unsupervised classification. 
3.1 Study Area and Data 
The study area we selected is the county of Jiangling in 
Hubei province, China. Jiangling county, located in 
middle Yangtze River Plain, is a major rice production 
county in Hubei province. The early rice is sowed in the 
third decade of March or the first decade of April and 
transplanted in the third decade of April of the first 
decade of May. The semilate rice is sowed in the first 
decade of May and transplanted in the third decade of 
June. According to the farming practice of the study area, 
Landsat-5 TM CCT, dated 8 June, 1992 when it was 
scene of Path-124, Row-39 
containing the whole county, was acquired from the 
clear and cloudless, 
Chinese Satellite Ground Station. The image processing 
system we used is ERDAS software and ARC/INFO GIS 
software is also supplementally used. Moreover, 1:50 
000 scale topographic maps, recent vegetation type 
maps, soil maps, land cover/use maps and other ancillary 
information were available. 
3.2 Experimental Procedure 
The whole experimental process includes 5 steps: 
Step 1 load Landsat- TM digital data(scene 124-39) 
Sun  workstation(INPUT). 
Points(GCPs), extracted from 1:50 000 topographic map 
to Ground Control 
in Gauss-Kreuger projection system, were used to 
conduct geometrical correction. Radiometric correction 
hasn't been done. 
Step 2  : display the administrative boundary of 
Jiangling county(ARC/INFO vector format) on the 
screen and cut down(CUTTER) the rectangular image 
circumscribing the study area with a rectangle box. The 
administrative boundary of Jiangling county was 
extracted form the same topographic map as did GCPs. 
Plate 1 is a standard false-color composite image of the 
242 
rectangular image containing the whole study area. The 
yellow line is the administrative boundary of Jiangling 
county. 
Step 3 
area outside the administrative boundary. A standard 
for strategy A, mask out the inappropriate 
statistical unsupervised classification was 
performed(CLUSTR) and the target area was clustered 
into 50 clusters. 
Step 4 
was conducted(CLUSTR) directly and the rectangular 
area was clustered into 50 clusters also. Then, cut out the 
for strategy B, unsupervised classification 
inappropriate area with the administrative boundary. 
Step 5 referring to the soil maps, land cover/use 
maps and topographic maps, the unsupervised 
classification result from step 4 was recoded into 10 
major land cover types. 
3.3 Results and Analysis 
Table 1 and table 2 list the unsupervised classification 
result of strategy A and strategy B respectively. Plate 2 
illustrates the condensed 10 clusters from step 5. In plate 
2, red color stands for early rice and pink semilate rice, 
shallow gray irrigated land and blue water. 
In table 2 only 14 clusters occupied more than 2 percent 
of the total non-zero pixels, while in table 1 there are 17 
clusters. This means pixels within the administrative 
boundary from strategy B has a better accumulation 
effect compared with that from strategy A which ismore 
evenly distributed. In table 1 of strategy A, the first 6 
clusters occupied more than 43 percent of the total non- 
zero pixels. It is hard to avoid class confusion when 
most pixels are aggregated inthe first several clusters.In 
table 2 of strategy B, the first 6 clusters only occupied 
23.28 percent of non-zero pixels. The peaks of the 
histogram are secretely distributed at cluster #7, #9#19, 
#22, #25, #32, #47 respectively. This means the main 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996