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!

Poster #113, Tectonic Geodesy

Exploratory visualization and analysis of high resolution GPS data across California

Ariel Espindola-Mercado, Amy L. Ferrick, & Brendan J. Meade
Poster Image: 

Poster Presentation

2020 SCEC Annual Meeting, Poster #113, SCEC Contribution #10644 VIEW PDF
As part of the 2020 SOURCES program our team forayed into the Nevada Geodetic Laboratory’s high frequency 5-minute GPS dataset spanning 2013-2017. This dataset, over 350 GB and 20 million separate files, has remained underutilized and difficult to access due to its large size and dispersed organization. By packaging these files into a more accessible dataset and performing exploratory analyses, we sought to illuminate how this data can be used to learn about crustal motions. Subsets of the dataset from the Western United States were organized and packaged into parquet files by taking advantage of Dask's parallel CPU processing in order to maximize the limited memory capacity of commonly... used computers. We then normalized and interpolated the displacement data onto a regular grid across Southern California to produce exploratory animated visualizations. For exploratory purposes we also investigated the viability of a dynamic mode decomposition (DMD) of the data with the goal of finding low dimensional representation of large-scale deformation features. With the Python scripts we authored, users are able to plot the three-component time series for a given GPS station, interpolate displacements onto a regular grid in a given region, and animate the evolution of the displacements over a given time interval. Our initial application of DMD to the time series indicates that, although DMD is ideal for time-dependent observational data, the complexity of underlying features may indicate that the underlying features cannot be effectively characterized using harmonic and decaying modes alone. Packaging the dataset in parquet files will make the dataset more accessible for sophisticated data analysis within the rapidly proliferating Python language, as opposed to the niche languages that similar analyses are currently being done on. Additional analyses are needed in order to determine the viability of a dynamic mode decomposition for revealing patterns in the data.