Seismic Analysis of the 2017 Oroville Dam Spillway Erosion Crisis

The outflow channel of the northern California (USA) Oroville Dam suffered catastrophic erosion damage in February and March, 2017. High discharges released through the spillway (up to 3,000 m3/s) caused rapid spillway erosion, forming a deep chasm. A repeat LiDAR survey obtained from the California Department of Water Resources indicates that the chasm eroded to a depth of 48 meters. A three-component broadband seismometer (STS-1) operated by the Berkeley Digital Seismological Network recorded microseismic energy produced by the flowing water, providing a natural laboratory to test methods for seismically monitoring sudden catastrophic floods and erosion. In this study, we evaluate the three-component waveforms recorded during five constant-discharge periods - before, during, and after the spillway crisis - each of which had a different channel geometry. We apply frequency-dependent polarization analysis (FDPA; following Park, 1987), which characterizes particle motion at each frequency. The method is based on principal component analysis on a spectral covariance matrix in one-hour windows and it produces the horizontal azimuth, vertical tilt, horizontal phase, and vertical phase of the dominant particle motion. The results indicate a greater vertical component (perhaps roughness-induced) of power at a broad range of frequencies at a given discharge after the formation of the chasm. As the outflow crater developed, the back-azimuth of the primary source of seismic energy changed from the nearby Thermalito Diversion Pool (188 degrees) to the center of the outflow channel (170 degrees). To further analyze FDPA results, we apply the 2D spectral-element solver package SPECFEM2D (Tromp et al. 2008), and find that local topography should be considered when interpreting the surface waveforms predicted by FDPA results. This research suggests that monitoring changing channel geometry and erosion in large-scale flood events may be enhanced by seismic FDPA analysis. The results of this work are compared and contrasted with 3-component seismic observations of cobble-bed stream floods in Maryland.