[Paper Review] Ambiguities in Determination Of Self-Affinity in the AE-index Time Series

N. W. Watkins, M. P. Freeman+2 more|arXiv (Cornell University)|Nov 7, 2000

Complex Systems and Time Series Analysis4 references4 citations

TL;DR

This paper demonstrates that the second-order structure function (S2) method—commonly used to infer self-affinity in time series—can produce misleading results when applied to non-self-affine signals like random pulse trains, especially over limited timescales. It shows that such signals can mimic fractal scaling, leading to erroneous conclusions about self-affinity in the AE index, and argues that the Hurst exponent slope is a more reliable discriminator when combined with additional physical context.

ABSTRACT

The interaction between the Earth's magnetic field and the solar wind plasma results in a natural plasma confinement system which stores energy. Dissipation of this energy through Joule heating in the ionosphere can be studied via the Auroral Electrojet (AE) index. The apparent broken power law form of the frequency spectrum of this index has motivated investigation of whether it can be described as fractal coloured noise. One frequently-applied test for self-affinity is to demonstrate linear scaling of the logarithm of the structure function of a time series with the logarithm of the dilation factor $λ$. We point out that, while this is conclusive when applied to signals that are self-affine over many decades in $λ$, such as Brownian motion, the slope deviates from exact linearity and the conclusions become ambiguous when the test is used over shorter ranges of $λ$. We demonstrate that non self-affine time series made up of random pulses can show near-linear scaling over a finite dynamic range such that they could be misinterpreted as being self-affine. In particular we show that pulses with functional forms such as those identified by Weimer within the $AL$ index, from which $AE$ is partly derived, will exhibit nearly linear scaling over ranges similar to those previously shown for $AE$ and $AL$. The value of the slope, related to the Hurst exponent for a self-affine fractal, seems to be a more robust discriminator for fractality, if other information is available.

Motivation & Objective

• To investigate the reliability of the second-order structure function (S2) as a test for self-affinity in the AE index time series.
• To demonstrate that non-self-affine signals—specifically random pulse trains—can produce near-linear scaling in S2 plots over finite dynamic ranges.
• To challenge the assumption that linear scaling in log(S2) vs. log(λ) implies self-affinity, especially when the range of dilation factors λ is limited.
• To argue that the Hurst exponent derived from S2 scaling is a more robust indicator of fractality when supported by additional physical or statistical evidence.
• To highlight the importance of distinguishing between fractal and non-fractal components in complex geophysical time series like the AE index.

Proposed method

• Construct a model of a non-self-affine time series composed of random, differentiable pulses with functional forms similar to those identified by Weimer in the AL index.
• Apply the second-order structure function S2(λ) = ⟨(X(t+λΔt)−X(t))²⟩ to this model and analyze its scaling behavior over various ranges of λ.
• Compare the scaling behavior of the model to that of true self-affine processes (e.g., Brownian motion) to assess similarity in S2 plots.
• Use the relationship between S2 and the autocorrelation function (ACF) to express S2(λ)/S2(1) in terms of ACF(λΔt) and ACF(Δt), enabling analytical comparison.
• Evaluate the slope of log(S2(λ)) vs. log(λ) to determine apparent Hurst exponent values and compare them to expected values for self-affine processes.
• Use the fact that the AE index is derived from AL and AU indices, and that Weimer’s pulse forms are present in AL, to justify the physical relevance of the pulse model.

Experimental results

Research questions

• RQ1 Can non-self-affine time series, such as random pulse trains, produce near-linear scaling in S2 plots over finite ranges of λ, mimicking self-affine behavior?
• RQ2 To what extent is the S2 method reliable for detecting self-affinity in the AE index when the dynamic range of λ is limited?
• RQ3 How do the scaling properties of pulse-based models compare to those of true self-affine processes like Brownian motion?
• RQ4 What role does the Hurst exponent slope play in distinguishing between self-affine and non-self-affine signals when S2 scaling is ambiguous?
• RQ5 Can the presence of non-fractal components (e.g., substorm-related energy release) in the AE index obscure or mimic fractal scaling in S2 analysis?

Key findings

• Non-self-affine time series composed of random, differentiable pulses can exhibit near-linear scaling of log(S2(λ)) vs. log(λ) over finite dynamic ranges, similar to that observed in the AE index.
• The apparent Hurst exponent derived from such scaling can be close to that of self-affine processes, leading to false positives in fractality detection.
• Pulses with functional forms matching those identified by Weimer in the AL index produce S2 scaling behavior that closely mimics the AE index’s reported scaling over similar λ ranges.
• The slope of the S2 scaling plot—related to the Hurst exponent—is a more robust discriminator of fractality than the linearity of the scaling plot alone.
• The AE index may be a hybrid signal, combining fractal-like behavior from solar wind-driven ionospheric currents with non-fractal components from magnetospheric substorm dynamics.
• Longer time series are required to distinguish between fractal and non-fractal components in AE data, as short segments may be well-fit by simple models like p-models of turbulence.