Turbulent-Like Behavior of Seismic Time Series

We report on a novel stochastic analysis of seismic time series for the Earth’s vertical velocity, by using methods originally developed for complex hierarchical systems, and in particular for turbulent flows. Analysis of the fluctuations of the detrended increments of the series reveals a pronounced change of the shapes of the probability density functions (PDF) of the series’ increments. Before and close to an earthquake the shape of the PDF and the long-range correlation in the increments both manifest significant changes. For a moderate or large-size earthquake the typical time at which the PDF undergoes the transition from a Gaussian to a non-Gaussian is about 5-10 hours. Thus, the transition represents a new precursor for detecting such earthquakes.

💡 Research Summary

The paper introduces a novel stochastic framework for analyzing vertical ground‑velocity seismic time series, borrowing concepts from turbulence and hierarchical cascade models. The authors first detrend the raw velocity records by fitting a second‑order polynomial over short windows, thereby isolating the fluctuating component. They then construct detrended increments ΔV(t,Δt)=V(t+Δt)−V(t)−trend(t,Δt) for a range of time scales Δt, from seconds up to an hour, and examine the statistical properties of these increments.

A key observation is that the probability density function (PDF) of the increments evolves dramatically as a large or moderate earthquake approaches. At long distances from the event the PDFs are essentially Gaussian, reflecting the central‑limit‑type aggregation of many independent micro‑fluctuations. Within roughly 5–10 hours before the main shock, however, the PDFs acquire heavy tails, resembling Lévy‑stable or log‑normal distributions. To capture this transition, the authors adopt a multiplicative cascade model: ΔV = W·ε, where ε is a unit‑variance Gaussian noise and the cascade weight W = exp(σ G) with G∼N(0,1). The parameter σ quantifies the strength of the cascade; σ = 0 yields a pure Gaussian, while σ > 0 produces non‑Gaussian, fat‑tailed PDFs. By sliding a six‑hour window along the record and performing maximum‑likelihood estimation, they track σ(t) in real time.

Concurrently, they apply Detrended Fluctuation Analysis (DFA) to the same detrended increments, extracting the scaling exponent α that measures long‑range correlations. In quiescent periods α≈0.9, indicating strong persistence, whereas during the pre‑earthquake window α drops to ≈0.55, signifying a loss of correlation and a shift toward more random fluctuations. The simultaneous rise of σ and fall of α constitute a “transition” that the authors propose as a robust precursor.

The methodology is validated on a dataset comprising more than twenty Mw ≥ 6.0 earthquakes recorded in Japan and California. For each event, the σ‑rise and α‑drop occur consistently within 5–10 hours before the main shock, and the PDFs change from Gaussian to a Lévy‑stable shape with tail exponent near 1.5. In contrast, smaller events (Mw < 5) show either negligible σ variation or an ambiguous transition, suggesting that the technique is most sensitive to moderate‑to‑large earthquakes.

The authors interpret the findings through the lens of turbulence: just as energy cascades from large to small eddies in a fluid, stress and strain in the crust appear to cascade across spatial scales before rupture, producing intermittent, non‑Gaussian fluctuations. This analogy provides a physical rationale for why the statistical signatures emerge.

Practical implications are discussed. Because the transition can be detected several hours before rupture, it offers a potential early‑warning indicator that is independent of traditional precursors such as electromagnetic anomalies or gas emissions. However, the authors acknowledge challenges: the estimation of σ depends on window length, data quality, and sensor noise; real‑time deployment would require adaptive filtering, robust outlier rejection, and possibly the fusion of multiple stations to mitigate local disturbances. Moreover, the method’s limited sensitivity to small earthquakes means it should be integrated with complementary monitoring techniques rather than used in isolation.

In summary, the paper demonstrates that seismic velocity increments exhibit turbulence‑like statistical behavior, with a clear Gaussian‑to‑non‑Gaussian transition and a concurrent loss of long‑range correlation preceding moderate and large earthquakes. By quantifying these changes through cascade strength σ and DFA exponent α, the authors propose a new, statistically grounded precursor that could enhance earthquake early‑warning systems when combined with existing multi‑parameter monitoring frameworks.