An Earthquake Detection Algorithm for Local Seismic Monitoring of Underground Caverns: The Sorrento Salt Dome, Louisiana Case Study

Abstract:

Southern Louisiana is dotted with salt domes. These features were formed as a result of the geologic alteration of the Louann salt, a widespread early Jurassic salt layer (initially ~1.5 km thick) that was deposited in a marine setting during the separation of North and South America. Salt mining has been ongoing in Louisiana for industrial brine extraction, as well as hydrocarbon storage. Tremors were felt around the Napoleonville salt dome, located near Bayou Corne, southern Louisiana, in May 2012, which led to the installation of USGS seismic stations in July 2012. An increasing number of long period seismic events were observed in the days leading up to the formation of an ~200-m-wide sinkhole on 3 August 2012. In late 2019, tremors were felt around the Sorrento Salt dome. We installed 12 USGS SmartSolo nodes recording at 500 Hz above the Sorrento dome in early February 2020 and recorded two months of data, with monitoring resuming in July 2020.

We expect that cavern integrity and the potential of collapse can be inferred based on an increase in the frequency of seismic events observed in the days leading up to cavern failure. Using data from the Sorrento array, we investigate the occurrence of significant seismic events that could be indicative of changes occurring within the salt dome. The application of short-term-moving-average to long-term-moving-average ratio (STA/LTA) detectors to seismogram ground motions has been used as a standard approach to detect local earthquakes, and we initially identified events using an STA/LTA detector with manual adjustments as needed. We then used a correlation detector with the identified events as templates to supplement the initial detections. We defined a detection as a correlation with the template greater than 0.6. The identified detections were then reviewed and will serve as a training dataset for a semiautomatic deep convolutional neural network (CNN) based technique, which will assist in detecting additional microearthquakes that may be indicative of deformation within the salt dome. The input data to the CNN model consists of multi-station, 3-component, short period waveforms recorded in February and March 2020. We will present preliminary results of microearthquake detections, spectrograms and the analysis of spatio-temporal patterns in the seismicity.