Preliminary Results and Discussion

Isodata Clustering
Figure 2 shows the results of isodata cluster analysis for the test area 1, showing the NDVI profiles of the 30 clusters with NDVI on a scale from 0 to 255. Such clusters have also been generated for test areas 2 and 3 respectively. Since these clusters are in large numbers, we need to make smaller number of clusters, which are more distinct and recognizable. These were then reduced to 10 cluster classes from each group by selecting clusters based on the distinct dynamic profiles and then some typical feature classes were selected to study their statistical characteristics and relative properties.


Figure 2: Results of isodata clustering for 30 clusters in test area 1
Properties of Typical Classes
Figure 3 shows the NDVI patterns of typical classes chosen for the purpose of drawing some conclusions about the similarity or differences of various feature classes. These typical classes solely based on the temporal patterns of NDVI, are assumed to be forest1, forest2, semi-desert and desert.


Figure 3: NDVI dynamics of typical features derived from the cluster analysis
Table 1 shows the statistical properties of these typical classes, Table 2 shows the correlation (R² value) between the classes and Table 3 shows the mean distance between the classes and Table 4 shows the mean absolute distance between the classes, that is the average of absolute difference of NDVI values between the classes for twelve months.
Table 1: Statistical properties of some typical classes

	Forest1	Forest2	Semi-desert	Desert
Mean	147.75	157.25	128.5	119.58
Max	167	173	147.0	134
Minimum	120	125	115.0	110
Max-Min	47	48	32.0	24
Max/Min	1.39	1.38	1.28	1.22
Variance	190.69	262.50	68.92	32.91
Std.Deviation	13.81	16.20	8.30	5.74

Table 2: Correlation (R² value) between typical classes

	Forest1	Forest2	Semi-desert	Desert
Forest1	1	0.402	0.148	0.04
Forest2		1	0.335	0.634
Semi-desert			1	0.699
Desert				1

Table 3: Mean distance between the clusters

	Forest1	Forest2	Semi-desert	Desert
Forest1	0	9.5	19.25	28.17
Forest2		0	28.75	37.67
Semi-desert			0	8.92
Desert				0

Table 4: Mean absolute distance between the clusters

	Forest1	Forest2	Semi-desert	Desert
Forest1	0	16.83	21.75	28.33
Forest2		0	29.08	37.67
Semi-desert			0	8.92
Desert				0

Indicator for Temporal Signature Similarity
A review of tables 2,3 and 4 clearly shows that for any two classes to be similar or closer, it is necessary that they have higher correlation and smaller distance between them. Therefore, a factor that could incorporate both the distance between the signatures and the correlation of the plots of their monthly pattern should be used for grouping similar clusters.