Data

Creating and Interpreting Digital Orthophotographs

Mosaicking the Imagery

Georeferencing/mosaicking of the imagery was performed using PCI OrthoEngine module. The NITF IKONOS imagery was orthorectified using the Rational Functions extracted from the NITF, then further supplemented with ground control collected via survey grade GPS and corrected for terrain displacement using the DEM's where available. When multiple scenes were available for a given area, these were collectively incorporated into the orthomosaic project through bundle adjustment. Each scene was exported as a separate orthorectified file for further image processing. In addition, the best segments of each scene were selected for creation of the final mosaic. Segments of each scene were selected to minimize sun glint, cloud interference, turbidity, etc. in the final mosaic. Where possible, parts of images obscured by sun glint or clouds were replaced with cloud/glint free parts of overlapping images. As a result, most mosaics have few or no clouds or sun glint obscuring bottom features. However, in some cases, clouds, sun glint, or turbid areas could not be replaced with overlapping imagery. In these areas, such obstructions were minimized but could not be eliminated completely.

Ground Control Points (GCPs) for Georeferencing

Fixed ground features visible in the IKONOS imagery were selected for ground control points (GCPs) which were then used to georeference the imagery (i.e., link the image pixels to a real world coordinate system such as Universal Transverse Mercator). The NOAA's National geodetic Survey and National Centers for Coastal Ocean Science occupied multiple location throughout Palau using survey grade GPS The computed root-mean-squares (RMS) of these residuals were only 17.4 cm in latitude, 38.2 cm in longitude, and 129.0 cm in geodetic (ellipsoidal) height for the survey. The GPS vectors were post-processed using the program GPSurvey version 2.35 and NGS' program ADJUST did the final network adjustment. GPS observations were adjusted using the Trimble base station (PBLS CORS ARP Palau) located at PALARIS is Koror, Palau. We obtained points with a wide distribution throughout the imagery whenever possible since this results in the most accurate registration throughout each image. Only ground control points for terrestrial features were collected due to the difficulty of obtaining precise positions for submerged features. For more details see the report IKONOS Imagery Ground Control Point Survey, 2005.

Image to Image Tie-Points

Image to image tie-points (distinct features visible in overlap areas of each frame such as street intersections, piers, coral heads, reef edges, and bridges) were then used to further co-register the imagery, especially for photos taken over open water where ground control points were not available (see Figure). Softcopy photogrammetry software has the ability to automatically find such features common to overlapping imagery but this automated function has mixed results for submerged features.

Georeferencing/mosaicking of the TIFFs was performed using PCI OrthoEngine. The NITF IKONOS imagery was orthorectified using the Rational Functions extracted from the NITF, then further supplemented with ground control and corrected for terrain displacement using the DEM's where available. When multiple scenes were available for a given area, these were collectively incorporated into the orthomosaic project. Each scene was exported as a separate orthorectified file for further image processing. In addition, the best segments of each scene were selected for creation of the final mosaic. Segments of each scene were selected to minimize sun glint, cloud interference, turbidity, etc. in the final mosaic. Where possible, parts of images obscured by sun glint or clouds were replaced with cloud/glint free parts of overlapping images. As a result, most mosaics have few or no clouds or sun glint obscuring bottom features. However, in some cases, clouds, sun glint, or turbid areas could not be replaced with overlapping imagery. In these areas, such obstructions were minimized but could not be eliminated completely.

Digital Elevation Models (DEM)

Pre-existing U.S. Geological Survey (USGS) 10 m2 resolution digital elevation models were used to correct for relief displacement (see Figure). Once a draft orthorectified mosaic was produced, a set of independent ground control points (i.e., check points) were used to measure the quality of each mosaic's rectification and ensure that it required horizontal and vertical spatial accuracy limits. If the spatial accuracy was not acceptable based on this comparison, additional modifications were made to the GCPs, tie-points, etc., until a satisfactory mosaic was created for each island. In general, mosaics were georeferenced such that pixels are positioned within one pixel width of their correct location.

Image Processing

Several intermediate, derived products were produced as the satellite imagery was processed for use in producing the benthic habitat maps. First, the raw satellite images were converted from Digital Numbers (DNs) to normalized reflectance. Normalized reflectance (or at-satellite reflectance) converts DNs into standardized, satellite-independent, comparable values. First developed for Landsat satellite imagery, the algorithm used to perform this conversion was modified for IKONOS image processing. As part of the conversion from DNs to at-satellite reflectance, the following equation is used (Green et al. 2000):

R = pi * L/ (Eo cos(theta0) 1/r²)
L = radiance (from calibration provided by Space Imaging).
r = earth-sun distance in Astronomical Units.
theta0 = the solar zenith angle
Eo = the mean solar exo-atmospheric irradiance in each band. (A convolution of the spectral response and solar radiation from Neckel and Labs (1984) was used to get Eo.)

The acquisition angles (ephemeris data) of the satellite relative to the ground at the time of image acquisition were also used. Calibration coefficients for the satellite, provided by Space Imaging, were used to calculate at-satellite radiance, which was then transformed to reflectance. The normalized reflectance imagery was then transformed into water reflectance (or the signal < 10 cm above the water surface). Water reflectance uses the near-infrared band to remove radiance attributed to atmospheric and surface effects (Stumpf et al. 2003). Water reflectance estimates how the signal (photons) received by the satellite is diminished as it passes through the atmosphere on the way down to the water-atmosphere boundary and on the way back up to the satellite after the signal leaves the water-atmosphere boundary. Water reflectance also estimates how the signal at the satellite is diminished by water vapor, clouds, specular effects at the water surface (wave surface glint), and other signal- absorbing and diffusing materials.

Finalizing the Process

Final mosaics were created in "img" file format (georeferenced image file) with the following projection parameters Universal Transverse Mercator (UTM) Zone 53 North, North American Datum 83.

These files are available on the Shallow-Water Benthic Habitats of the Republic of Palau DVD and at the NOAA’s Biogeography Program Web site. These mosaics were color-balanced in order to provide the most seamless, cloud-free product available for use (see Figure).