I do the odd bit of consulting work from time-to-time when clients would rather pay for image processing work instead of working their way up the learning curve themselves. I recently had a small project from a media company that needed imagery around ancient Petra in Jordan. They wanted both low resolution and high resolution images. I suggested Landsat 8 as an option for 15 meter imagery by pan sharpening multispectral bands 2, 3 and 4 with the panchromatic band. The client certainly liked the price (Landsat 8 scenes are free). The technique, described in earlier articles worked out pretty well, as the Landsat 8 sensors are excellent. A sample from the pan sharpened scene LC81740392013185LGN00 is shown below. The first image is a scaled-down version of the full scene.
Landsat 8 Pan Sharpened Scene
The next image is a detail from the area around the modern town of Petra and the ancient city in the valley just to the west.
Landsat 8 Pan Sharpened Petra Detail
I wondered if I could offer the client an image with 50% improved ground resolution by pan sharpening the Landsat 8 multispectral bands with an EO-1 ALI panchromatic band. I have described the EO-1 ALI instrument in previous articles. EO-1 was a test bed for Landsat 8, testing many of the technologies such as the push broom sensors that were eventually incorporated into Landsat 8. As I discovered during this exercise, the Landsat 8 sensors seem significantly better than the EO-1 sensors. However the EO-1 ALI has a 10m panchromatic band as compared to the 15m Landsat 8 panchromatic band. (I decided not to start another of my rants on that subject as I have made my frustration clear in several earlier articles).
You might wonder why not just pan sharpen the EO-1 multispectral bands with the EO-1 panchromatic bands. The reason is that I have not had a lot of luck doing this. My opinion is that the EO-1 multispectral sensor is just not very good. Most of my attempts at pan sharpening the EO-1 multispectral bands have produced washed-out images that are not visually pleasing. The pan-sharpened EO-1 image is shown below.
EO1 Multispectral Bands Pan Sharpened with EO1 Panchromatic Band
Since EO-1 collects imagery on demand, coverage is sparse, but it tends to be clustered over areas that people tend to be interested in. I logged into the USGS GLOVIS site and discovered that ancient Petra, south of the Dead Sea in Jordan is one such area. I decided to see how pan sharpening by this technique would work.
Since I had already downloaded a Landsat 8 scene covering my area of interest, I proceeded to download the EO-1 scene EO1A1740392010221110KF. Fortunately, the EO-1 scene fit entirely within the extents of my Landsat 8 scene, allowing me to create an image covering the full extents of the EO-1 image. I did this by opening an EO-1 band in PANCROMA and noting the corner coordinates. These happened to be:
ULX : 710385
ULY : 3441315
LRX : 781815
LRY : 3323385
I closed the EO-1 scene by selecting 'Close Graphics Window and Reset' and then opened my three Landsat 8 multispectral bands 2, 3 and 4 (visible blue, green and red, respectively). I then subsetted the Landsat bands to match the extents of the EO-1 band by selecting 'Subset Utilities' | 'Subset Images' | 'Subset Three Bands'. This option allows me to subset three bands to the same corner coordinates in one operation. (Since I was producing images for graphics application, I subsetted to 8-bit subset bands.
When the Subset Data Entry Form became visible, I checked the 'Select by Coordinates' radio button and cut and pasted the corner coordinates listed above into the appropriate text boxes. I clicked 'Enter' and my images were automatically subsetted to the correct coordinates, saving them by selecting 'File' | 'Save Subset Images' | 'GeoTiff'.
The subsetted Landsat 8 bands were off by one column pixel so I fixed that using the PANCROMA image resizing utility. I selected 'Preprocess' | 'Resize Images' | 'Resize Three Images' | 'Scale'. This utility can be used to match row and column counts for the very common "off-by-one" problem. This action resulted in the EO-1 panchromatic band having exactly 3X the row and column counts as the subsetted multispectral bands.
I then pan sharpened the bands using the PANCROMA four file HSI algorithm. (The five band XIONG and AJISANE algorithms are not needed because the EO-1 panchromatic band, like that of the Landsat 8 panchromatic band (and unlike the Landsat 7 panchromatic band), is sensitive to the full visible spectrum and is not sensitive to the infrared spectrum. I experimented with both the Laplacian and Bilinear interpolation methods. Both seemed to work OK. A sample of my pan sharpened scene is shown below. First, a scaled version of the full scene.
Landsat 8 Pan Sharpened with EO1 Full Scene
The image below shows the Landsat 8 bands pan sharpened with the EO-1 panchromatic band. The pan sharpened Landsat 8 scene is shown next to it again for reference.
Landsat 8 Pan Sharpened with EO1 (10m resolution) compared to Landsat 8 pan sharpened scene (15m resolution)
My conclusion from this exercise is that it is certainly possible to pan sharpen the 30m Landsat 8 multispectral bands with the 10m EO-1 ALI panchromatic band. The improvement is not as striking as I expected, despite the higher resolution of the EO-1 panchromatic sensor. This is mainly due to the outstanding quality of the Landsat 8 sensors. However it does not take too much effort to use EO-1 data if it is available and the improvement may be worth the effort.