Sublacustrine landslide and tsunami models from Lake Quinault, Washington

A topobathymetric digital terrain model (DTM) was generated by merging several topographic light detection and ranging (LiDAR) datasets and a multibeam bathymetric dataset. Bathymetry was collected in 2023 by the NOAA Navigation Response Team Seattle (NCEI, 2023), and topographic LiDAR data were accessed through the Washington LiDAR portal (Washington Geological Survey, 2012, 2018, 2019, and 2020). If needed, datasets were reprojected into NAD83/UTM Zone 10 N [EPSG:26910]. Topographic LiDAR data were provided in the NAVD88 vertical datum. Bathymetric soundings were collected in a vertical datum based on averaged water level data at the outlet of Lake Quinault from 2016-2021 at USGS streamgage 12039500 (NCEI, 2023). The bathymetric vertical datum was adjusted to NAVD88 by applying a vertical offset of +56.317 meters (m), based on a modeled orthometric height provided in the hydrographic survey report (NCEI, 2023). Hydroflattened "water surfaces" were removed from LiDAR datasets overlapping Lake Quinault, resulting in a ring of no data values between the bathymetric and topographic data, representing shallow water depths where the multibeam survey could not reach. A bilinear interpolation was used to fill in the missing data using the SciPy package in Python. Artifacts from interpolation are apparent in the merged DTM, especially along the broad shallows of the northeastern Lake Quinault shoreline. Merged datasets were regridded to a final grid resolution of 2.74 m, based on the resolution of the coarsest LiDAR dataset used (Washington Geological Survey, 2020).

A portion of the topobathymetric DTM was extracted in the vicinity of McCormick Creek along the northern shore of Lake Quinault in order to run the BingClaw landslide model. The geomorphic expression of slide runout from a previous slope failure are visible in the bathymetric data. A DTM representing a possible pre-slide lake-bed surface was generated by tracing the footprint of the observed runout debris, removing this region from the topobathymetric DTM, and performing a spline interpolation to fill in the gap. A minimum volume for the observed runout debris of 210,000 cubic meters was estimated by differencing the smoothed pre-slide DTM with the modern DTM. A raster of initial slide thickness was created by tracing the upslope region of the observed runout and filling this region with an initial guess of 5 m of sediment that tapers to 0 m thickness along the edges of the traced region. Including the modern delta, the initial modeled slide volume provided in the BingClaw example is 260,000 cubic meters.

A suite of BingClaw models was run with the modeled slide thickness inputs, varying rheologic input parameters until the modeled runout roughly matched the observed perimeter of runout debris. In the example provided in this data release, an initial yield stress of 5,000 Pascals, residual yield stress of 500 Pascals, and remolding coefficient (gamma) of 0.005 was used. The model was run using a spatial resolution of 10 m, and model results were written to output files every 2 seconds over 120 seconds of simulation time. The resulting model outputs of evolving slide thickness were used to force the GeoClaw tsunami model. BingClaw outputs were converted to GeoClaw inputs using a python script. Simple plots of slide thickness were generated using python scripts (setplot.py) provided with the BingClaw model.

GeoClaw models were run using the BingClaw model outputs as initial water level conditions based on the method of Kim and others (2019). For each output timestep in the BingClaw model, the change in water depth due to the change in landslide thickness is used as a change in the GeoClaw water surface elevations, assuming hydrostatic pressure. This coupling scheme does not directly account for longitudinal momentum transfer from the slide to the generated tsunami waves. Two types of hydrodynamic tsunami propagation models were run: a non-dispersive wave model using the nonlinear shallow water equations (swe) and a dispersive wave model using the Serre-Green-Naghdi Boussinesq-type equations (bouss). Both models use the shallow water equations for inundation and in modeled water depths less than 5 m. Both models were run with adaptive mesh refinement (AMR) at refinement level resolutions of 1000, 200, 40, 10, and 5 m. The parameter for sea level in GeoClaw was set to 57 m to represent the lake water surface elevation. Model figures and animations were plotted using the included "tsunami_plots.ipynb" Python Jupyter Notebooks.

Files needed to run BingClaw and GeoClaw models along with example modeled outputs, were compiled into compressed archives for distribution. A supplemental information file "readme.md " describes in greater detail the file structure of the included compressed archived, and limited instructions on usage.