Coastal bluff point clouds derived from SfM near Elwha River mouth, Washington from 2016-04-18 to 2020-05-08

Description

Methods

Photos of the bluff were taken from a boat or while walking along the beach (Miller, et al., 2022). Where possible, photos were geotagged with information from a stand-alone RTK GNSS, although not all survey dates had GNSS geolocation information. Photos from Miller, et al. (2022) were aligned using the USGS published workflow (Over, et al., 2022) with Agisoft Metashape Professional 1.8.5. All photos from all 30 survey dates were aligned within a single chunk in a 4D approach described by Wernette, et al. (2022) with the default parameters of Over, et al., (2022). A two-step filtering process was applied to the aligned sparse point clouds, with a reconstruction uncertainty (Re) value of 10 and a projection accuracy (Pa) of 3. Dense point clouds were then generated with ultra-high quality and aggressive filtering. Finally, point clouds were filtered by their point confidence to eliminate points with a confidence of 2 or lower.

No ground-truth dataset was available for any of the point clouds, nor were ground-control points (GCPs) available. Wernette, et al. (2022) found that the 4D with differential GNSS geotagged photos has a median accuracy of 0.38 m from an independent LIDAR point cloud approximately 2 years prior to the point cloud survey date. However, given the likelihood of bluff change within the two years between the first SfM point cloud and the LIDAR point cloud, the true accuracy of the SfM point clouds is likely better than 0.38 m. The relative accuracy of all point clouds to each other is significantly greater, although this is challenging to quantify without ground-truth observations. In this way, the point clouds presented here may be used with greater confidence to track relative changes from one survey to another, but caution should be taken when comparing these point clouds against other independent point cloud dataset.