Exploration of Feature Engineering Techniques and Unsupervised Machine Learning Clustering Algorithms for Geophysical Data on Levees

Current levee inspection practices consist mainly of visual inspections of the levee surface and sparse, instrumentation monitoring of a single data point at the location of installation. With an assigned grade of “D” from the United States Army Corps of Engineers for levees in the United States, many of these critical systems are in poor health and with significant deterioration putting millions of people at flood risk. There is a need for better levee inspection practices to assess the health of these systems and to create more standardized inspection practices among different organizations as these systems continue to age and deteriorate. By using a combination of UAV-enabled sensing techniques and geophysical techniques, both the surface and subsurface of levees can be spatially and temporally mapped in a non-invasive way, significantly improving current inspection practices. UAV LIDAR, in combination with two geophysical techniques, multichannel analysis of surface waves (MASW) for shear wave velocity and electromagnetic induction (EMI) for apparent resistivity, was used along a stretch of levee on Grand Island in the Sacramento-San Joaquin Delta in California, just south of Sacramento, to assess the feasibility of using these methods to map the health of sandy levees. The subsurface data was used as inputs into unsupervised machine learning algorithms to assess the health and structures of the subsurface. Preliminary analyses have shown that unsupervised machine learning can be used to cluster subsurface data by depth below ground surface.