Refraction-corrected bathymetric point cloud from the UAS survey of the coral reef off Waiakane, Molokai, Hawaii, 24 June 2018

Aerial imagery was collected using a Department of Interior-owned 3DR Solo quadcopter fitted with a Ricoh GR II digital camera featuring a global shutter. The UAS was flown on pre-programmed autonomous flight lines which were oriented roughly shore-normal and were spaced to provide approximately 75 percent overlap between images from adjacent lines, at an approximate altitude of 100 meters above ground level (AGL). The camera was triggered at 1 Hz using a built-in intervalometer. Before each flight, the camera’s digital ISO, aperture, and shutter speed were adjusted for ambient light conditions. A total of five flights were conducted for the survey between 16:40 and 17:45 UTC (06:40 and 07:45 HST). Flight F01 was a reconnaissance flight, and no mapping imagery was collected during the flight. Flights F02 and F03 were conducted at an approximate altitude of 100 meters above ground level (AGL), resulting in complete coverage of the mapping area with a nominal ground-sample-distance (GSD) of approximately 2.5 centimeters per pixel. Flights F04 and F05 were conducted using the same flight lines and altitudes of F02 and F03, but the camera was fitted with a circular polarizing filter to reduced reflections and provide improved imaging of the seafloor through the water surface. Only the images collected with the polarizing filter (F04 and F05) were used for the creation of the point cloud presented in this data release.

Survey control was established using twenty temporary ground control points (GCPs) distributed throughout the survey area. The GCPs were placed using a combination of kayaking, wading, and snorkeling. The GCPs consisted of: nine submerged targets consisting of small (80 centimeter X 80 centimeter) square tarps with black-and-white cross patterns anchored to the shallow (less than 1.5 meters deep) seafloor using 2 pound fishing weights; nine sub-aerial targets consisting of orange plastic five-gallon bucket lids (32 centimeter diameter) painted with a black “X” pattern and affixed in a horizontal orientation to vertical rebar stakes placed in areas of reef rubble to provide the targets with sufficient elevation to remain above the water surface during the survey; and two sub-aerial ground targets consisting of small (80 centimeter X 80 centimeter) square tarps with black-and-white cross patterns placed in the sand at the shoreline. Two of the submerged targets were disturbed by waves or currents during the survey and were not used for SfM processing. All GCP positions were measured using post-processed kinematic (PPK) GPS, using corrections from a GPS base station on a temporary benchmark (MK02) located approximately 1 kilometer away from the study area. Reference coordinates for MK02 were established using the mean position derived from four static GPS occupations with durations greater than 4 hours submitted to the National Geodetic Survey Online Positioning User Service (NGS OPUS).

To measure approximate water surface elevations during the survey, a pressure sensor was temporarily deployed at a central location within the mapping area approximately 450 meters offshore. Water depths were recorded at 1 Hz and were adjusted to compensate for atmospheric pressure using an atmospheric pressure sensor concurrently deployed nearby. The elevation of the pressure sensor port was measured using the same PPK GPS used for the GCPs. The water surface elevation (WSE) was calculated using the sum of the measured ellipsoidal height of the sensor and the mean water depth measured during the duration of the UAS flights. To correct for potential low-frequency water level fluctuations during the survey (such as those caused by tidal fluctuation, or wind and wave setup), separate WSE were calculated for flights F02 and F03 (unfiltered, “non-polarized” imagery) and for F04 and F05 (imagery collected with a circular polarizing filter). Wave action was minimal during both time periods: depths varied by 0.138 meters with a standard deviation of 0.021 meters, and 0.170 meters with a standard deviation of 0.020 meters during the acquisition of the non-polarized and polarized imagery, respectively. Longer-frequency water level changes were also found to be minimal during the time period. Despite the different time periods, the resulting final mean WSE of the time periods for F02 and F03, and F04 and F05 differed by less than 0.001 meters. The final resulting mean WSE for both time periods was calculated as 16.327 meters above the NAD83 GRS80 ellipsoid.

The image files were renamed using a custom python script. The file names were formed using the following pattern Fx-YYYYMMDDThhmmssZ_Ryz.*, where: - Fx = Flight number - YYYYMMDDThhmmssZ = date and time in the ISO 8601 standard, where 'T' separates the date from the time, and 'Z' denotes UTC ('Zulu') time. - Ry = RA or RB to distinguish camera 'RicohA' from 'RicohB' - z = original image name assigned by camera during acquisition - * = file extension (JPG or DNG)
The approximate image acquisition coordinates were added to the image metadata (EXIF) ('geotagged') using the image timestamp and the telemetry logs from the UAS onboard single-frequency 1-Hz autonomous GPS. The geotagging process was done using the GeoSetter software package. To improve timestamp accuracy, the image acquisition times were adjusted to true ('corrected') UTC time by comparing the image timestamps with several images taken of a smartphone app ('Emerald Time') showing accurate time from Network Time Protocol (NTP) servers. For this survey, + 00:00:02 (2 seconds) were added to the image timestamp to synchronize with UTC time. The positions stored in the EXIF are in geographic coordinates referenced to the WGS84(G1150) coordinate reference system (EPSG:7660), with elevation in meters relative to the WGS84 ellipsoid.
Additional information was added to the EXIF using the command-line 'exiftool' software with the following command: exiftool ^ -P ^ -Copyright="Public Domain. Please credit U.S. Geological Survey." ^ -CopyrightNotice="Public Domain. Please credit U.S. Geological Survey." ^ -ImageDescription="Low-altitude aerial image of the shallow coral reef off Waiakane, Molokai, HI, USA, from USGS Unmanned Aircraft System (UAS) survey 2018-617-FA." ^ -Caption-Abstract="Coral reef off Waiakane, Molokai, HI, USA, from USGS survey 2018-617-FA." ^ -Caption="Aerial image of the shallow coral reef off Waiakane, Molokai, HI, USA, from USGS Unmanned Aircraft System (UAS) survey 2018-617-FA." ^ -sep ", " ^ -keywords="coral reef, Molokai, Maui County, Hawaii, Waiakane, 2018-617-FA, Unmanned Aircraft System, UAS, drone, aerial imagery, U.S. Geological Survey, USGS, Pacific Coastal and Marine Science Center" ^ -comment="Low-altitude aerial image from USGS Unmanned Aircraft System (UAS) survey 2018-617-FA." ^ -Credit="U.S. Geological Survey" ^ -Contact="pcmsc_data@usgs.gov" ^ -Artist="U.S. Geological Survey, Pacific Coastal and Marine Science Center"

Structure-from-motion (SfM) processing techniques were used to create the point clouds in the Agisoft Photoscan/Metashape software package using the following workflow with the JPG images from F04 and F05 (images collected with a polarizing filter): 1. Preliminary image alignment and sparse point cloud error reduction was performed using shoreline imagery to develop an a priori camera lens model. The lens model was then fixed for the subsequent SfM processing steps. 2. Additional image alignment for all the images was performed with the following parameters - Accuracy: 'high'; Pair selection: 'reference', 'generic'; Key point limit: 0 (unlimited); Tie point limit: 0 (unlimited). 3. Sparse point cloud error reduction was performed using an iterative gradual selection and camera optimization process in which all sparse points exceeding a Reconstruction Uncertainty of 10 were removed from the sparse point cloud. Additional sparse points obviously above or below the true surface were manually deleted after the last error reduction iteration, and a final camera optimization was performed. 4. Ground control points (GCPs) were manually marked for all GCPs. All submerged GCPs were disabled in subsequent processing steps to serve as validation check points. 5. A final camera optimization was performed, and a dense point cloud was created using the 'high' accuracy setting, with 'aggressive' depth filtering. 6. Some manual editing was performed to delete obvious high and low noise from the point cloud. Classification was run on the point cloud to identify 'low noise' and 'ground' points. 7. The estimate camera positions were exported for use in further processing. 8. The point cloud was exported to an LAZ file for further processing.

Post-processing of the point cloud was performed to reduce the effect of refraction at the air-water interface using the multi-view refraction correction python script (py_sfm_depth.py) described in Dietrich (2017a). The script was used as part of the following workflow: 1. The point cloud was thinned to 5 cm point spacing using the LAStools 'lasthin' utility using the 'lowest' operator, keeping only points that were classified as 'ground'. 2. The thinned point cloud was exported to a csv text file using the LAStools 'las2txt' utility. 3. The water surface elevation of 16.327 meters relative to the GRS80 ellipsoid, was added as a field in the point cloud text file. 4. Multi-view refraction correction was performed on the point cloud text file using the 'py_sfm_depth' script (Dietrich, 2017b), with the estimated camera positions from the SfM processing steps, above. 5. The resulting refraction-corrected point cloud was converted to LAZ format using the LAStools 'txt2las' utility.