Automatic Decomposition of Geodetic Time Series for Studies of Ground Deformation

Abstract

Geodetic measurements of surface deformation have been used for several decades to study how the Earth's surface responds to a wide range of geophysical processes. Geodetic time series acquired over a finite spatial extent can be used to quantify the time dependence of surface strain for a wide range of spatial and temporal scales. In this thesis, we present a new method for automatically decomposing geodetic time series into temporal components corresponding to different geophysical processes. This method relies on constructing an overcomplete temporal dictionary of reference functions such that any geodetic signal can be described by a linear combination of the functions in the dictionary. By solving a linear least squares problem with sparsity-inducing regularization, we can limit the total number of dictionary elements needed to reconstruct a signal. In Chapter 2, we present the development of this method in the context of transient detection, where we define transient deformation as nonperiodic, nonsecular accumulation of strain in the crust. The sparsity regularization term automatically localizes the dominant timescales and onset times of any transient signals. We apply this method to Global Positioning System (GPS) data for a slow slip event in the Cascadia subduction zone while incorporating a spatial weighting scheme that filters for spatially coherent signals. In Chapter 3, we use a combination of unique space geodetic measurements and seismic observations to study the 2014 collapse of Barðarbunga Caldera in Iceland associated with a major eruption event. The eruption sequence, which involved deflation of a magma chamber underneath the caldera and emplacement of a dike leading to lava flow, resulted in rapid subsidence of the glacier surface overlying the caldera and wide-scale ground deformation encompassing the rift zone associated with the dike emplacement. We present a model of the collapse that suggests that the majority of the observed subsidence occurs aseismically via a deflating sill-like magma chamber. In Chapter 4, we extend upon the transient detection framework presented in Chapter 2 to study complex surface deformation over groundwater basins near Los Angeles, California. We develop a distributed time series analysis framework based on the sparse estimation techniques of Chapter 2 and apply it to an 18-year interferometric synthetic aperture radar (InSAR) time series covering the Los Angeles area. We compare long- and short-term ground deformation signals to hydraulic head data from monitoring wells to understand the mechanical link between pressure variations in subsurface aquifers and observed ground deformation.