Digital elevation models (DEMs) of coastal North Carolina, from 2020-05-08 to 2020-05-09

On a workstation (AMD Threadripper 3960X /AsRock TRX40 Creator motherboard/256GB 3000MHz RAM) using Win10x64, aerial imagery and associated positional data (Kranenburg and others 2022a; 2022b; 2022c; 2022d) and a single ground control point (Brown and others, 2021) were brought into an Agisoft Metashape Pro (v. 1.6.5) project to align in a 4D manner (see Sherwood and others, 2019). Only one ground control point, GCP 34, was used as a control point to help 'lock' in the camera positions, whereas the rest of the points were used as independent vertical check points (that is, they were not placed in Metashape). See Kranenburg and others, 2022a, for Metashape Reference Settings inputs. The additional imagery improves the horizontal and vertical accuracy, but similar products can be reproduced without them. The following steps were performed on the Metashape project in the geographic coordinate system NAD83(2011) in ellipsoid height following the general guidance of Over and others (2021): 1. Separate camera models and camera groups were created for each flight date aligned in 4D. Note that multiple imagery datasets are used in the overall processing effort to produce the end individual survey products that this metadata file covers. 2. Imagery (with positions) are aligned in 4D to create a point cloud using a 'High' alignment setting, keypoint limit of 70,000, tiepoint limit of 0, generic preselection selected, and reference preselection via source selected. The tiepoint accuracy was set to 1 pixel. 3. The resultant point cloud was filtered using one iteration of the 'Reconstruction uncertainty' filter at a level of 10, one iteration of the 'Projection accuracy' filter at a level of 3, and two iterations of the 'Reprojection accuracy' filer at a level of 0.3 (in the second iteration the 'fit additional corrections' option was turned on). With each filter, iteration points are selected, deleted, and then the camera model is optimized to refine the focal length, cx, cy, k1, k2, k3, p1, and p2 camera model coefficients. 4. Natural breaks in the final alignment product were identified based on bodies of water or the extreme northern and southern ends of a flight (see Product_Boundary_Map.jpg at https://doi.org/10.5066/P9K3TWY7). The breaks and dates of imagery in the 4D process are used for ease and speed of processing. The boundaries used for the 2020-05 dataset are the Virginia-North Carolina border vicinity to Oregon Inlet, Oregon Inlet to Hatteras Inlet, Hatteras Inlet to Ocracoke Inlet, Ocracoke Inlet to Ophelia Inlet, Ophelia Inlet to Cape Lookout.
For each region, sub-chunks were created for each mission (one or two flight-days, usually adjacent) by iteratively copying the original 4D chunk and removing all camera groups not part of a single mission. Note that this particular metadata file only covers the 2020-05 dataset. Processing took place starting in June 2021 and into February 2022, the latest processing in YYYYMM is given in the Process Date.

From the previous process step, each 'chunk' is processed as follows to create the final DEM products for each flight date: 1. Targeted imagery sets in the aligned chunk are copied from the workstation to a new Agisoft Metashape Pro. project on the U.S Geological Survey Tallgrass high-performance computing network for processing on CPU and GPU nodes (https://www.usgs.gov/core-science-systems/sas/arc/machine-access). 2. If multiple flight dates were flown for this chunk, imagery from one date is targeted, and the rest are disabled (this is then switched for each flight date in the chunk after the final step and steps 3-6 are repeated). 3. A high quality dense point cloud is generated for the chunk with 'mild' filtering and the options to calculate point color and point confidence selected. 4. Dense point cloud is filtered using the 'Filter by Confidence' tool using the range 0-2, these points selected at this range are classified as low noise. 5. A DEM is built from the dense point cloud with interpolation disabled and without the points classified as low noise. 6. DEM is exported in NAD83(2011) UTM Zone 18N and NAVD88(m) with 1-m cell size and region boundaries rounded to the nearest 10 m interval. West and south bounds are rounded up and east and north bounds are rounded down. 7. In order to minimize elevation differences between flights in the 4D reconstruction, all DEMs were then adjusted to a common datum by adding the median signed difference between GCPs (Brown and others, 2021) and the Oregon Inlet to Hatteras Inlet DEM elevations (GCP - DEM) produced by each survey. That is, for DEMs from 2019-10-11 to 2020-05-09, differences were calculated for all GCPs and all Oregon Inlet to Hatteras Inlet DEMs and then these values were ranked, and the middle value was used (or the middle two values were averaged, for an even number of differences) to adjust each DEM in the chunk, not just the Oregon Inlet to Hatteras Inlet DEM. The median value for this 4D reconstruction was -0.03402. 8. Each adjusted DEM was then masked by a human-edited shapefile in QGIS (v. 3.2.2.) to remove high uncertainty regions. Note that the shapefiles are used as part of the workflow to produce the final product and are not provided, if this extent or shapefile needs to be used by someone else it can be recreated in a GIS setting by extracting the outline(s) of the provided raster(s), for example the raster to vector tool in QGIS. Uncertainty was visualized using a hillshade of the DEM with an elevation factor of 5 and an additional raster that highlights areas with a high standard deviation away from the adjusted DEM elevation values. The shapefiles were edited by either S. Bosse, J. Over, E. Lyons, or J. Favela. 9. Final clipped DEMs for the 2020-05 dataset are turned into cloud-optimized GeoTIFFs (COG) using gdal_translate with the following command:for %i in (.\*.tif) do gdal_translate %i .\cog\%~ni_cog.tif -of COG -stats -co BLOCKSIZE=256 -co COMPRESS=DEFLATE -co PREDICTOR=YES -co NUM_THREADS=ALL_CPUS -co BIGTIFF=YES (v. 3.1.4 accessed October 20, 2020 https://gdal.org/), where i is the name of each geoTIFF section.