Project Methods
Data Conditioning Methods
Data Processing Overview
The image processing scheme was developed by BAE Systems Spectral Solutions such that atmospherically corrected, calibrated data in units of remote sensing reflectance could be produced and then used with the ArcGIS Coral Reef Digitizer Extension software developed by NOS. The main components of the scheme were atmospheric correction and deglinting (Fig.1).
Figure 1. Image processing flowchart for IKONOS and Quickbird data
Atmospheric Correction
Once the imagery was evaluated for overall quality, it was processed for mapping using the program ATCOR2™, an atmospheric correction software plugin for ERDAS IMAGINE PRO V.8.7 ( Leica Geosystems Geospatial Imaging, LLC .) that corrects for aerosols and water vapor and outputs a radiometrically corrected image in reflectance units. Some key steps for using ATCOR2 are shown below.
- The Solar Zenith value in ATCOR2 is the angle of the sun off-nadir. Space Imaging, however, reported the sun elevation (i.e., the angle of the sun from horizon). Thus, we calculated the solar zenith angle following equation 1:
Solar Zenith (degrees) = 90 - Sun Angle Elevation Eq (1)
Where the sun angle elevation was provided by Space Imaging in the metadata files.
- The tilt angle pertains to the angle of the sensor off-nadir. Space Imaging reported the sensor tilt angle as Nominal Collection Elevation in degrees from the horizon. We calculated the angle by Equation 2.
Tilt Angle (degrees) = 90 - Nominal Collection Elevation Eq (2)
Where the nominal collection elevation was provided by Space Imaging's metadata. Unfortunately, the only tilt angles considered in ATCOR2 for this option were 10, 20, and 30 degrees. The calculated tilt angle was rounded to the closest default angle (e.g., if the tilt angle was 17°, the closest default angle was rounded to 20°).
- The direction (N, S, E, W) to select for this option was determined by the relative azimuth between the nominal collection azimuth of the sensor and the solar azimuth, calculated using Equation 3.
Relative Azimuth = | Nominal Collection Azimuth - Solar Azimuth | Eq (3)
A relative azimuth of 0° = S, 30° = E, 120° = N, 150° = W. All other angles were rounded to the nearest defined angle to determine the direction (e.g., if the relative azimuth was 130°, the closest defined angle was 120° so the direction would be assigned as N).
- Aerosol type was selected as “midlat_summer_marit”
- Haze removal was not performed before correction as the function only worked over land and caused problems over water.
- Output reflectance files were in percent, multiplied by a scale factor (normally 10) and saved in integer format. To get remote sensing reflectance (upwelling radiance / downwelling irradiance), the data needed to be divided by (pi*scale factor*100), resulting in units of per-steradian (sr-1). Example spectra are shown in Figure 2.
Figure 2. Example remote sensing reflectance spectra of seafloor and ground cover.
Pan Sharpening
The purpose of pan-sharpening was to spectrally sharpen low spatial resolution image data with high spatial resolution image data. The 4-band color low-resolution (4m) multispectral (MSI) IKONOS imagery was merged with the high-resolution (1m) single-band grayscale panchromatic IKONOS imagery, with nearest neighbor resampling to the high-resolution pixel size. The pan sharpening process was carried out by Geo Eye and Digital Globe for their data respectively.
An issue noted with the pansharpened files was that they may not appear as sharp as the 4m spatial resolution MSI image before processing. This was a result of the slight temporal difference (sometimes up to ½ second) between the MSI image and panchromatic images, an issue known to Space Imaging but not fixed during the contract period. The detail (i.e., resolution) of the output file, however, should look much more refined after this process.
During orthorectification, digital imagery is subjected to algorithms that eliminate each source of spatial distortion. The result is a georeferenced digital mosaic of several imagery scenes with uniform scale throughout the mosaic. After an orthorectified mosaic is created, visual interpreters can accurately and reliably delineate the boundaries of features in the imagery as they appear on the computer monitor using a software interface such as the Habitat Digitizer. Through this process, natural resources managers and researchers are provided with spatially accurate maps of habitats and other features visible in the imagery.
Glint Removal
Images with moderate amounts of glint were corrected using an automated glint-removal algorithm written in Matlab and utilized the differences in the near-infrared band to distinguish glint from water, land, and the seafloor. The idea behind the code is that pixels will have a variable fraction of specular reflection caused by the angle of wavelets in relation to the sun. The fraction is proportional to the amount of signal in the near-infrared (NIR) band, which would benegligible in the ocean (Hochberg et al. 2003).
In order to calculate the amount of glint, pixels in the image were segmented based on thresholds of NIR to find those with the highest signal. Land, vegetation and very shallow water often had a high NIR value also and were masked out of glint calculations using a band ratio threshold of NIR verses blue. The glint pixels were averaged and the minimum value of the remaining “background” pixels (not glint, land, vegetation, shallow water) were subtracted to get a glint spectrum. The amount of glint in each pixel was calculated using the NIR band by first subtracting the “background” pixel NIR value, then dividing by the glint NIR value. Glint was removed from all bands by subtracting the glint spectrum in all bands scaled by the ratio of glint in the NIR.
The deglinting procedure was carried out with atmospherically corrected MSI and pansharpened data, and only with images that had glint pixels that would hinder the visibility of bottom features. Pansharpening the image after deglinting the four-band multispectral image would result in reintroduction of glint, so deglinting was always the last step. A problem with deglinting pansharpened data was that some spectral artifacts were introduced by the pan-sharpening and the algorithm needed hand-tuning to remove these artifacts.
Deglinting Process
In order to calculate the amount of glint, pixels with the highest 5% of NIR signal were segmented into the “glint subset”. Land, vegetation and very shallow water often had a high NIR signal also and were masked out of glint calculations using band ratio thresholds. Pixels with zeros in all bands, created during image mosaicking, were masked as well. See Glint Removal Algorithms Section in Project Completion Report: Mapping of Benthic Habitats for the Main Hawaiian Islands for more details on this topic.
Final Imagery Mosaics
Final mosaics were created in "tif" file format (georeferenced image file) with the following projection parameters Universal Transverse Mercator (UTM) Zone 5 North, North American Datum (NAD) 83 for the island of Hawaii, and UTM Zone 4 North, NAD 83 for the islands of Maui, Kahoolawe, Lanai, Molokai, Oahu, Kauai, Niihau and Kaula. These were created by fusing imagery collected for this project over open water with cloud free LANDSAT terrestrial imagery
These files are available on the Shallow-Water Benthic Habitats of the Main Hawaiian Islands 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).