Pan Sharpening Landsat with SPOT^®

One of the main advantages of pan sharpening Landsat band files with the SPOT^® panchromatic band as compared to pan sharpening with the corresponding Landsat band is the significantly higher resolution of SPOT^®. Landsat band files are approximately 30m resolution with a 15m panchromatic band. SPOT^® on the other hand has 20m band files and a 10m panchromatic band. The disadvantage of SPOT^® is that it does not have a visible blue band. As a result, if you want to create pan sharpened natural color RGB images from SPOT^® , you have to first synthesize the natural color bands from the three false color bands using a utility such as the one offered by PANCROMA. This can work reasonably well but even after a fair amount of effort the pan sharpening results may never be perfect. So one is left with the choice of either a true color pan sharpened RGB image at the lower 15m resolution or else a synthetic RGB image at the higher 10m resolution.

A third possibility is to use the SPOT^® panchromatic band to pan sharpen Landsat band files. The latest PANCROMA version has the necessary utilities to do this, and the results can be quite satisfactory, as shown by the example described below. SPOT^® data is relatively widely available and free data can be obtained by anyone for all of Canada through the national GeoBase archive. Free Landsat data is available from several sources for the entire earth. (See the Data web page at this site for a complete list of sources.)

The first step in the pan sharpening process is to locate your Landsat and SPOT^® files. You will need three Landsat band files and a corresponding SPOT^® panchromatic band. This procedure works best if the following conditions are met:

• Use ortho images if available.

• Both images should share the same projection and if possible the same datum.

• Both images should be relatively cloud free.

• It is OK if there is some mountain snow but it is better if there is none. If snow coverage exists, it should be similar in both images.

• The SPOT^® scene should be completely within the Landsat scene.

The first step is to select your SPOT^® and Landsat scenes for pan sharpening. A good source for SPOT data is the Canadian GeoBase archive. As mentioned above, try to select them so that the Landsat scene completely encloses the SPOT^® scene. That is, the SPOT scene should be located completely within the bounds of the Landsat scene. This is not too hard to do because Landsat scenes cover an area approximately 185 km by 185 km while SPOT^® scenes cover a much smaller area of 60 km by 60 km. The images to the right show how the SPOT^® scene fit inside the LANDSAT scene for my example.

Note: If you are downloading data from the GeoBase archive, make sure you download the UTM projection version. The latitude and longitude projection will not be oriented 'North is Up' will be rotated with respect to the Landsat image.

Now select and download the three Landsat multispectral band files, i.e. band1, band2 and band3 (blue, green and red) according to the rules above. (You do not need the Landsat panchromatic band). My Landsat scene had corner coordinates:

ULX is: 250890
ULY is: 6475260
LRX is: 512220
LRY is: 6233430

Now subset the Landsat bands by opening the files as usual in PANCROMA. (Note: see the Tutorial Article on the latest PANCROMA utility for automatically subsetting and registering two sets of three band files.) I subsetted the Landsat image to the SPOT^® set of corner coordinates. (Of course you can subset both scenes to any consistent set of corner coordinates.) In my case I first determined the corner coordinates using the 'Compute Maximum Common Extents' utility. To do so I first opened the panchromatic band an one of the Landsat band files.

I selected 'Band Combination'|'Subset Images'|'Compute Maximum Common Extents'. PANCROMA computed the coordinates of the maximum overlap area between the two images. I then selected 'Band Combination'|'Subset Images'|'Subset Three Bands'. When the Subset Input Box appears check the 'Select by Coordinates' radio button. The upper left and lower right corner coordinates have already been entered into the text boxes by the 'Compute Maximum Common Extents' utility. These are the corner coordinates of the SPOT^® scene. Now select the 'Enter' button. The data entry box will disappear and the subset images will be displayed. Now save the three subset images in GeoTiff format by selecting 'File'|'Save'|'Save Subset Images'|'GeoTiff'.

The next step in the pan sharpening process is to compare your subset Landsat and SPOT^® panchromatic images. They should cover exactly the same area. The SPOT^® rows and columns should also be exactly three times those of the Landsat Subset images. Often the rows or columns will be off by one. This must be corrected by resizing whichever is larger down to the smaller dimension. If the panchromatic band is larger, resize it by selecting 'Pre Process' | 'Resize Images'. Select 'Resize One Image' if the panchromatic band is larger or else 'Resize Three Images' if the Landsat bands are larger.

Once you have checked your images and resized if necessary, the Landsat band files can be pan sharpened using the SPOT^® panchromatic band as usual. In my case I entered the three subsetted Landsat band files (blue green and red) and the resized (by one row) SPOT^® panchromatic band. The images to the right show the result of my efforts to pan sharpen Landsat bands p051r020_7dt20010814_z10_10.tif, p051r020_7dt20010814_z10_20.tif and p051r020_7dt20010814_z10_30.tif downloaded from the GLCF website with SPOT^® panchromatic band s5_12508_5721_20061008_p10_1_utm10.tif downloaded from the Canadian GeoBase archive of SPOT^® imagery. The first grayscale image is the SPOT^® panchromatic band. The second is the Landsat band3 grayscale image that has been subsetted so that its corner coordinates match that of the SPOT^® panchromatic image exactly. The third image is the RGB image pan sharpened using the HSI transform method. (Note that the black collar area of the SPOT^® panchromatic image is preserved, as the black collar pixels are merely pan sharpened to black.)

If desired you can also use the PANCROMA Brovey, XIONG or AJISANE pan sharpening algorithms. For the latter two, PANCROMA will subset a four file bundle of the three visible spectrum bands plus the NIR band. After subsetting, the pan sharpening proceeds as normal except of course that the SPOT panchromatic band is again used¹.

¹Note: The XIONG algorithm is particularly useful when pan sharpening using the Landsat 7 panchromatic band. The reason is that the Landsat 7 panchromatic band spans wavelengths of 0.52 to 0.90 microns. This covers the green, red and NIR bands. This means that a Landsat panchromatic band contains no blue color information. Pan sharpening using this band can lead to spectral distortion. You can easily notice this when pan sharpening scenes with a high blue spectral component, i.e. forested scenes. Desert scenes in contrast that lack a lot of blue are not so much affected. The PANCROMA XIONG and AJISANE algorithms are designed to help correct this.

The SPOT^® panchromatic band spans a spectrum of 0.48 to 0.71 microns. This almost exactly covers the blue, green and red spectral bands. Pan sharpening with the SPOT^® panchromatic band does not produce the kind of spectral distortions common with Landsat 7, and the XIONG algorithm may actually produce poorer results than the four file HSI algorithm. It is generally helpful to check the specifications of the panchromatic band in whatever data set you are using in order to select the most appropriate pan sharpening algorithm.