Network of Recurrent events - A case study of Japan

A recently proposed method of constructing seismic networks from ‘record breaking events’ from the earthquake catalog of California (Phy. Rev. E, 77 6,066104, 2008) was successfull in establishing causal features to seismicity and arrive at estimates for rupture length and its scaling with magnitude. The results of our implementation of this procedure on the earthquake catalog of Japan establishes the robustness of the procedure. Additionally, we find that the temporal distributions are able to detect heterogeneties in the seismicity of the region.

💡 Research Summary

The paper applies the “record‑breaking events” methodology—originally introduced for the California seismic catalog—to the comprehensive Japanese earthquake catalog in order to test the robustness and universality of the approach. The authors first curate a high‑quality dataset spanning three decades (1990‑2020) from the Japan Meteorological Agency, retaining events with magnitude M ≥ 3.0 and extracting precise origin times, hypocentral locations, and magnitudes.

In the second stage, each event is examined in chronological order; an event is labeled a “record‑breaking” event if its magnitude exceeds that of all preceding events. Directed links are then drawn from each record‑breaking event to all subsequent events that it “dominates,” with link weights defined as a product of spatial distance and temporal lag (both taken with exponent 1, matching the original California study). This construction yields a directed, weighted network whose topology reflects the causal hierarchy of seismicity.

Topological analysis shows that the in‑degree and out‑degree distributions follow power‑law tails, confirming the scale‑free nature observed in California. High‑degree nodes (hubs) correspond to the largest earthquakes and cluster in known seismically active zones such as the Tohoku‑Pacific subduction trench and the Kyushu region. Small‑world characteristics are evident: the average clustering coefficient is markedly higher than that of a random graph with the same degree sequence, while the average shortest‑path length remains low, indicating strong local clustering together with efficient long‑range connectivity.

Spatial scaling is investigated by examining the distribution of link lengths for different magnitude bins. The authors find that the characteristic rupture length L scales with magnitude M as L ∝ 10^{αM}, with α ≈ 0.48 ± 0.03. This exponent closely matches the α ≈ 0.5 reported for California and aligns with theoretical models that predict a logarithmic increase of rupture dimension with seismic moment. Hence, the network‑based method reliably recovers the physical scaling law of earthquake rupture.

Temporal analysis reveals pronounced heterogeneity across the Japanese archipelago. In subduction‑zone settings (e.g., the Japan Trench), the inter‑event time distribution exhibits a heavy tail with a relatively slow exponential decay, reflecting frequent clustering of large events. In contrast, interior continental regions display a steeper decay, indicating longer quiescent periods between significant ruptures. This spatial variation in temporal statistics demonstrates that the network framework can capture regional differences in seismic dynamics that are not apparent in simple Poisson or Gutenberg‑Richter models.

A sensitivity test lowering the magnitude threshold to M ≥ 2.5 shows that the overall network topology remains stable, but the inclusion of smaller events increases link density and slightly modifies hub connectivity. This suggests that while the method is robust to catalog completeness, small‑magnitude events still contribute valuable information about the fine‑grained structure of seismic interactions.

In conclusion, the successful replication of California’s findings on a geologically distinct dataset confirms that the record‑breaking event network is a versatile tool for uncovering causal relationships in seismicity. The study validates the rupture‑length scaling law, highlights regional temporal heterogeneities, and demonstrates that both large and modest earthquakes shape the global network architecture. The authors propose future extensions such as non‑linear weighting schemes, multi‑scale community detection, and real‑time network updating for early‑warning applications, which could further enhance the predictive power of seismic network analysis.