Results, Analysis and Discussion - Historical Land Use/Land Cover Classification Using Remote Sensing - page 186

Fig. 6.11 The three wide major LULC-changes that resulted from applying the pre-classification
change detection approach for the period (1975-2007)
Evaluation of changes/accuracy. This showed the accuracy of the produced
change detection results by assessing the occurred errors which resulted from the
adopted approach in change detection process and by using the general principle of
the error matrix, i.e., How effective was the applied approach to detect the changes?
As the change classes are relatively general, it was possible to perform the
accuracy visually using the remotely sensed data itself. Some 250, 200 and 150
testing points were automatically distributed (random-points) for the three resulted
change classes respectively (i.e., no change, natural vegetation to bare areas, and
bare areas to cultivated areas) over the resulted thematic map (Fig. 6.11 ). Then,
the initial remotely sensed image, the final remotely sensed image and the resulted
thematic map from the differencing process were geographically linked with each
other. This was, of course, after the geographic registration (see Chap. 5.2.1 ) .
After that, came the manual step, where each point on the thematic map was
compared to its land use/land cover in 1975 and again in 2007, using the visual
interpretation of LANDSAT-MSS-June data and the LANDSAT-TM-August data.
The point resulted change class (e.g., bare areas to cultivated) was found to be
correct only when its use/s in both comparison dates (1975 and 2007) were
compatible with the defined and resulted description of the change class. For
example, it's the point resulted change class use in 1975 had to be as a bare area
and then as a cultivated area in the year 2007. However, the thematic map that
represented the change classes did not include all the possibilities of LULC-