Spatiotemporal distribution of Oklahoma earthquakes: Exploring relationships using a nearest‐neighbor approach

Abstract

Determining the spatiotemporal characteristics of natural and induced seismic events holds the opportunity to gain new insights into why these events occur. Linking the seismicity characteristics with other geologic, geographic, natural, or anthropogenic factors could help to identify the causes and suggest mitigation strategies that reduce the risk associated with such events. The nearest‐neighbor approach utilized in this work represents a practical first step toward identifying statistically correlated clusters of recorded earthquake events. Detailed study of the Oklahoma earthquake catalog's inherent errors, empirical model parameters, and model assumptions is presented. We found that the cluster analysis results are stable with respect to empirical parameters (e.g., fractal dimension) but were sensitive to epicenter location errors and seismicity rates. Most critically, we show that the patterns in the distribution of earthquake clusters in Oklahoma are primarily defined by spatial relationships between events. This observation is a stark contrast to California (also known for induced seismicity) where a comparable cluster distribution is defined by both spatial and temporal interactions between events. These results highlight the difficulty in understanding the mechanisms and behavior of induced seismicity but provide insights for future work.

Earthquake Declustering Using the Nearest‐Neighbor Approach in Space‐Time‐Magnitude Domain

“…The other mode is located considerably closer in time and space to the origin; it corresponds to clustered events. This bimodality has been documented in multiple regions and on multiple scales (Gentili et al, , ; Gu et al, ; Kossobokov & Nekrasova, ; Moradpour et al, ; Peresan & Gentili, ; Ruhl et al, ; Schoenball et al, ; Trugman et al, ; Vasylkivska & Huerta, ; Zaliapin & Ben‐Zion, , , , ). The bimodality of observed seismicity facilitates cluster detection and declustering.…”

“…The problem of detecting earthquake clusters (which is different from although closely related to declustering) was shown recently to be effectively addressed by a nearest‐neighbor analysis in space‐time‐magnitude domain (Zaliapin et al, ; Zaliapin and Ben‐Zion, , , , , ). This cluster identification methodology is effective in diverse settings, including tectonic seismicity (Gentili et al, , ; Gu et al, ; Kossobokov & Nekrasova, ; Moradpour et al, ; Peresan & Gentili, ; Reverso et al, ; Ruhl et al, ; Trugman et al, ; Zaliapin et al, ; Zaliapin & Ben‐Zion, , , ), induced earthquakes (Goebel et al, ; Martínez‐Garzón et al, ; Schoenball et al, ; Schoenball & Ellsworth, ; Teng & Baker, ; Vasylkivska & Huerta, ; Zaliapin & Ben‐Zion, ), synthetic seismicity in Epidemic Type Aftershock Sequence (ETAS) models (Gu et al, ; Zaliapin et al, ; Zaliapin & Ben‐Zion, ), and laboratory rock fracture experiments (Davidsen et al, ).…”

Narrow Spatial Aftershock Zones for Induced Earthquake Sequences in Oklahoma

“…We obtain consistent results as in Figures and S10 using the relocated catalog, suggesting that the influence of relative location differences on spatial decay is insignificant in Oklahoma (Figures S12 and S13). Lastly, we test whether our spatial decay results are sensitive to the choice of cluster identification technique by utilizing the network‐tree algorithm from Zaliapin et al () since it has often been applied in Oklahoma (Vasylkivska & Huerta, ; Zaliapin & Ben‐Zion, ). We use the network‐tree approach as applied in Oklahoma by Goebel et al () to identify clusters and find rapid spatial density decay of aftershocks that is, qualitatively, in close agreement with the results of Figures and S11 (Figures S14–S16).…”

Probabilistic identification of earthquake clusters using rescaled nearest neighbour distance networks

“…A d f value of 1.6 is used throughout this work as suggested by Zaliapin & Ben-Zion (2013a, 2016a. Zhang & Shearer (2016) tested several variations of this value in Southern California and found that the cluster identification was not greatly affected by alternate choices, though Vasylkivska & Huerta (2017) and Peresan & Gentili (2018) both found that in some areas this could have a significant impact on the binary separation threshold for causally related events.…”

“…Events at the same location can be set artificially to a very small spatial separation as in our algorithm, but this will cause a significant stretching of the clustered η j mode, which will then have an affect on the mixture model fit by increasing the length of the tails of the distribution and/or adding a third mode. Though the locations are the most uncertain component of the η j , Vasylkivska & Huerta (2017) suggest that the rescaled spatial distance is critically important in some areas, particularly those which experience induced seismicity. Care should be taken when dealing with small η j events in the catalogue by identifying if the events can reasonably be removed, or if a small perturbation in location should be applied to reposition all events to account for location uncertainty.…”