Landcover change detection using digital analysis of remotely sensed satellite data: A methodological study
Methodology
Two Landsat Thematic Mapper scenes of diffrent dates, Septamber 1988 and April 1989, were acquired for the study site representing we and dry season respctively. Both were cloud-free for the substances selected. The two sets of Landsets TH data were coregistered to a 15-meter UTH projection resulting in a single date set.
A supervised classification was performed on the meridian system for the subscence of each date. Field survey was undertaken to study vegetation cover, terrain and other related information. This procedure allowed identifying the training samples to be selected. The training samples selected were based on landcover and terrain. Signature set components of bands 92,3,4,5 and 7) obtained from training sample were statistically tested to measure discrimination between classes. When satisfactory discrimination was obtained, classification was performed to assign the pixels into the classes having nearest mean vector. piele falling outside data boundary as defined by standard deviation were classified as unknown or full. Classification results give 11 classes for the September 1988 image and 14 classes for the April 1989 image. Thus, the digital number of pixels for the September 1988 classified image ranges from 1 to 11 and 1 to 14 for those of the April 1989 classified image.
The two classified images were digitally combined into one image. this is accomplished using a formula I
1 x (ncls
2) + I
2 , where I
1 and I
2 represent the digital numbers of the September and April images respectively and nc1s2 is the number of classes in the April image. The matrix in figure 1 reveals how the formula produces a different digital number for each possible combination of classes. The result is a classified image created by a process that provides a unique method of referring back to the original images.
April Image classes (1-14)
September
Image Classe
(1-11) |
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
| 1 |
15 |
16 |
17 |
18 |
19 |
20 |
21 |
22 |
23 |
24 |
25 |
26 |
27 |
28 |
| 2 |
29 |
30 |
31 |
32 |
33 |
34 |
35 |
36 |
37 |
38 |
39 |
40 |
41 |
42 |
| 3 |
43 |
44 |
45 |
46 |
47 |
48 |
49 |
50 |
51 |
52 |
53 |
54 |
55 |
56 |
| 4 |
57 |
58 |
59 |
60 |
61 |
62 |
63 |
64 |
65 |
66 |
67 |
68 |
69 |
70 |
| 5 |
71 |
72 |
73 |
74 |
75 |
76 |
77 |
78 |
79 |
80 |
81 |
82 |
83 |
84 |
| 6 |
85 |
86 |
87 |
88 |
89 |
90 |
91 |
92 |
93 |
94 |
95 |
96 |
97 |
98 |
| 7 |
99 |
100 |
101 |
102 |
103 |
104 |
105 |
106 |
107 |
108 |
109 |
110 |
111 |
112 |
| 8 |
113 |
114 |
115 |
116 |
117 |
118 |
119 |
120 |
121 |
122 |
123 |
124 |
125 |
126 |
| 9 |
127 |
128 |
129 |
130 |
131 |
132 |
133 |
134 |
135 |
136 |
137 |
138 |
139 |
140 |
| 10 |
141 |
142 |
143 |
144 |
145 |
146 |
147 |
148 |
149 |
150 |
151 |
152 |
153 |
154 |
| 11 |
155 |
156 |
157 |
158 |
159 |
160 |
161 |
162 |
163 |
164 |
165 |
166 |
167 |
168 |
Figure.1. matrix of the possible range of numbers resulting from the dignitally combined September 1988 and April 1989 images
The large range of digtal numbrs in this image are than regrouped into a smamm number of more meaningful classifications
Results and Discussion
The change detection procedure used has involved a combination of classified images derived for each date. This approach as described above, was performed on an initian supervised classification of Land sat TM data from 2 dates and subsequently mathematically combined the two classified images. Hence the output image was greatly dependent upon the accuracy of the classified images field survey together with the knowledge of relationships among landform, soil and land use etailed the regrouping of the classes with a meaningful legend. The change image obtained from this analysis offered the landuse /landcover types for 2 season with respect to terrain. it should be noted that difficulties in distribution of riparian from other types of vegetation and floodplain from lower terrace found when
taking into account only radiometric values. These problems can be solved by further analysis of the change image. figure 2 shows the false colour composite of the subscene for 2 date images contrasts in wetness and vegetation covers between 2 images could be discernible
Fig.2. False colour composite of the 2 substances.
Table.1 Area occupied by each Class of the change image (September 1988 and April 1989)
| Glase |
landuse land cove |
Terrain |
Area (km) |
| 1 |
-Null |
- |
0.1056 |
| 2 |
Overlap |
- |
- |
| 3 |
Water in April and September |
- |
127043 |
| 4. |
Water in September, no water in April |
- |
16,2781 |
| 5. |
water in April, no water in September |
- |
4,4347 |
| 6 |
Rice cultvation in September
and rice stubble in April |
Lower terrace |
405,3868 |
| 7 |
Moderate to dense ground cover
of vegetation in April and September |
Lower terrace |
8,7387 |
| 8 |
Slight ground cover of vegetation
in April and September |
Lower terrace |
0.045 |
| 9 |
Slight ground cover of vegetation
in April and September |
upper terrace |
51.56 |
| 10 |
Moderate to dense ground cover of
vegetation in April and
September |
Upper terrace |
248.5287 |
| 11 |
Moderate to dense ground cover of
vegetation (tree +annual field crops) in April
and slight vegetation in September |
Upper terrace |
248.5287 |
| 12 |
Moderate to dense ground cover of
vegetation September and slight
vegetation in April |
Upper terrace |
125,695 |
| Total |
|
|
873,63127 |
The most extensive area of the September substances was covered by rice at the vegetative stage together with statement water. The field crops, forest remnants, shrub and woody weeds occupied the remaining area (upper terraces) was recognized. Table 1 summarizes the area occupied by each class of the change image (April and September). It was not the intention of this study to monitor the chgange of specific landuse type. The change of landcover/landuse of the entire study area was the main interest and concern . Figure 3 fillustrates the change image in the study area.
Fig. 3 Yasothon Landcover change
To assess the accuracy of the change it was compared with the existing maps in combination with field surveys. A randow check of the classes in the change image was done for both seasons and satisfactory results were obtained. In this regard, it should be observed that in term of landuse/ lancove change monitoring, the change image offered an advantage in identifying the vegetation density. Indentification of upper and lower tesraces was inferred from the soil soisture condition of the September image. The April suscene image had a relatively narrow range of reflectance in the identifcation of terrain: the upper terrace could not be discriminated from the lower terrace by using the dry season image.
Consideration of the radiometric value could only allow for the recoginition of a very limted type of terrain. The parameters used in the classified image should inlude, not only landcover and soil moisture conditions, out other sources of information as well . However this approach was based on the use of purely digital radiometric values.
