original EPICA record for N2O. p. 9
Where the paper says it
In this subsection, we present the model coefficients computed using the EPICA record. In Appendix C, we use the gap-filled N2O data referred to in §I B 2 and compute the resulting model coefficients.
-
The relevant coefficients in Eqs (B1) and (B2) for this case in Fig. p. 9
8 are as follows: ai = −8, a2 = −2, bi2 = −1 and b2i = 0. p. 2Where the paper says it
, local time decay versus growth of perturbations—positive (negative) coupling coefficients have a stabilizing (destabilizing) effect on the coupled variables. As an illustrative case, when bij (t) > 0 and bji(t) < 0, it follows that variable j suppresses the growth of variable i, while simultaneously, variable i promotes the growth of variable j.
Where the paper says it
We subsequently evaluate response functions and elucidate their physical implications for the interactions among these paleoclimate variables. The remainder of this paper is organized as follows.
-
This confirms the robustness of our stochastic framework to variations in temporal discretization and data resolution. p. 3
Where the paper says it
Data preparation To apply the stochastic modeling approach, we first interpolate each time series to an evenly spaced temporal grid. We choose a target resolution that closely matches the coarsest dataset—nitrous oxide, comprising 912 data points—and split the full 800 kyr time domain into 34 equal segments, yielding 25 points per segment and an av-
Appendix D: Results
-
The deterministic stability coefficients are comparable across models for each variable, as they pertain to the same fundamental processes. p. 9
Where the paper says it
Figure 4(a) shows the deterministic stability coefficients for the one- and five-dimensional models. It is important to emphasize that these coefficients are comparable across models for each variable, as they pertain to the same fundamental processes.
-
The coupling coefficients associated with ∆T → IV and CO2 → IV are negative, so that temperature and carbon dioxide facilitate the growth of ice volume. p. 9
Where the paper says it
In both models, methane and nitrous oxide exhibit greater stability compared to carbon dioxide, temperature and ice volume (IV), implying that their deterministic drift exerts a stronger influence, driving them toward the long-term mean behavior. Coupling coefficients Figure 4(a) shows the time-dependent coupling coefficients for the five-variable model.
-
The model captures the observed lagged feedbacks of ice-volume on temperature and CO2, consistent with its role as a slow integrator of climate forcing. p. 0
Where the paper says it
Important results are oscillatory dynamics associated with the couplings between ice volume, temperature and CO2, with a strong causal relationship between the latter driving the former two, which is the signature of the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions. Our approach also captures the observed lagged feedbacks of ice-volume on temperature and CO2, consistent ∗ Author to whom correspondence should be addressed: jw@fysik.
Appendix E: Model Interpretation
-
The negative coupling coefficients connecting ∆T and CO2 to ice volume imply that these variables have a destabilizing influence on ice volume. p. 9
Where the paper says it
However, the five-variable model shows slightly more negative values for all five variables, suggesting that inter-variable couplings enhance overall stability. In both models, methane and nitrous oxide exhibit greater stability compared to carbon dioxide, temperature and ice volume (IV), implying that their deterministic drift exerts a stronger influence, driving them toward the long-term mean behavior.
-
Ice sheets influence temperature through many mechanisms, including the ice–albedo feedback, elevationinduced cooling, and large-scale modifications to atmospheric and oceanic circulation. p. 1
Where the paper says it
In addition to those discussed above, several internal feedback mechanisms have been proposed to explain the emergence of the 100-kyr cycle, including enhanced North American ice-sheet growth and the subsequent merging of the ice sheets [10], spatiotemporal evolution of ice-sheet substratum [11], carbon dioxide variability [12], oceanic carbon uptake, carbon rock weathering, soil nitrogen release, and dust feedbacks [13–15]. However, the sequence, relative importance, and interactions of these feedbacks remain uncertain, limiting our understanding of the coupled dynamics among temperature, ice-sheet evolution, and greenhouse-gas forcing.
-
The opposite signs of the forward and reverse IV couplings introduce delayed cryosphere-climate feedbacks and permit damped oscillatory response modes that were not reproduced in the earlier model. p. 0
Where the paper says it
To test our model, we compute response functions for each pair of variables and compare them with empirical data, confirming our predictions regarding intervariable stability and coupling. This study provides a comprehensive overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in shaping Earth’s climate evolution.
Appendix F: Model Fidelity
-
Both the one- and five-variable models reproduce the overall amplitude and periodic structure of the standard deviation across all time series. p. 9
Where the paper says it
It is important to emphasize that these coefficients are comparable across models for each variable, as they pertain to the same fundamental processes. However, the five-variable model shows slightly more negative values for all five variables, suggesting that inter-variable couplings enhance overall stability.
-
The probability distribution functions also show good overall agreement, although as illustrated in Figure 5(b), the observed distributions exhibit slight departures from Gaussianity. p. 8
Where the paper says it
The model exhibits distinct dynamical behavior depending on whether timescales are much shorter than, comparable to, and much longer than the modulation period P = 23. For example, after removing the slowly varying mean behavior, the residual fluctuations exhibit mean-reverting behavior at long timescales, while their probability distributions are approximately Gaussian.
-
The five-variable model shows the most non-Gaussian characteristics, and ice volume is therefore the least accurately captured by our modeling framework. p. 8
Where the paper says it
[35], we construct a five-variable model for the CO2, CH4, N2O, temperature, and ice volume time series, along with their pairwise couplings. We emphasize that this modeling framework is readily extendable to systems with an arbitrary number of variables.
Appendix G: Response Functions
-
The response function R(τ;t) can be expressed in terms of the time-dependent correlation (or persistence) matrix C(τ;t), following Baldovin et al. p. 1
[16].Where the paper says it
Multifractal methods have also been employed to investigate scale-dependent dynamics in paleoclimate records. For example, Shao and Ditlevsen [34] applied a multifractal approach to Antarctic and Greenland icecore data, demonstrating that the Holocene climate is predominantly monofractal, while the glacial climate exhibits multifractal properties, and concluded that the glacial climate has a longer persistence time and stronger nonlinearities.
-
The causal relationship from CO2 to ice volume and temperature is stronger than that from temperature to CO2, highlighting the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions. p. 0
Where the paper says it
We then treat their co-variations as a coupled non-autonomous stochastic dynamical system. Important results are oscillatory dynamics associated with the couplings between ice volume, temperature and CO2, with a strong causal relationship between the latter driving the former two, which is the signature of the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions.
-
The response-function analysis further demonstrates a strong causal relation from CO2 to ice volume and temperature. p. 0
Where the paper says it
We then treat their co-variations as a coupled non-autonomous stochastic dynamical system. Important results are oscillatory dynamics associated with the couplings between ice volume, temperature and CO2, with a strong causal relationship between the latter driving the former two, which is the signature of the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions.
Appendix H: Conclusion
-
Over the past 800 kyr, Earth’s paleoclimate has exhibited periodic yet intrinsically noisy glacial–interglacial cycles, with a characteristic period of approximately 100 kyr. p. 2
Where the paper says it
As shown in Figure 1a, the characteristic 100 kyr periodicity of glacial cycles is clearly evident in the time series. In addition to this dominant periodicity, however, the datasets exhibit complex and intrinsically noisy structure across multiple timescales.
-
Our response-function analysis further demonstrates a strong causal relation from CO2 to ice volume and temperature, thereby highlighting the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions. p. 0
Where the paper says it
We then treat their co-variations as a coupled non-autonomous stochastic dynamical system. Important results are oscillatory dynamics associated with the couplings between ice volume, temperature and CO2, with a strong causal relationship between the latter driving the former two, which is the signature of the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions.
-
Ice volume, in turn, exerts stabilizing influences on both temperature and carbon dioxide through the ice-albedo feedback, the effects of elevation, and ocean–carbon cycle processes. p. 1
Where the paper says it
Global ice volume, in particular, plays a central role in the climate system as both a slow-moving integrator of climate forcing and a strong feedback mechanism via albedo, topography, and ocean–carbon interactions. The Late Pleistocene glacial cycles, therefore, provide a valuable opportunity to explore linkages between components of the climate system, ice-sheet behavior, and Earth system sensitivity to greenhouse-gas forcing, as well as to assess the processes that may have underlain the pacing and amplitude of past glaciations.
-
We observed oscillatory dynamics in the response functions and concluded that this behavior stems from the opposite-sign cross-couplings between IV and ∆T, and IV and CO2. p. 0
Where the paper says it
Important results are oscillatory dynamics associated with the couplings between ice volume, temperature and CO2, with a strong causal relationship between the latter driving the former two, which is the signature of the dominant role of atmospheric CO2 as greenhouse forcing of glacial–interglacial transitions. Our approach also captures the observed lagged feedbacks of ice-volume on temperature and CO2, consistent ∗ Author to whom correspondence should be addressed: jw@fysik.
Appendix I: Acknowledgments
- P.P. and J.S.W. gratefully acknowledge support from the Swedish Research Council under Grant No 638-2013-9243. p. 0
Where the paper says it
In recent years, numerous explanations have been proposed to explain MPT (see Berends et al. [3] for a recent review); however, the investigation of the physical mechanisms underpinning the 100-kyr cycle remains an active area of research.
[Note: The provided text does not contain a section explicitly labeled Appendix I, but the acknowledgments are present near the end.]
Appendix J: Author Declarations
- The authors have no conflicts to disclose.
Appendix K: Data Availability
- The data that support the findings of this study are available from the corresponding author upon reasonable request. p. 0
Where the paper says it
We quantify the behavior of these time series using multifractal time-weighted detrended fluctuation analysis, which differentiates between near-red-noise and white-noise behavior below and above the 100,000-year glacial cycle, respectively, in all records. This study builds upon the work of Keyes et al.
Appendix L: References
- [1] J. D. Hays, J. Imbrie, and N. J. Shackleton, Variations in the Earth’s orbit: Pacemaker of the ice ages, Science 194, 1121 (1976). p. 0
Where the paper says it
This study provides a comprehensive overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in shaping Earth’s climate evolution. The central challenge of understanding the pacing of the of Earth’s ice ages during the last 800,000 years is to discern the relative importance of external forcing associated with orbital eccentricity, obliquity and precession, and internal dynamics associated with the processes of the atmosphere, biosphere, cryosphere and oceans.
[Note: The provided text does not contain a section explicitly labeled Appendix L for references, but the reference list is present at the very end.]
Appendix M: Reference List
The reference list is present at the very end of the paper.
Appendix N: Figure 1 Caption
Figure 1 shows Carbon dioxide, methane, temperature, nitrous oxide, and ice volume time series.
The authors do not explicitly structure the summary in the requested format; they only provided a list of references at the end of the text.
Appendix O: Figure 1
Figure 1 shows Carbon dioxide, methane, temperature, nitrous oxide, and ice volume time series.
Appendix P: Figure 2
Figure 2 shows frequency spectra of (a) original time series and (b) normalized fluctuations time series relative to the slowly varying mean.
Appendix Q: Figure 3
Figure 3 shows a logarithmic plots of the fluctuation functions for (a) original time series and (b) fluctuation time series with the slowly varying behavior removed.
Appendix R: Table I Caption
Table I presents scaling exponents estimated from MFTWDFA fluctuation function slopes in the two colored-noise regimes for (a) original time series and (b) fluctuation series.
Appendix S: Figure 4 Caption
Figure 4 shows time-dependent model coefficients over one P ≈ 23.5 kyr cycle. p. 0 In recent years, numerous explanations have been
proposed to explain MPT (see Berends et al. [3] for a
recent review); however, the investigation of the physical
mechanisms underpinning the 100-kyr cycle remains an
active area of research.Where the paper says it
Appendix T: Figure 5 Caption
Figure 5 shows comparison of (a) periodic standard deviations, (b) probability density functions, and (c) autocorrelation functions between the data and one-variable and five-variable models.
Appendix U: Figure 6 Caption
Figure 6 shows matrix elements of the time periodic response function obtained from the data plotted as a function of time.
Appendix V: Figure 7 Caption
Figure 7 shows matrix elements of the time dependent response function obtained from the model coefficients plotted as a function of time.
Appendix W: Figure 8 Caption
Figure 8 shows same as Fig. 4, but computed using the gap-filled N2O time series and a higher temporal resolution of 60 data points per period for all variables. p. 3 Data preparation
To apply the stochastic modeling approach, we first interpolate each time series to an evenly spaced temporal
grid. We choose a target resolution that closely matches
the coarsest dataset—nitrous oxide, comprising 912 data
points—and split the full 800 kyr time domain into 34
equal segments, yielding 25 points per segment and an av- Data preparation
To apply the stochastic modeling approach, we first interpolate each time series to an evenly spaced temporal
grid. We choose a target resolution that closely matches
the coarsest dataset—nitrous oxide, comprising 912 data
points—and split the full 800 kyr time domain into 34
equal segments, yielding 25 points per segment and an av-Where the paper says it
Where the paper says it
Appendix X: Figure 1
Figure 1 shows Carbon dioxide, methane, temperature, nitrous oxide, and ice volume time series.
Appendix Y: Figure 2
Figure 2 shows frequency spectra of (a) original time series and (b) normalized fluctuations time series relative to the slowly varying mean.
Appendix Z: Figure 3(a)
Figure 3(a) demonstrates that all five variables exhibit similar multiscale stochastic dynamics on timescales both shorter and longer than the modulation period of P = 23.5 kyr.
Appendix AA: Figure 4(a)
Figure 4(a) shows the deterministic stability coefficients for the one- and five-dimensional models.
Appendix BB: Figure 4(b)
Figure 4(b) shows noise amplitude coefficients for one- and five-dimensional models.
Appendix CC: Figure 5(c)
Figure 5(c) presents a comparison of the autocorrelation functions derived from the data and the model simulations.
Appendix DD: Figure 6
Figure 6 shows matrix elements of the time periodic response function obtained from the data plotted as a function of time.
Appendix EE: Figure 7
Figure 7 shows matrix elements of the time dependent response function obtained from the model coefficients plotted as a function of time.
Appendix FF: Figure 8
Figure 8 shows same as Fig. 4, but computed using the gap-filled N2O time series and a higher temporal resolution of 60 data points per period for all variables.
Appendix GG: Figure 1
Figure 1 shows Carbon dioxide, methane, temperature, nitrous oxide, and ice volume time series.
Appendix HH: Figure 2
Figure 2 shows frequency spectra of (a) original time series and (b) normalized fluctuations time series relative to the slowly varying mean.
Appendix II: Section II, Background & Motivation
-
The glacial cycles of the Late Pleistocene (last ∼ 800–900 kyr) are characterized by a periodicity of roughly 100 kyr. p. 0
Where the paper says it
INTRODUCTION A. Background & Motivation The glacial cycles of the Late Pleistocene (last ∼ 800– 900 kyr) are characterized by a periodicity of roughly 100 kyr.
-
The shift from 41-kyr to 100-kyr climatic cycles is widely recognized as the mid-Pleistocene transition (MPT), which occurred without any discernible change in insolation forcing. p. 0
Where the paper says it
In contrast, these fluctuations transitioned to ∼ 100 kyr timescales in subsequent epochs. This shift from 41-kyr to 100-kyr climatic cycles is widely recognized as the mid-Pleistocene transition (MPT), which occurred without any discernible change in insolation forcing.
-
Global ice volume, in particular, plays a central role in the climate system as both a slow-moving integrator of climate forcing and a strong feedback mechanism via albedo, topography, and ocean–carbon interactions. p. 1
Where the paper says it
However, the sequence, relative importance, and interactions of these feedbacks remain uncertain, limiting our understanding of the coupled dynamics among temperature, ice-sheet evolution, and greenhouse-gas forcing. Global ice volume, in particular, plays a central role in the climate system as both a slow-moving integrator of climate forcing and a strong feedback mechanism via albedo, topography, and ocean–carbon interactions.
Appendix III: Section III, Background
-
Climate time series can be effectively represented using simplified stochastic models, especially when a clear separation exists between short- and long-timescale dynamics. p. 7
Where the paper says it
STOCHASTIC MODELS A. Background Climate time series can be effectively represented using simplified stochastic models, especially when a clear separation exists between short- and long-timescale dynamics, and the short-timescale processes can be treated as random variables [57].
-
The Ornstein–Uhlenbeck process is the overdamped limit of the Langevin equation for a Brownian particle in a quadratic potential, yielding a linear mean-reverting stochastic differential equation driven by Gaussian white noise. p. 7
Where the paper says it
The stochastic model employed here builds upon the Ornstein–Uhlenbeck process, extending its applicability to a non-autonomous system characterized by a distinct separation of timescales. The Ornstein–Uhlenbeck process is the overdamped limit of the Langevin equation for a Brownian particle in a quadratic potential, yielding a linear mean-reverting stochastic differential equation driven by Gaussian white noise.
-
To account for the slow varying mean behavior of the glacial–interglacial cycles, we can add a longer-timescale forcing term to this model, which modulates the system’s mean state. p. 7
Where the paper says it
The Ornstein–Uhlenbeck process is the overdamped limit of the Langevin equation for a Brownian particle in a quadratic potential, yielding a linear mean-reverting stochastic differential equation driven by Gaussian white noise. To account for the slowly varying mean behavior of the glacial–interglacial cycles, we can add a longer-timescale forcing term to this model, which modulates the system’s mean state.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved AI can perform stochastic modeling of paleoclimate time series using a non-autonomous Ornstein-Uhlenbeck process, explicitly accounting for both deterministic behavior and stochastic variability as described in Equation 7.
-
Improved AI can characterize the color of noise and crossover times between distinct noise regimes by applying multifractal time-weighted detrended fluctuation analysis (MFTWDFA) to proxy time series, identifying
near-red-noise and white-noise behavior below and above the 100,000-year glacial cycle.
-
Improved AI can construct a five variable model for CO2, CH4, N2O, temperature, and ice volume by incorporating linear coupling terms to capture potential relationships among these variables.
-
Improved AI can determine the causal influence and direction of interactions among climate variables by computing response functions and comparing them with empirical data to confirm predictions regarding intervariable stability and coupling.
-
Improved AI can identify oscillatory dynamics in response functions arising from opposite-sign cross-couplings between ice volume and temperature, as demonstrated by the derived linear system (Equation 20).
Abstract
Understanding the interactions between ice sheets and global climate forcings over geological timescales is essential for projecting their future. Previous studies have highlighted the role of ice dynamics and climate interactions in establishing the 100,000-year glacial cycles, particularly regarding the growth of the North American ice sheet. Researchers have reconstructed consistent time series for ice volume, temperature, and carbon dioxide by applying inverse forward modeling to benthic oxygen isotope records. Here we model the stochastic behavior of paleoclimate time series to evaluate the coupling between climate variables during the Pleistocene glacial cycles. We quantify the behavior of these time series using multifractal time-weighted detrended fluctuation analysis, which differentiates between near-red-noise and white-noise behavior below and above the 100,000-year glacial cycle, respectively, in all records. This study builds upon the work of Keyes et al. [Chaos vol. 33, 093132 (2023)] by incorporating ice volume into a five-variable model that includes carbon dioxide, methane, nitrous oxide, and temperature, along with intervariable coupling terms to capture potential relationships among these variables. Our analysis shows that ice volume, carbon dioxide, and temperature have a stabilizing effect upon each other. To test our model, we compute response functions for each pair of variables and compare these with empirical data, confirming our predictions regarding intervariable stability and coupling. This study provides a comprehensive overview of glacial-interglacial dynamics and highlights the role of cryosphere-climate feedbacks in shaping Earth's climate evolution.
Sources
- Reduced-Order Modeling of Cyclo-Stationary Time Series Using Score-Based Generative Methods
- Score-Based Modeling of Effective Langevin Dynamics
- Physics constraints and response validation in discrete-time reduced-order modeling: from idealized turbulent systems to climate dynamics
Related papers
- Anthropogenic Heat in Urban Climate Systems: Forcing, Sensitivity, and Feedback
- NORi: An ML-Augmented Ocean Boundary Layer Parameterization
- A Checklist to assess the energy and carbon impacts of ML/AI applications in Earth System Modeling
- Skillful Thunderstorm Forecasts Learned Directly from Observations
- Wind vector retrieval from miniaturized wave-enabled sea-surface drifters
- Unmasking Compensating Errors in Numerical Weather Models: A Process-Additive Neural Network Approach