3.3.2. Compressed data statistics
In an attempt to condense the original data by removing the range of grayscale values, the four spectral bands (masked) were compressed from their original 11-bit format with 2048 possible values each, to create two new data sets containing equidistant bins of 20 and 100 values each. The same four original statistics above (mean, standard deviation, number of unique values, and Shannon index of unique values) were computed for each of the 8 new bands. This procedure produced 32 potential measures of spectral 10-DAB (Appendix A).
3.3.3. Unsupervised classifications
This study employed four different unsupervised classification schemes. The classification scheme used to generate the mask image (described previously) resulted in 100 signatures. Of these, 22 were selected for inclusion in the mask; the remaining 78 non-mask classes comprised the first classification scheme. The other three classification schemes were created from the masked IKONOS image bands to produce 10, 100, and 1000 class signatures. Using the same classification methods as when generating the mask, the classes in each case were initialized along the principal axis and were calculated with an ISODATA algorithm using a convergence threshold of 0.95. For each of the four classified images, the number of unique classes and a Shannon index were computed using pixels within each of the field plots. A total of 8 spectral diversity indices were thus created using this procedure (Appendix A).
In an attempt to condense the original data by removing the range of grayscale values, the four spectral bands (masked) were compressed from their original 11-bit format with 2048 possible values each, to create two new data sets containing equidistant bins of 20 and 100 values each. The same four original statistics above (mean, standard deviation, number of unique values, and Shannon index of unique values) were computed for each of the 8 new bands. This procedure produced 32 potential measures of spectral 10-DAB (Appendix A).
3.3.3. Unsupervised classifications
This study employed four different unsupervised classification schemes. The classification scheme used to generate the mask image (described previously) resulted in 100 signatures. Of these, 22 were selected for inclusion in the mask; the remaining 78 non-mask classes comprised the first classification scheme. The other three classification schemes were created from the masked IKONOS image bands to produce 10, 100, and 1000 class signatures. Using the same classification methods as when generating the mask, the classes in each case were initialized along the principal axis and were calculated with an ISODATA algorithm using a convergence threshold of 0.95. For each of the four classified images, the number of unique classes and a Shannon index were computed using pixels within each of the field plots. A total of 8 spectral diversity indices were thus created using this procedure (Appendix A).