LiDAR-Based Change Detection for Earthquake Surface Ruptures
Determination of near-fault ground deformation during and after an earthquake is key to understanding the physics of earthquakes. Observations with high spatial resolution and accuracy of natural faults that have experienced earthquakes are required to understand the mechanism of earthquake surface ruptures, which are presently poorly understood. Significant damage due to earthquakes, coupled with poor comprehension of the patterns of uplifting, subsidence, and lateral slipping of surface ruptures, requires the ability to rapidly characterize three-dimensional (3-D) deformation over large areas to study stress change on faults and after-slip activity. As Earth observation data becomes ubiquitous, remote sensing based change detection is becoming more and more important for scientific applications, environmental policy and decision-making. Light Detection and Ranging (LiDAR) is a proven approach to creating fast and accurate terrain models for change detection applications. An Airborne LiDAR Scanning (ALS) system, which combines a scanning laser with both GPS and inertial navigation technology, can create a precise, three-dimensional set of points for features on the Earth’s surface. Temporal differencing of repeat ALS surveys is a potential method for determining near field surface deformation caused by earthquakes. Early studies have shown the potential of obtaining displacement fields by differencing repeat LiDAR scans. However, the overall methodology has not received sufficient attention and optimal methods of estimating 3-D displacement from ALS are needed. The primary contribution to knowledge presented in this dissertation is an experimental evaluation of the performance of state-of-art 3-D registration methods for the determination of near-fault earthquake deformation, including closest point criterion based and probability based 3-D registration. Rigid and non-rigid algorithms are both taken into consideration to determine deformation. The convergence behavior, computational time and matching accuracy of every method is assessed in terms of stability and robustness to transformation, noise level, and sampling rate. Moreover, the best performing method is closely examined to determine the optimal scheme and parameters used to autonomously estimate spatially varying earthquake introduced 3-D displacements. A new solution, the Anisotropic Iterative Closest Point (A-ICP) algorithm, has been developed to overcome the difficulties associated with sparse legacy pre-event datasets. The method incorporates point accuracy estimates to both speed up the Iterative Closest Point (ICP) method and improve the matching accuracy. In addition, a new partition scheme, known as a “moving window,” is proposed to handle the large spatial scale point cloud coverage over fault zones and to enhance the change detection results for local, varying surface deformation near the fault. Finally, to make full use of the multi-return and classified point clouds, the enhancement of anthropogenic features in change detection is examined in near-fault displacement research.