My obsessive quest for free satellite data has unearthed what is without doubt the best open source for Landsat TM and Landsat ETM on the planet, the University of Maryland's Global Land Cover Facility (GLCF). This site has added significantly to its collection since I first wrote about it in October, 2002. The size of its collection is simply awesome.

Their dataset is conveniently searched and accessed using the recently upgraded and improved Earth Science Data Interface (ESDI). Other GIS database administrators should take note of what the University has done with this user interface. It probably represents the state of the art. The archive includes thousands of Landsat TM and Landsat ETM files, and many Landsat MSS, MODIS and other images. GLCF is possibly the most important remotely sensed data image archive on the internet and will soon attain international prominence, if it has not already.

Although this site duplicates some of the TM coverage discussed in the earlier EarthSat article spatially, the site offers all Landsat TM bands instead of the three bands that EarthSat offers. In addition, the site offers a large number of the impressive enhanced (ETM) Landsat product which includes the 15m pan chromatic band, effectively doubling the maximum resolution of the images produced.

This makes the production of true color composites with common image processing tools like Paint Shop Pro much easier. Very extensive image processing would be required to accomplish this using the three bands offered by EarthSat. Since one of the three EarthSat sensors is not in the visible spectrum, there is really no set procedure for doing this. Each image requires essentially custom processing in order to achieve the goal.

I searched the database for an image that is starting to become something of a standard for image comparison, namely Cape Cod Massachusetts. I downloaded the row/path 11/31 bands 1,2,3, 4 and 5. I first tried making the composite from the specified 4,3, and 2 RGB bands but for some reason this did not yield very good results, all the green areas of the image being blue. After some trial and error, I used bands 5, 2 and 1 with a lot of image processing to yield the result shown at upper right. Compare this to the 7, 2, 4 EarthSat image of the same area shown in my earlier article.

The GLCF dataset also includes 841 Landsat Enhanced Thematic Mapper images. These have the same seven bands as the TM images with the 15m resolution band 8 added. As a result, 15m true color RGB composite images can be constructed using the techniques similar to those described in earlier articles. The GLCF dataset is undoubtedly the largest single source of free high-resolution Landsat imagery currently available with the possible exception of the Geogratis set described in my earlier article. ESDI also contains a very limited number of Landsat MSS images.

The second image shown to the right is a Landsat ETM composite of the Puget Bay area of the Northwestern United States constructed by combining the 1, 2, and 3 bands in RGB order. Below it is a full scale 30m resolution chip from the large Landsat ETM image shown immediately above it. Note the wake of the ocean-going vessel at the upper left of the image.

The final image is a detail from a 15m RGB composite produced from the number 8 panchromatic band. The HSL technique described in my earlier article did not produce anything close to a true color composite for reasons that have to do with the subtleties of the HSL color model and the characteristics of the particular panchromatic image I used. As a result, I tried a different technique that produced a better result. This consisted of averaging the 30m RGB composite image with the 15m gray scale so that the four distinct images were merged to produce the composite. This can be done conveniently using the Image|Arithmetic utilities in Paintshop Pro. The result of this combination is definitely of higher resolution than the 30m composite (which you can easily see by zooming the 30m image to match its scale to the 15m image below), but slightly degraded from the 15m panchromatic source. The spectral fidelity has suffered as well. Clicking on the thumbnail will link to a full 15m resolution chip.

The production of 15m RGB composites from Landsat ETM bands is worth an article itself since (as I mentioned in my earlier article) there are several techniques described in the literature. Several readers of my articles have already pointed out better ways to construct such composites. I hope to publish more on this subject on this site in the future.

Note: The University has put a lot of work into the GLCF site since I first wrote this article last year. As a result, this outstanding site is even better and more reliable than before. I cannot praise this outstanding scientific resource enough.