International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
crop area of every type using high spatial resolution images like
Landsat TM or CBERS CCD data, even using very high spatial
resolution data like IKONOS or Quickbird image at proper date.
In some area, crop acreage could be estimated precisely using
remotely sensed data on proper date. But this only limited to
some special crop type in some special area. For example,
winter wheat acreage estimation using NDVI from Landsat TM
in North China could get a high accuracy above 90% in some
area (Xu Xiru 1991). This is because winter wheat there has no
other fusion crop types. This method could not be used in south
China for low proportion of winter wheat and the fusion of seed
rape. Meanwhile, the threshold of NDVI used to extract winter
wheat area varies in different area and different year. At the
same time, tests showed that, identification of different crop
type have a very low accuracy for reason of the cross-planting
and inter-planting phenomena. We have taken a test in Kaifeng
area, Henan province. The accuracy of summer-harvest and
fall-harvest crops is no more than 90% (Li Qiangzi, 2002).
Therefore, crop acreage estimating techniques using optical
data cannot meet the CCWS running request.
So, we need a new methodology integrating remote sensing and
other techniques to estimate with in precise. In this paper,
supported by remote sensing, we provided a stratified two-stage
sampling methodology to meet the running of CCWS,
integrated remote sensing and sampling techniques together.
Using sampling technique in CCWS, not only solved the
difficulties of full-coverage image receiving, but also provided
an efficiency way to control the estimating accuracy.
In the methodology, we used a two-stage sampling procedure to
estimate crop acreage supported by the 100,000-scaled China
Land Resource Database (Liu Jiyuan, 1996). In the first
sampling procedure, we built the cluster-sampling frame by
using 1:100000 scaled map-sheet and selected remotely sensed
data under cluster-sampling technique. Although it is difficult to
extract crop area from remotely sensed data like Landsat TM,
but it is easy to estimate crop proportion accurately (Li Qiangzi,
2004). Using the ability of estimating total area of all crops,
complete estimating the total proportion of crop on the farm in
every second-stratum. In the second sampling procedure, we
use transect sampling to survey crop type proportion. At last,
crop acreage is the product of crop proportion on the arable area
and crop type proportion of every crop type on the planted area
and the arable area.
For the need of sampling, we brought into the stratification. We
divided China cultivated region from three different levels.
After the stratification, we get 11 first-level strata 44 second-
level strata and 102 third-level strata and we estimating crop
proportion at the second-level strata and surveying crop type
proportion at the third-level strata.
But when estimating rice acreage, we always use SAR data and
optical data together. SAR could extract rice proportion directly
with high accuracy (Liew S C, Kam S, Tuong T et al 1998), so
there is no need of ground survey for crop type proportion, and
rice acreage can be calculated using arable area multiple rice
proportion directly.
3. STRATIFICATION
For crop acreage estimation, stratification is an effective way to
increase the estimation accuracy and efficiency. After
stratification, the planting structure, crop rotation system and
crop calendar development is similar within every stratum. That
will be benefit to cluster sampling for crop proportion
monitoring and transect sampling for crop type proportion
survey.
254
During the first level stratification, the main stratification tools
used including atmosphere temperature, precipitation, solar
eradiation, soil types and physiognomy properties and crop
rotation. After clustering the stratification tools and removing
small polygons, we get 11 first-level strata. :
During the second level stratification, the main stratification
tools are the proportions of different crop type at county level,
here we considered the main 4 crop types, including rice, wheat,
maize and soybean. During the procedure, we extract strata
boundary from county boundary for the reason of statistics data
collected on county level. At last, we get 44 second-level strata.
During the third-level stratification, the main stratification tools
used mainly is the farm consistency. Firstly we extract cultivate
area from 100,000 scaled China Land Resource Database and
calculate cultivating consistency in every 1 KM * 1 KM grid,
then, reclassified the cultivating consistency into 4 levels (>
80%, 50-80%, 15-50%, 0-15%) and used in stratification. After
the procedure, we get 102 third-level strata (See figure 1).
Stratification tools
Atmosphere temperature, precipitation,
solar eradiation, soil types and
physiognomy properties and crop
rotation.
rice, wheat, maize and soybean
proportion at county level,
Level
First
Second
Table 1. Factors used in stratification
Figure 1. Stratification used for crop acreage estimation
4. CROP PROPORTION MONITORING
4.1 Cluster Sample Design
China Crop Watch System re-divided strata into clusters and
selected clusters enough to meet the requirement of given
accuracy at second-level strata.
Firstly, we built sampling frames on 1:100000 scaled standard
topographic map sheet. A cluster is a standard 1:1000000 scaled
map sheet with area from 1270 to 1950 KM2 correspond to à
1/16 Landsat TM view which area is about 1977 KM2. Because
the area equality, we use the sampled cluster to select remotely
sensed images to monitor crop proportion (see figure 2).
Internation
THO
Figure
At last, we
is calculate
n= ?
|
Ü p
n, = c P
ü
Where: n i
sampling, !
d are obtai
given samp
estimation
pre-samplii
accuracy, v
the last san
all over Ch
variance b
size of sim
sampling ai
During the
than one s
When sam]
to l. At las
4.2 Image
For crop p:
images wei
— Atmos
data to ren
03i. CO2,
Modtran.
—Geometi
the coordin
arise from
satellite, th
rotation of
procedure,
——SAVI le
reflect groi
classificatic
following e
SAVI = ((1
--Image €
bands,
