The general strategy by which Landsat MSS data is processed to detect and interpret change is here expanded for the interested reader. All the processing described here is with accomplished digital data using a computer, and it must be painstakingly meticulous.
Preprocessing: A series of cosmetic adjustments are made to the earlier data (especially Landsat-1) to rescale all channels to 8 bit (Digital Number values range from 0-255) and to remove a sensor defect called six-line striping. This is difficult to completely remove and it is still present in some images used in this book and on the CD-ROM.
Registration: The 1972 and contemporary scenes are (spatially) registered so that the pixels of both images exactly overlay one another. This is not a simple task and it is one made more difficult by very small changes in the orbit details of the Landsat satellite series over the years. Landsat 1-3 had a coverage repeat cycle of 18 days whereas Landsat 4-5 was placed in a lower orbit and had a repeat cycle of 16 days. This caused a significant change in orbit geometry and resulted in the imaging paths of Landsat 1-3 Landsat 4-5 never exactly coinciding. We have attempted to use the scenes with the greatest overlap wherever possible.
Normalisation: The two scenes have to be normalised to each other so that parts of the landcover that do not change have the same brightness values. Normalisation removes atmospheric influence, as well as variation introduced by differing sensor calibrations of the various Landsat space craft. Most of the initial scenes, 1972-74 were acquired by Landsat 1 while the contemporary scenes were acquired by Landsat 4 or 5. The normalisation technique used was to find invariant dark areas (eg. lakes, rock outcrops) and bright areas (eg. beach sand) common to both scenes; based on the assumption that these areas had not changed, we numerically adjusted the contemporary scene to the 1972 image.
The contemporary Landsat scene was always chosen to have a similar solar elevation as the 1972 scene. The solar elevation determines how brightly illuminated the whole scene is and how much shadowing is present among the landcover. The influence of this shadowing cannot be corrected, so it was matched by choosing scenes with similar solar elevation. Even though the overpass time of the Landsat satellite series has remained constant at approximately 0930 hours AEST, the position (elevation) of the sun in the sky varies over the annual cycle - the sun is highest in the sky in summer and lowest in winter.
Differencing: With the scene pair normalised, differences in the brightness values (DNs) in each of the four MSS wavebands represent changes only in the landcover. All other influences have been removed or at least greatly reduced. The contemporary scene brightness values for each pixel, channel by channel, are subtracted from the 1972 image to produce an image that represents change alone. This image is called the Difference image.
Display: The radiometric MSS Difference image can be displayed as a false colour composite (FCC) using the same techniques as for the 1972 or contemporary MSS scene. The spectral data (colours) are only initially difficult to interpret; with experience landcover change is easily detected and interpreted. The comprehensive set of examples from the Perth scene in the next few pages are good examples which you should study before moving on with the remaining chapters.