Attribute_Accuracy_Report:
No accuracy calculations are performed on raw data. Camera EXIF position data are generally accurate to within at least a few hundred meters, and time data are accurate to within a few minutes. Camera clock was UTC through 7 March 2018, and switched to PST starting 29 March 2018 (to simplify processing across UTC date changes). Some images are over-exposed, some images are under-exposed, and no effort is made to optimize exposure through post-processing, nor to eliminate imagery with acquisition errors. These data represent raw imagery as-collected and subsequently used in Structure-from-Motion (SfM) photogrammetric processing workflows.
GNSS position accuracy is limited by Precise Point Positioning (PPP) techniques and is assumed to be about 10cm horizontal and 15cm vertical based on post-processing reports, multiple comparisons with ground control points, and residual error after alignment for photogrammetry. Accuracy reports from each PPP post-processing report are included with GNSS data when published for a survey.
RGB values are 8-bit, generally not saturated or underexposed. In a few cases of high or low light conditions, some imagery may be under- or over-exposed.
GNSS values are checked for consistency and outliers. Bad data are discarded and missing data are interpolated or replaced with calculated positions from 4D SfM (see Over and others 2021 for methods).
Data are considered complete for the information presented, as described in the abstract. Users are advised to read the metadata record carefully for additional details, and to examine EXIF data for individual flights and images if required to determine camera settings such as exposure mode and shutter speed.
Horizontal_Positional_Accuracy:
Horizontal_Positional_Accuracy_Report:
A formal accuracy assessment of the horizontal positional information in the dataset has not been conducted. Most images have EXIF stamps from a consumer-grade GPS, updated every few seconds. Position accuracy for these images is generally within about 200m. Some images may lack GPS position in EXIF data, but in most cases we apply interpolated positions or positions derived through onboard survey-grade GNSS or 4D structure-from-motion. High-precision positions are produced as a separate product for most flights during post-processing. Interpolated positions are a linear interpolation of XYZ coordinates between images with known coordinates.
Vertical_Positional_Accuracy:
Vertical_Positional_Accuracy_Report:
A formal accuracy assessment of the vertical positional information in the dataset has not been conducted. Most images have EXIF stamps from a consumer-grade GPS, updated every few seconds. Vertical accuracy for these images is not assessed for accuracy. Some images may lack GPS position in EXIF data but in most cases we apply interpolated positions or positions derived through onboard survey-grade GNSS or 4D structure-from-motion. High-precision positions are produced as a separate product for most flights during post-processing. Interpolated positions are simply a linear interpolation of XYZ coordinates between images with known coordinates.
Process_Step:
Process_Description:
Acquisition -- Each image is acquired in camera raw format. Camera is triggered by an intervalometer, usually integrated with a survey-grade GNSS and triggered using a pulse-per-second (PPS) signal from the GNSS, routed through a microcontroller to the camera remote shutter, and recording an event mark to the GNSS when the shutter is actuated, through the flash-sync port on the camera. A consumer-grade GPS is typically attached to the camera keeping camera time synced with UTC time and providing approximate coordinates for EXIF data.
Process_Date: Not complete
Source_Produced_Citation_Abbreviation: raw imagery
Source_Produced_Citation_Abbreviation: raw GNSS data
Process_Step:
Process_Description:
Image renaming -- Raw Images are renamed using Phil Harvey's EXIFTOOL (
https://exiftool.org) with the following command:
exiftool -r -d %Y%m%d%H%M%S "-testname<CAM${SerialNumber;s/\d{4}//}_${SubSecDateTimeOriginal}_${SubSecTimeOriginal}%-c.%e" -ext nef .\*
files follow the following naming convention: {CAM###}_{YYYYMMDDHHMMSS_ss}.jpg, where {CAM###} is the last 3 digits of the camera serial number, preceded by the letters “cam”, and where {YYYYMMDDHHMMSS_ss} is the image acquisition time in {YearMonthDayHourMinuteSecond_hundredths} expressed in 24-hour time, as recorded by the camera’s internal clock and written to the SubSecondDateTime field in the image EXIF data. For example, CAM001_202009182311_50 would be the timestamp for an image with a SubSecondDateTime EXIF time/date stamp of September 18th, 2020 at 11:11.50 pm (see accuracy report for details on time accuracy).
Process_Step:
Process_Description:
Image conversion -- renamed images are batch- converted from .camera raw format to a format compatible with photogrammetry processing software. Software versions, conversion date, and other parameters are recorded in individual image EXIF data.
Process_Date: Not complete
Process_Step:
Process_Description:
Image QA/QC -- original and converted images are counted, and the beginning and end of the survey are reviewed to cull extraneous images such as camera testing or inadvertently triggered images and place both raw and converted images in an "outtakes" folder for a given flight (these images might not be released and are not processed). A multipage contact sheet of images is created to allow quick review ensuring that (a) conversion was successful with no corrupted imagery, and (b) "extra" imagery not capturing the intended view is culled and placed in the "outtakes" folder. If corrupted converted images are found, imagery is re-converted. If extra images are found, raw and converted images are moved to the "outtakes" folder. This step is iterated until a collection of valid images remains. Note that in some cases, images of water may be retained if there are visible or potentially visible bathymetric features, aquatic vegetation, or objects in the scene that may be useful for future data applications.
Process_Date: Not complete
Process_Step:
Process_Description:
GNSS data processing -- when collected, GNSS data are post-processed to derive the precise position of ground control points, check points, and/or the GNSS antenna at the time of image acquisition. If necessary for rapid processing, approximate (PROVISIONAL) positions are produced from rapid ephemeris data. After precise ephemerides are available, GNSS data are post-processed to calculate GNSS trajectory, and precise positions are derived from station occupations or event marks recorded with precision timing, placed along the GNSS trajectory. Processing workflows, accuracy estimates, software and version are detailed in GNSS metadata. Camera and ground control positions are post-processed together with images, camera lever arm data and photogrammetric software to derive accurate camera position and pose.
Process_Date: Not complete
Process_Step:
Process_Description:
Imagery EXIF coordinates reconciliation -- Note that EXIF camera positions are not intended to be suitable accuracy for precise photogrammetric surveys, but they are used to populate the EXIF metadata field and locate the image in space relative to other images for reference purposes. GNSS/SfM position data published separately from images should be used for SfM reconstruction if available. Camera positions are not always recorded for individual images due to GNSS collection conditions, malfunctioning equipment, or other reasons.
Depending on availability of position data, coordinates for image EXIF position data are recorded in the following order of preference, with the appropriate indicator written to the EXIF:GPSAreaInformation tag:
1) If the image position is recorded to the EXIF field with the attached consumer-grade GPS device, that position is used. (EXIF:GPSAreaInformation="camera-integrated GPS")
2) If a survey is missing only a few image positions, positions are interpolated or extrapolated from consumer-grade GPS positions using adjacent images and a time-averaged velocity and direction algorithm. (EXIF:GPSAreaInformation="interpolated" or (EXIF:GPSAreaInformation="extrapolated")
3) If consumer GPS data are missing but GNSS event marks are present, GNSS data are used to fill in the EXIF field. (EXIF:GPSAreaInformation="GNSS event mark")
4) If a survey is missing many or all image positions, and/or another flight with valid position data exists for the same area, 4D SfM positions are used to fill in the EXIF field. These positions may be processed at lower accuracy/resolution, or with fewer images, than final positions used in photogrammetry, and are intended only to approximately locate images in space. (EXIF:GPSAreaInformation="structure-from-motion")
5) If a survey does not have image positions, but synthetic (derived from sources such as lidar or other imagery and elevation data sources) or manually placed ground control points have been measured, the imagery may be processed with 3D or 4D SfM techniques using ground control points to derive EXIF camera positions. These positions may be processed at lower accuracy/resolution, or with fewer images, than final positions used in photogrammetry, and are intended only to approximately locate images in space. (EXIF:GPSAreaInformation="structure-from-motion")
6) If no other imagery exists or has not been processed for a survey area, or if imagery needs to be published without accurate SfM processing to support emergency response or timely best science, then an approximate location may be assigned to some or all images based on best estimate of location. This may include assigning a single location to all images to approximately locate them based on the geographic area of a survey. (EXIF:GPSAreaInformation="estimated approximate location")
Process_Date: Not complete
Process_Step:
Process_Description:
Photogrammetric Processing -- 3D or 4D structure-from-motion (SfM) processing, as described in Over and others (2021) and Warrick and others (2017), is used to derive camera positions for surveys missing position data, and to produce camera models, camera orientation information, point clouds, surface models, depth maps, and orthocorrected imagery. SfM products are released with metadata detailing processing steps and workflows.
Process_Date: Not complete
Process_Step:
Process_Description:
Machine Learning -- Machine learning techniques as described in Buscombe and others (2022) and Buscombe and Goldstein (2022) are used to build, train and evaluate models, and generate labeled data. Labeled and segmented data products are released with metadata detailing process steps and workflow.
Process_Date: Not complete
