Burstiness and memory of large subduction earthquakes: insights from paleoseismology and analogue modelling

Constraining the timing of large subduction earthquakes remains a fundamental yet unresolved problem in seismic hazard assessment. Although paleoseismic records from many subduction margins suggest predominantly quasi-periodic recurrence of great earthquakes, the large variability observed among different segments and regions raises the question of whether such patterns reflect intrinsic megathrust behavior or, instead, the limitations of the available records. Here we investigate the robustness and interpretability of earthquake recurrence metrics by combining global paleoseismic datasets with scaled seismotectonic models of the subduction megathrust seismic cycle.

We characterize earthquake recurrence using two complementary statistics: burstiness (B), which quantifies the degree of periodicity and clustering of inter-event times, and the memory coefficient (M), which captures temporal correlations between consecutive recurrence intervals. Mapping paleoseismic records from multiple subduction zones onto the M–B plane reveals that most segments exhibit quasi-periodic behavior (B < 0), but span a wide range of memory values, from strongly negative to strongly positive. Notably, this diversity shows no systematic dependence on subduction rate, earthquake rate, or record length, and adjacent segments along the same margin may occupy markedly different regions of the M–B plane.

To assess whether this apparent variability reflects differences in fault dynamics or observational bias, we analyze long, continuous earthquake sequences generated by scaled seismotectonic models. Despite large contrasts in asperity number, size, and along-strike strength heterogeneity, experimental sequences cluster within a relatively narrow domain of the M–B plane. Through controlled subsampling tests, we show that catalog incompleteness, limited along-strike coverage, and short observation windows can substantially shift M and, to a lesser extent, B. 

The analysis of experimental data provides useful constraints on the limits of our ability to infer long-term earthquake recurrence from paleoseismic records, with important implications for probabilistic seismic hazard assessment.