Multivariate statistical appraisal of regional susceptibility to induced seismicity: application to the Permian Basin, SW United States

Induced earthquake sequences are typically interpreted through causal triggering mechanisms. However, studies of causality rarely consider large regions and why some regions experiencing similar anthropogenic activities remain largely aseismic. Therefore, it can be difficult to forecast seismic hazard at a regional scale. In contrast, multivariate statistical methods allow us to find the combinations of factors that correlate best with seismicity, which can help form the basis of hypotheses that can be subsequently tested with physical models. Such a statistical approach is particularly important for large regions with newly-emergent seismicity comprising multiple distinct clusters and multi-faceted industrial operations. Recent induced seismicity in the Permian Basin provides an excellent test-bed for multivariate statistical analyses because the main causal industrial and geological factors driving earthquakes in the region remain highly debated. Here, we use logistic regression to retrospectively predict the spatial variation of seismicity across the western Permian Basin. We reproduce the broad distribution of seismicity using a combination of both industrial and geological factors. Our model shows that hydraulic fracturing and/or hydrocarbon production from the Wolfcamp Shale is the strongest predictor of seismicity, although the physical triggering process is unclear due to uncertain earthquake depths. We also find that the proximity to neotectonic faults west of the Delaware Basin is another important factor that contributes to induced seismicity. This higher tectonic stressing, together with a poor correlation between seismicity and large-volume deep salt-water disposal wells indicates a very different mechanism of induced seismicity compared to that in Oklahoma