Pan Sharpening Tests
NASA released the first Landsat 8 scene on March 30, 2013. (Note that the satellite is technically called the LCDM until it is released to the USGS.) Because the satellite is not yet on its correct orbit path, there are slight geometric differences between this scene and that expected later this year. The scene appears to include the Denver, CO international airport. This article features the results from some initial tests on the data. The first image is an RGB composite made using bands 2, 3 and 4 in the blue, green and red channels. This image is shown below.
Landsat 8 30m RGB Color Composite Image (Credit: NASA).
I ran some tests to determine how the narrowing of the panchromatic band would affect pan sharpening results as compared to Landsat 7 data. The Landsat 7 panchromatic band spans wavelengths 0.52 microns to 0.90 microns. The Landsat 8 panchromatic band spans 0.50 microns to 0.68 microns. Landsat 8 includes slightly more of the visible blue band, but more importantly it excludes nearly the entire near infrared (NIR) band. The Landsat 7 panchromatic band has been problematic for pan sharpening because of its poor match with the visible wavelengths. This has resulted in color distortions that can be quite severe, especially for vegetated areas that tend to be highly reflective of IR radiation. Correcting these distortions takes special algorithms that either subtract the infrared component using the NIR band or that can infer the correct color levels using other techniques.
The first pan sharpened image shown below was produced using a simple four band HSI method. This turned out to produce a very good match with the RGB composite standard.
Landsat 8 15m HSI Pan Sharpened Image (Credit: NASA).
The next image was produced using the PANCROMA™ ELIN algorithm. ELIN uses a complex local optimization strategy that can produce very close color matches to the RGB composite standard regardless of the characteristics of the panchromatic image. It is ideal for Landsat 7 data that has a panchromatic band highly sensitive to NIR wavelengths that can cause color fidelity issues. ELIN also produced a high fidelity image, but not really better than that produced by the HSI algorithm.
Landsat 8 15m ELIN Pan Sharpened Image (Credit: USGS).
The XIONG algorithm subtracts a user-specified fraction of the NIR band from the panchromatic band before pan sharpening. The result at a relatively low subtraction level was interesting. Subtracting the NIR band made the image worse, giving an un-natural cast to the cloud shadows. The color balance of the lake at the upper left of the image is not quite right either. This is the expected result when subtracting the NIR information from a panchromatic band that is not sensitive to the NIR band.
Landsat 8 15m XIONG Pan Sharpened Image (Credit: USGS).
The AJISANE algorithm transforms the image data to a different color space than the algoriths used for the previous image before substituting the panchromatic information and transforming back. As is usual for this method, the pan sharpened images are a little dark and need the color channels boosted equally to produce an optimal image. This color tones from this method appear OK but somewhat muddier than the other images. The interesting thing about this image was the slightly higher resolution of the pan sharpened image as compared to all the previous ones. More fine details in the terminal building and on the runways are visible from this image.
Landsat 8 15m AJISANE Pan Sharpened Image (Credit: USGS).
This is only a single scene, and it was obviously acquired in the winter or late spring in the western part of the United States. As a result, there is little vegetation in the image that would have presented a more interesting test. However, it appears as if the narrower Landsat 8 panchromatic band may yield pan sharpened images with more natural color tones, using simpler pan sharpening methods.
Target Acquisition Tests
The next test used the eight Landsat 8 multispectral bands for target identification. The chosen target was the cloud shadows evident over and in the vicinity of the airport. I first sampled the areas in the image obscured by cloud shadows using the PANCROMA™ Point Spectrum Generator™. I then identified the best match using the PANCROMA™ Spectral Angle Analyzer™. Note that the analysis was conducted using scaled digital numbers (DNs) rather than reflectance values because the necessary equations to transform the DNs to top of atmosphere reflectance have not yet been released. However, the method proved reasonably effective at discriminating the cloud shadows from the other features in the image. The Point Spectrum and the Spectral Angle plot are shown below.
Cloud Shadow Point Spectrum .
Spectral Angle Plot (Credit: USGS).
This is a very high-level preview of Landsat 8 data. However it may provide some indications of the characteristics of the data that should be forthcoming in large volume around the middle of 2013.