RGB and thermal imagery, SfM products, and GPS data collected during UAS operations at Marsh Island, New Bedford, MA on July 2nd, 2024

GROUND CONTROL: Ten AeroPoint GCPs were spaced out over the field site and left on for at least two hours to collect GNSS data. After collection and connected to Wifi, the AeroPoint data were uploaded and run through a post-processing kinematic algorithm of the CORS network to get a global accuracy. Internal accuracy is reported in xyz variance and with a baseline distance. The data were exported in the horizontal datum NAD83(2011) to produce latitude, longitude, and ellipsoid heights, and then UTM19N and vertical datum NAVD88 using Geoid 18 was used to produce easting and northing and orthometric heights. These were exported to a CSV file and named 2024004FA_MI_Jul_AeroPoints.csv.

UAS FLIGHTS: The Skydio X10 is a small UAS with an integrated VT300-L camera system. The radiometric thermal camera is a FLIR Boson+ with less than 30 millikelvin sensitivity. The Skydio X10 also has two RGB cameras, one has a 50 degree field of view (narrow) and the other has a 93 degree field of view (wide). A single flight in auto mapping mode was flown at 85 meters with 70% desired sidelap and overlap and 0.67 cm ground sampling distance. The UAS was flown at 5 m/s using the narrow camera and thermal camera. Two sets of thermal images are taken, but only the set with the radiometric data (SX10r) is provided, as the second set is redundant. Note that all images are automatically tagged by the internal GPS in WGS84 + EGM96 but the positions have been converted to NAD83(2011) and NAVD88 geoid 18 in the imagery locations file (2024004FA_MI_Jul_ImageryLocations.csv) and this is accounted for when transforming to NAD83(2011)/UTM19N and NAVD88 Geoid 18 in the products.

RAW IMAGERY: The SkyDio images (thermal images are radiometric JPEGs (RJPEG) and RGB images are regular JPEGs) were processed to add additional information required by the USGS to the EXIF headers using ExifTools (https://exiftool.org/, version: 12.06), and the files were renamed to a unique identifier using Namexif (http://www.digicamsoft.com/softnamexif.html, version 2.1) to avoid any possibility of duplicate names. These steps are described here. 1. ExifTools was used to tag each photo headers following the Imagery Data System EXIF Guidance. Attributes (e.g. Credit, Copyright, UsageTerms, ImageDescription, Artist, etc) were stored in a csv file and written to each image with the command:' exiftool -csv="C:\directory\name\EXIF.csv" C:\directory\name\of\photos *.JPG ' To read out the photo information to a csv when in the directory with the photos the command is: exiftool -csv *.JPG > .\allheaders_out.csv 2. All the images were renamed with Namexif (https://us.digicamsoft.com/softnamexif.html v 2.1 accessed April 2020) to ensure unique filenames and compliance with the USGS Coastal and Marine Hazards and Resources Program's best practices for image naming convention. Images were renamed with the field survey ID prefix; flight number, and ID that distinguishes USGS cameras by make/camera number, the image acquisition date, coordinated universal time (UTC) in ISO8601 format, and a suffix with the original image name. For example, image name '2024004FA_f01SX10r_20240702T165822Z_S#######', 2024004FA is the field activity ID, f01 is the flight number, SX10r is a Skydio X10 radiometric image, 20240702 is the UTC date in the format YYYYMMDD, and a 'T' is used to separate UTC date from UTC time in format HHMMSS followed by a Z. The S######## is the original raw photo name appended to the end of the new filename.

PHOTOGRAMMETRY: The products were created in Agisoft Metashape Pro v. 2.0.1 using the following general steps (see Over and others, 2021 for a more detailed SfM methodology explanation and https://support.skydio.com/hc/en-us/articles/27236378437019-Scan-reconstruction-best-practices-with-Skydio-X10 for processing the SX10 images): 1. A project was created in Metashape and thermal and RGB imagery was imported. 2. Photos were aligned at a low accuracy and then GCPs were automatically detected in the point cloud. GCP positions (2024004FA_MI_Jul_AeroPoints.csv) were added to the project in the reference systems NAD83(2011)/UTM Zone 19N and NAVD88 (geoid 18). The horizontal and vertical accuracies for the GCPs were set to 0.01/0.02 m, respectively, and the camera position accuracy for the images was set to 10 m. The photos were then re-aligned with high accuracy (the pixels were not subsampled) using a keypoint limit of 40,000 and unlimited tie points. 3. The alignment process matched pixels between images to create point clouds and put the imagery into a relative spatial context using the GCPs. The resultant point clouds were filtered using one iteration of the 'Reconstruction uncertainty' filter at a level of 14, one iteration of the 'Projection accuracy' filter at a level of 4, and three iterations of the 'Reprojection accuracy' filter to get to a level of 1. With each filter, iteration points are selected, deleted, and then the camera model was optimized to refine the focal length, cx, cy, k1, k2, k3, p1, and p2 camera model coefficients. 4. At this point, multiple ‘chunks’ were created so that a high-quality dense cloud with a low-frequency filtering algorithm could be made for both the radiometric and RGB images. The dense point cloud was then edited by visual inspection to remove points with a low confidence near the edges and near water bodies. 5. A DSM is built from each dense point cloud and then an RGB averaged orthomosaic is built from the DSM with refined seamlines. The RGB DSM and orthomosaic were exported at 5 cm resolution (2024004FA_MI_Jul_SX10_DSM_5cm.tif and 2024004FA_MI_Jul_SX10_RGBavg_Ortho_5cm.tif). The thermal orthomosaic was converted from Kelvin to Celsius using the band math (0.01*thermal_band)-273.15 and exported as a single band raster at 10 cm resolution keeping only values within the ranges of 5 and 50 to remove outlier values (2024004FA_MI_Jul_SX10_IR_Ortho_10cm.tif).

CLOUD OPTIMIZATION: All products were DEFLATE compressed and turned into a 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.