2014


Brian Chung

Brian Chung


Years participated in RESESS:
2014
2013


Poster
Download PDF »

An Overview

Major: Mathematics; Geological Science minor
Academic Affiliation: Boston College
Research Mentors: Corne Kreemer
Communication Mentor: Kathy Kelsey


Biography

Inspiration from a gifted, passionate high school earth science teacher was the catalyst for New Yorker Brian Chung to pursue an undergraduate education at Boston College connecting mathematics and geophysics. Additional interests in Slavic and violin studies have led Brian to seek opportunities in applying his abilities in areas of social significance. As a UNAVCO research intern stationed at the Nevada Bureau of Mines and Geology at the University of Nevada Reno, he conducted theoretical research on the applications of nonparametric statistics in short GPS time-series analysis with seasonal signal. Brian and his research mentor, Dr. Corné Kreemer, assessed the performance of the nonparametric Theil-Sen slope estimator with seasonal adjustments as compared to the traditional, parametric, least-squares estimator with seasonal and non-seasonal adjustments.


Abstract

A common element of time-series data in fields such as geophysics, hydrology, and astronomy is seasonal signal. This variation obscures the trends that exist in the time-series. Slope estimates can have low bias as long as the data span is long enough to outweigh the effects of the variation of the seasonal signal. The traditional method of calculating slopes of time-series is least-squares regression. Blewitt and Lavallée [2002] conclude that least-squares slope estimates that account for seasonal signal produce reasonable biases for data spans above 2.5 years long. They add that least squares slope estimates not accounting for seasonal signal produce acceptable biases for data spans beyond 4.5 years. The Theil-Sen method is a nonparametric approach to time-series regression that does not depend on the bold assumptions of parametric methods like least-squares. This method is also robust to outliers, whereas least-squares can be heavily influenced by the presence of outliers.

Geodesists calculate tectonic plate velocities based on the position data that come in the form of slopes in time-series. Technological malfunctions as well as natural, anthropogenic, and economic factors often limit the length of time spans over which one can obtain useful velocity estimates. Van Dam et al. [2001], note the effects of seasonal atmospheric and hydrospheric loading on the lithosphere, which alters GPS position measurements. A challenge at the forefront of geodesy is generating accurate and meaningful plate velocity estimates from short timespan time-series. The seasonal Theil-Sen method is henceforth examined as a slope estimator for GPS time-series shorter than 2.5 years.

Through Monte Carlo experiments on synthetic time-series and application to real data, the seasonal Theil-Sen estimator is found not to be a viable method for short time-series slope analyses. The seasonal and non-seasonal least-squares estimators along with Blewitt and Lavallée’s dataspan requirements continue to be the most effective slope estimation methods time-series. The Theil-Sen estimator, however, shows promise is specific types of time-series with phenomena—like steps.


Presentation