The Role of Fluids in Triggering Earthquakes: Observations From Reservoir Induced Earthquakes

We relocate micro-earthquakes induced by the Açu reservoir in Brazil and observe seismicity migration consistent with pore-pressure diffusion. Reservoir induced seismicity provides a natural laboratory in which to investigate the spatio-temporal evolution and triggering of earthquakes caused by fluid diffusion. Our results can be used to quantify and model pore-pressure diffusion, and to investigate the role of fluids in triggering earthquakes in other tectonic settings. Do Nascimento et al. (2004) recorded and located 267 earthquakes (M ≤ 2.1) beneath the Açu reservoir between 1994-1997. The seismicity increased several months following annual water level peaks, implying that fluid pressure diffusion is the principal triggering mechanism. The small station spacing and very low-attenuation, Precambrian basement rock enabled location of the earthquakes with uncertainties of only a few hundred meters. The earthquakes were located in three clusters, and the time delay to activation of each cluster increased with the depth of the cluster. The location uncertainties were too large to resolve any seismicity migration within a single cluster. We relocate these earthquakes using waveform cross-correlation to obtain groups of similar events, and improve picks. We also apply time-corrections for the stations with poor absolute timing. We begin by relocating 173 earthquakes from the largest cluster using only S-P times because of the poor timing. We obtain more tightly clustered locations with uncertainties on the order of 10-50 m (sub-sample pick accuracy in 200 samples/s data). We observe temporal migration of the earthquakes within the cluster, both along strike, and to increasing depth. We observe a seismicity migration rate between 5 and 15 m/day. The rate is highest during the time of peak seismicity rate, and there is some suggestion that the rate decreases with increasing depth. We then include the best constrained time-corrections, and apply our methods to the entire data set.