Inputs and outputs of 3D point-cloud comparisons (M3C2) from repeat bathymetric and topographic surveys from 2019 – 2022 of the French Broad River, Asheville, NC

This data release provides the input and output datasets used to inform geomorphic change detection (GCD) analyses from repeat streambank and streambed surveys of a one-kilometer reach of the French Broad River (FBR) where the NCDOT initiated a highway widening and bridge construction project in 2020 near Asheville, North Carolina (Whaling, 2023a–d; Whaling and Wagner, 2021; Whaling and others, 2023; Whaling and Fitzgibbon, 2024; Whaling and Little, 2024). The Multiscale Model-to-Model Comparison (M3C2) algorithm (Lague and others, 2013) implemented in CloudCompare was used to quantify changes in streambed and streambank morphology associated with construction activities. Because streambanks were surveyed annually in winter and bathymetry biannually in both winter and summer, most winter streambank surveys were paired with the contemporaneous winter bathymetry to make “annual” assessments, which capture approximately yearly changes in both the streambanks and streambed. The only exception was November 2019, when no winter bathymetry was collected; in this case, the streambank survey was paired with the nearest bathymetry from June 2019 for comparison. In contrast, shorter-term “inter-survey” assessments compared summer bathymetry with the following winter bathymetry to capture streambed change over shorter intervals. In total, six M3C2 cloud-to-cloud comparisons were performed: two “annual” streambank-and-streambed analyses (June 2019 bathymetry + November 2019 streambanks vs. December 2020 bathymetry + December 2020 streambanks; December 2020 bathymetry + December 2020 streambanks vs. December 2021 bathymetry + December 2021 streambanks), three “inter-survey” streambed-only analyses (December 2020 vs. June 2021; June 2021 vs. December 2021; December 2021 vs. June 2022), and one “total-change” analysis comparing the earliest pre-construction surveys (June 2019 bathymetry + November 2019 streambanks) to the most recent surveys during ongoing construction (June 2022 bathymetry + December 2021 streambanks).
To generate the input point clouds for M3C2 (provided in input_M3C2.zip), streambed and streambank datasets were first resampled to a uniform 0.25 meter spacing to support algorithmically consistent volume-change computations. Streambed datasets were resampled using Inverse Distance Weighting (IDW) interpolation within the spatial extents defined by polygons in clipping_polygons.zip, followed by point cloud extraction at grid nodes. Densely sampled streambank datasets were resampled directly using mean rasterization on grids oriented to preserve vertical bank geometry. Grid orientation was defined based on major changes in channel trajectory (e.g., a large bend in the river), necessitating the use of three separate grids. A polygon shapefile provided with this data release (lidar_segments.zip) was used to clip the lidar datasets into these segments prior to resampling to each grid. Resampled streambed datasets were used directly as input to M3C2 for the three “inter-survey” comparisons. For the “annual” comparisons that analyzed both streambed and streambank change, each resampled streambed dataset was merged with its corresponding resampled streambank dataset to form a single input for M3C2: June 2019 bathymetry + November 2019 streambanks vs. December 2020 bathymetry + December 2020 streambanks; December 2020 bathymetry + December 2020 streambanks vs. December 2021 bathymetry + December 2021 streambanks; June 2019 bathymetry and November 2019 streambanks.
The M3C2 algorithm incorporates 3D uncertainty via precision maps (PM), allowing for a statistically robust level of detection (LoD) at 95% confidence (James and others, 2017). Uncertainty was attributed to each point based on data source characteristics. Systematic error assumptions included primarily horizontal lidar uncertainty and vertical bathymetric uncertainty. Lidar error was characterized by horizontal survey and inter-survey misalignment components, while bathymetric error was estimated from cross-check transects and DEM interpolation differences. The CSV file provided with this data release summarizes these 3D point uncertainties used in LoD computation (point_uncertainty.csv).
Each of the six M3C2-PM comparisons performed in CloudCompare produced six output point clouds (provided in output_M3C2.zip) attributed with M3C2 distances, associated distance errors (95% confidence), and significance flags. These results are used in the companion SIR to quantify volumetric geomorphic change and to spatially map erosion and deposition over the study period, as summarized in the accompanying xlsx file (volume_change.xlsx) and illustrated in cover.png, respectively. The supporting R code, provided in supporting_code.zip, performs input dataset resampling and uncertainty attribution, enables survey-area-wide volume computations, and supports targeted GCD analyses that isolate areas affected by direct construction, thereby excluding construction impacts from the geomorphic response in support of the companion SIR.