The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
ашВЩмиИШимШИИИ 
where d = (1 - S)Y, Y = the total value of the collectivity 
t = H when the number of samples is bigger than 45 
71 
RSAC has used the above formulas to calculate the least 
samples in stratified sampling of early rice of 2007 in China 
using FAM and SCM, and the result is listed in table 3. Based 
on the methods of stratified sampling, the sampling proportion 
is different between the FAM and the SCM. The former is 
0.0377, and the latter is 0.0231. Sampling with FAM needs the 
least samples of 94, and with SCM only 58. Sampling with 
SCM seems better than with FAM at the point of sampling 
proportion. However, the numbers of sampling units distributed 
to six layers change more violently with SCM than FAM, 
which leads the least samples of six layers to the same situation. 
layer 
total units 
the least samples 
by FAM 
by SCM 
by FAM 
by SCM 
1 
927 
1014 
35 
24 
2 
506 
755 
19 
18 
3 
358 
60 
14 
1 
4 
280 
394 
11 
9 
5 
227 
201 
9 
5 
6 
199 
73 
8 
2 
total 
2497 
2497 
94 
58 
Table 3. The result of calculating the least samples 
3.1.6 Sampling using RS and GIS: When the job of stratified 
sampling has been done, the next step is to show the layers in 
GIS as figure 2. Then, some RS images covering the sampling 
units should be ordered according to the least samples of every 
layer, and imaging date should be in early planting days of the 
surveyed crop. The area covered by images ordered should not 
be less than the area of the least samples of every layer. 
Because of the clouds covered in the paddy region of three 
seventh during the period of surveying early rice of 2007 in 
China, RSAC only ordered 36 SPOT images in stratified 
sampling. On the basis of this status, the surveying collectivity 
was adjusted and sampling units was stratified once again. The 
3.2.1 The shape and size of sampling unit of GRS: The 
sampling units of GRS can be called sampling frames. The 
sampling frames are designed as polygons that are located on 
farmland by RSAC (Chen, 1990). The polygons are mainly 
made up of natural borderlines coming from land cover such as 
road, dyke, ribbing, etc. Each polygon area is about 25 hectare. 
The structure is shown below. The two sampling frames used 
by RSAC to survey the area of rice are distributed in 
Guangdong province of China. In the map the codes indicate 
different land cover. The code of 1100 indicates paddy field, the 
code of 2000 indicates fallow, the code of 8001 indicates dykes, 
the code of 7000 indicates roads, the code of 1800 indicates 
vegetables, the code of 3000 indicates garden plots, and the 
code of 1901 indicates other plots of crops. 
new number of the collectivity is 1699, and the summation of 
least samples of six layers is 74 using FAM, it is 56 using SCM. 
However, the total number of sampling units is 201 in the actual 
application. 
3.1.7 Estimating the collectivity: Interpreting RS images 
covering the sampling units is an important step before 
estimating the surveying collectivity (Thomas, 2002). The 
quality of interpretation is closely related to the result of 
surveying (Yang, 2002). The key point of interpretation is to 
discriminate the crop and get the crop area of each sampling 
unit covered by RS images. After the total sampling units 
covered by RS images have been interpreted, the total area of 
surveying crop is to be estimated using the following formula: 
Ÿ- tiffin, 
(8) 
h=l 7=1 
Where y M = the crop acreage of unit ,■ of the h layer 
N = total number of sampling units of the h layer 
y = estimate value of total area of the collectivity 
L = total number of the layers, 
h = 1,2,... ,L 
n h ~ the amount of sampls of the h layer 
RSAC used two continuous years’ RS data to calculate the crop 
area variation rate based on SSM. Using FAM, the variation 
rate of area of early-rice covered the four-sevenths paddy fields 
of China is -2.48% from 2006 to 2007 and the confidence 
interval is from -8.91% to 4.41% while the confidence is 95%. 
Using SCM, the variation rate is -2.30% and the confidence 
interval is from -9.51% to 5.49% while the confidence is 95%. 
3.2 Ground random sampling using GPS 
Ground Random Sampling (GRS) using GPS is an independent 
method adopted by RSAC. On one hand, using GRS can make 
up for the lack of RS such as images covered by clouds. On the 
other hand, ground sampling can provide independent 
information of agricultural condition such as crops area, and 
crops geographical position information, which provides 
reference to the interpretation of RS images. 
Figure 2. The structure of sampling frames 
1341