High resolution double-difference relocations of earthquakes in and offshore Puerto Rico and Virgin Islands during the deployment of ocean bottom seismometers from mid-2015 to mid-2016

Process Step 1: The original earthquake event data was downloaded online from the International Seismic Centre - ISC. This was saved as both as text and QuakeML files. The information downloaded is the event origin time and location, depth, magnitude, location errors if any, associated P and S-waves of the event at seismic stations. To obtain the event list information, the following search parameters were submitted: Latitude ranges 16.5 to 20.5 degrees north, Longitude ranges -69.5 to -63.0 degrees, all available depths and earthquake local magnitudes, ML, with start date of 2015-05-13 00:00:00 to end date of 2016-06-01 23:59:59. A total of 4007 earthquake events are in this list. This process step and all subsequent process steps were performed by the same person, Claudia Flores, unless otherwise noted.

Process Step 2: Picking Seismic Phases from the known events on the Ocean Bottom Seismometers - OBS. Python version 3.6.7 and ObsPY version 1.1.1 (Beyreuther et al., 2010) are the main tools used to manipulate seismic waveforms and make calculations. Using the known locations of the earthquake events from the PRSN and the IASP91 velocity model, a list of predicted seismic phase arrivals for the OBS are calculated for input into an auto-picker. FBpicker is a module inside the Python software tool PhasePApy (Chen and Holland, 2015) that is used to pick the P-wave arrivals on the OBS. The same code was used for S-wave phases, but problems arose due to the noise from P-wave arrivals conflicting with the predicted S-wave arrival, therefore auto-picking S-wave phases was abandoned. The file EQ_20150513_20160601.csv merges both the known P and S wave arrivals from the ISC, PRSN (see sources) from Step 1, and the auto-picked P-wave arrivals from the OBS stations. An additional piece of information missing from the ISC, PRSN and OBS pick data is the confidence of the seismic wave pick. An assigned picking weight was calculated based on the distance between the seismic station and the earthquake event in degrees. For distances less than 1.0 degrees a pick weight of 1.0 is assigned. For distances between 1.0 to 2.0 degrees, a pick weight is determined by the following function where: pick weight = COS(distance in degrees/0.0238), which results in a range of pick weights between 1 to 0.10. For distances equal or greater than 2.0 degrees a weight of 0.10 is assigned. One decimal degree is approximately 111.19 km.

Process Step 3: Waveform Cross-Correlation calculations on select PRSN stations and OBS stations. The purpose of waveform cross-correlation is to provide a similarity percentage between earthquake events recorded at a seismic station. The more similar two waveforms recorded at a seismic station, the more likely that the two events originated from the same area. Waveform cross-correlations were calculated by matching waveforms at each station for the events with the restriction that event hypocenters compared cannot be more than 50 km from each other, and a match is considered above 60% similarity between waveforms. All waveforms were filtered using a Butterworth filter with corners between 4 - 16 Hz and cut at 20 seconds starting at 5 seconds before making the event comparisons at each station. PRSN stations used in the cross-correlation are selected based on the noise level observed in the seismic waveform after filtering. Results are in the file EQ_CrossCorrelation.csv.

Process Step 4: Earthquake Relocation using Double-Difference Algorithm. The data from the previous processing steps, the earthquake event information and corresponding waveform cross-correlations, were input for the earthquake relocation code HypoDD version 2.1b. See Waldhauser (2001) for details on how to use the HypoDD software and Waldhauser and Ellsworth (2000) for the double-difference algorithm details. The results depend on a seismic velocity model, and one was developed that combined the velocity model from the Mendoza Model (see sources) for depths to 31 kilometers depth and the IASP91 velocity model for depths 32 to 250 kilometers, included in file VelocityModel.csv. All 4007 events were used as input and the conjugate gradients method for solving the equations to handle the large volume of events. Details on methods for solving the double-difference algorithm are discussed both in Waldhauser and Ellsworth (2000) and Waldhauser (2001). Of the original 4007 events only 3719 events were relocated and are in the file EQ_Relocations_HypoDD.csv.

Process Step 5: Review of seismic phase arrivals on a subsection of the full earthquake event list. Due to time constraints in reviewing each event only a total of 1038 events ranging from May 13, 2015 to August 13, 2015. The purpose of reviewing each event was to add additional P and S phase data that were missed by the auto-picking methods. The auto-picking method generated a P to S pick ratio of 14042:8561. After human review, the P to S pick ratio improved to 18218:16905 for the same number of events. Snuffler, a module inside the seismic processing software Pyrocko version 2019.05.03, is used to import waveforms from various seismic stations sorted by distance to visually inspect each earthquake event and associated seismic wave arrivals (Heimann, et al, 2017). The output from the software is reformatted into a file for purposes of distribution, EQ_Reviewed_Phases.csv.