! 2004 
— 
1 tools 
solar 
| crop 
10ving 
cation 
level, 
wheat 
Strata 
'S data 
strata. 
1 tools 
Itivate 
se and 
1 grid, 
els (» 
. After 
, 
lon 
rs and 
given 
andard 
scaled 
id to à 
ecause 
motely 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
Couniry Boundary 
Provincal Boundary Et 
Sampling Frame 
Stratum 
   
E Arable area 
Figure 2. Sampling frame used for the estimation of the 
proportion of all crop 
At last, we selected sample clusters randomly. And sample size 
is calculated using the following equation. 
n, 
HE ST 
jl 
N 1) 
(2 p(1- p) 
muU LW) 
d 
Where: n is sample size, nO is sample size of simple random 
sampling, N is population, d is predefined sampling error, t and 
d are obtained from the assumption of normal-distribution. Here 
given sampling accuracy is 95%, and d = 0.05, t, = 1.96 at 95% 
estimation level. At the same time, give p assurance 0.75 by 
pre-sampling. Then n = 264. In order to ensure the sampling 
accuracy, we give the calculated sample size a 5% above. And 
the last sample size equal 278, total sampling ratio is about 9% 
all over China. Here, we did not calculate sample size based on 
variance between clusters and within cluster because sample 
size of simple random sampling is larger than that of cluster 
sampling and could result in a more accuracy sampling. 
During the sample selection, map sheets bestriding on more 
than one stratum is partition into the strata with larger area. 
When sample size of a stratum is less than 1, let the sample size 
to 1. At last, sample size all over china is 282. 
42 Image processing 
For crop proportion monitoring, Landsat TM or CBERS CCD 
images were processed under the following 7 steps: 
— Atmospheric Correction. Use meteorological observation 
data to remove the effects of atmosphere components, such as 
03, CO2, water and aerosol et al. the algorithm used here is 
Modtran. 
—Geometric Correction. Use topographic maps to determine 
the coordinates and to correct the distortions in satellite images 
arise from sensor, changing altitude, height and speed of the 
satellite, the angle of the orbit path in relation to earth, and the 
rotation of the earth under the satellite. In the correction 
procedure, UTM system and polynomial model were applied. 
—SAVI layer Calculation. SAVI is a good derived index to 
reflect ground features in the imagery and will be used in the 
classification process later. Here, SAVI is calculated using the 
following equation, 
SAVI = ((TM4 — TM3) / (TM4 + TM3 + 0.5) + 1) x 100 (5) 
--Image Composition. Combine SAVI layer with the 6 TM 
bands. 
285 
—-Non-arable Area Removal. Remove non-arable area assisted 
by land-use map assisted by land-use maps. This is done to 
reduce calculation time for the classification process, and to 
make sure that the calculate area relates only to the arable area. 
—Unsupervised Classification. Use an automatic clustering of 
ISODATA algorithm to create classes. This classification 
procedure can be repeatable, and can be accepted by the 
running of CCWS. 
— Labeling. After the classification, every cluster is identified 
to specify crop area or not. Here veteran personnel especially 
ground survey person take part in the labeling procedure. 
4.3 Crop proportion calculation 
CCWS estimates crop proportion using unequal cluster- 
sampling estimation methods. That is, Crop proportion p is 
estimated by equation (3) 
n 
S d: 
= i=l 
> nm; 
z (3) 
Where: p is crop proportion, and n is cluster number, also 
means the number of image frames, a; is all the crop pixels in 
the images and m is all the pixels in the images. And the 
variance of estimation is estimated by equation (4) 
lar Ÿ (a, - pm, y 
a 
p 
nm. nl (4) 
] Di 
m -—-— m; 
Where: "= means average crop pixel numbers in every 
Jdzn/N HE iur 
image | means sampling ratio. At last, estimation 
accuracy is estimated by equation (5) 
AQ) 71-1, Mp) / p ái 
Where: la 
distribution. 
Here, crop proportion calculation include 3 levels, firstly it is 
accounted on image level using equation above, and then 
account to strata level by weighted average, the weight here is 
the arable land area of every image in the stratum. At last, crop 
proportion on province level is accounted also by weighted 
average methods, the weight here is the arable land area of 
every stratum in the province. 
is obtained from the assumption of normal- 
4.4 Image selected and date request for crop proportion 
monitoring 
CCWS mainly estimate acreage of 7 crops, including winter 
wheat, spring wheat, maize, early rice, middle rice or single rice, 
later rice and soybean. 
CCWS mainly selected Landsat TM as image resource to 
monitor crop proportion. Sometimes, CBERS CCD had also 
been used to assistant Landsat TM. At the same time, it is 
difficult to get optional data in the limited time slot for 
meteorological reason in some regions, especially when rice 
acreage monitoring in summer in south China. For the running 
of CCWS, radar data is also used for rice acreage monitoring in 
low optical data acquisition area, especially in south China. 
For crop proportion monitoring, data phase request mainly 
depend on the spatial distribution of crops, phonological 
calendar, crop rotation custom and growing stage. In China, 
crop plant phenomena had been very complex except summer-