il 2004
M—ÀÀ
select
2 strata
It from
is the
. When
'S with
ed 2%
. Here,
tio n =
ngth is
ransect
ground
region
am for
control
ncy for
ı patent
team to
tion of
m (Wu
ong the
ture by
picture
m.
ted for
of crop
yi. Nis
ypes in
ly it is
evel by
evel is
here is
n road
mpling
port the
1 China
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
State Statistic Bureau, road density of almost all the provincial
administrative units in China is above the required level. In
Shandong province, the average road density is 47%, and all
cities except Dezhou have a density more than 4%, required for
the 2% sampling ratio. It is a truth that the road density may be
less than 2% in some area in some province, but we could get a
fact by experience that the convenient road or pathway between
villages can also be used as transect belt. We had taken a
ground survey in Taigu region in Shanxi province. Within a 5
KM by 5 KM test area, there are about 255 KM roads that can
be used as transect belt, the density is 10.2 high (Li Qiangzi,
2004).
Another argumentation we must answer is that the crop plant
structure may have difference with that in other and whether
transect along roads may draw a bias for crop proportion survey.
This phenomenon exists in some area. We can seen that there
are more vegetable areas and fruits farms along the road, but we
have not carried out a farther research about this, but we
mapped out a remedial schema for transect selection. That is,
keeping away from highways when building transect-sampling
frames and roads should clipped into segments no longer than 2
kilometres.
5.5 Crop type proportion sampling precise assessment
In order to assess the precise of crop type proportion survey, we
began to set test areas from 2000. Jiangning test area in Jiangsu
province and Dehui test area in Jilin province set in 2000, and
Shuangyang test area also in Jilin province, Luancheng test area
in Hebei province and Xinzhou test area in Hubei province set
in 2002. Every test area is 5 KM * 5 KM size.
In every test area, firstly we build transect frames and select
randomly the transect belt and carry out the ground survey with
help of GVG survey system. After inside interpretation and
statistics, we get the crop type proportion surveyed from
transect sampling. At the same time, our survey team went to
every field and wrote down the crop type assisted by high
spatial resolution images or big scale land-use maps, after area
caleulation and area account, we got the exactly crop type
proportion. At last, by comparing the two crop type proportion
value from different ways, we assessed the precise of transect
sampling technique and GVG survey system (see table 2).
Testarea — 2000 | 2001 2002
Rice | Rice | Rice Maize Winter wheat
Dehui 94.3 | 94.4 - - -
Shuangyang - - 95.9 025 -
Luancheng - - - 100.0 98.9
Xinzhou - - 89.1 -
Jiangning 96.0 | 99.9 | 92.4 f
Table 2: Transect sampling precise validated from 2000 to 2002
We found that rice proportion survey precise changes between
89.1% and 99.9%, most survey precise are higher than 94%.
Winter wheat proportion survey precise is also very high and
can be 98.9%. Maize proportion survey precise is 92%. All the
test results showed us that crop type proportion survey through
ranseet sampling and GVG survey system is precise enough to
meet the running request of CCWS.
257
Another precise test result from Taigu test area in Shanxi
province told us that, Although maize and soybean proportion
precise are more than 95% high for their not low proportion, but
using transect sampling and GVG survey system to survey crop
type proportion may survey off some very low proportion crops,
and this lead to a system error. That is, crop type proportion
surveyed may be a very small higher than the “true value”. This
system error should be calibrated by experience and be
eliminated later (Li Qiangzi, 2004).
6. CROP AREA ESTIMATION PROCEDURE USED BY
CCWS
Under the total crop area estimation methodology, CCWS
estimate crop area at strata level in first, then deduced to county
level or city level. At last, crop areas were summed up at
province level.
Because the methodology has a little difference between using
optical images and using radar images, we must discuss it here
respectively.
When using Radar images to monitoring rice area, the
estimation is very simple, that is,
Rice area — arable area * rice proportion
Here rice proportion is also monitored using cluster sampling
and radar images.
When using optical images, crop area estimation includes two
steps. First, CCWS estimate crop total area by multiple arable
area and crop proportion monitored using cluster sampling and
remote sensing,
Crop total area = arable area * crop proportion
Then, CCWS estimate the area of every main monitoring crop
type by multiple crop total area and crop type proportion
estimated using transect sampling and surveyed by GVG survey
system.
Crop area = Crop total area * crop type proportion
After the estimation at strata level, CCWS deduced crop area
estimation to county level or city level. Here, CCWS first
deduced crop proportion and crop type proportion to county
level or city level, that is, all counties in the strata has the same
crop proportion or crop type proportion with the strata, and the
city level crop proportion and crop type proportion is the arable
area weighted average of all counties in the city. Then, county
level of city level estimation is based on the corresponding
arable area, crop proportion and crop type proportion.
At last, province level crop area are accounted using weighted
average method, here the weight is the arable area or crop area
within the province. For the crop area is more feasible than
arable area, if crop areas in the last can be used, we use crop
area as weight. For the crop area is always not available, we
have not accounted province level crop area from strata level.
7. A CASE STUDY OF EARLY RICE AREA
ESTIMATION IN 2003
CCWS began estimate the early rice area in April. Ground
survey teams began to execute sampling groundwork in Fujian,
Jiangxi, Guangdong, Guangxi, Hainan, Yunan province during
the second 10 days in April, survey teams to Zhejiang, Anhui,
Hubei, Hunan, began their fieldwork in the second 10 days in
May. All ground survey works finished in the first 10 day. After
inside interpretation and accounting, crop type proportion data
were transferred to CCWS during the second 10 days in June.
Images orders were sent out during the first 10 days in June,
including 14 Landsat TM images and 2 Radarsat Scan SAR
images (see Figure 3). After 10 days processing, crop
