Exciting news! We're transitioning to the Statewide California Earthquake Center. Our new website is under construction, but we'll continue using this website for SCEC business in the meantime. We're also archiving the Southern Center site to preserve its rich history. A new and improved platform is coming soon!

Earthquake catalog reconstruction from analog seismograms: Application to the Rangely Experiment microfilms

Kaiwen Wang, William L. Ellsworth, Gregory C. Beroza, Gordon Williams, Miao Zhang, Dustin Schroeder, & Justin L. Rubinstein

Published August 10, 2018, SCEC Contribution #8384, 2018 SCEC Annual Meeting Poster #058

Before the digital era of seismic recordings, decades of seismic data were recorded in analog form and read manually by analysts. Despite the abundance of analog recordings, surprisingly few recent efforts have been made to analyze them. One common format is the 16 mm Develocorder film which recorded up to 20 channels. As the channels are closely spaced, signals commonly overlap with the adjacent channels making it difficult to convert the images into vector time series, especially for earthquakes. In this study, we discard the idea of converting analog data to vector time series, and instead build a set of image-based methods for detection, location and characterization. We validate the automatic processing workflow on one month of Develocorder films from the Rangely earthquake control experiment run by the U. S. Geological Survey from 1969 to 1975 in an oil field in western Colorado. Each day-long microfilm is scanned into ~700 images that total about 10GB in size. To detect events in this large and unlabeled image dataset, we first apply an unsupervised learning algorithm, Principle component analysis (PCA), to represent image features in lower dimensions and then use a Support vector machine (SVM) classifier to separate out earthquake events in the feature space defined by PCA. We next apply a STA/LTA picker which runs on the scaled image and its horizontal gradient to generate P-wave picks. The picks are fed into a grid-search associator to form individual earthquakes that are input to a location program. We also apply 2-D image correlation around the pick time to measure differential arrival times for double-difference relocation and estimate similarity between event pairs for clustering purpose. By running the workflow described above, we build a catalog of ~200 events based on the Develocorder films. Among them ~40 local events cluster at the injection wells, which is consistent with the original analysis of Raleigh et al. (1976). To locate these events, we also use 2-D image correlation to read the time code, correct for the film misalignment, and recover station locations from the original map. A distinct advantage over other approaches is that we do not need to convert the analog data to time series, but can work directly with the image itself. Our processing techniques have the potential for wide application to the decades of historical data before the digital era in seismology.

Wang, K., Ellsworth, W. L., Beroza, G. C., Williams, G., Zhang, M., Schroeder, D., & Rubinstein, J. L. (2018, 08). Earthquake catalog reconstruction from analog seismograms: Application to the Rangely Experiment microfilms. Poster Presentation at 2018 SCEC Annual Meeting.

Related Projects & Working Groups