The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
data provided by the operating department subordinated to the
government. In order to obtain the data of crops acreage every
year in good time, RSAC has to consider two factors as the
operating department, which are time and money. Since the
spatial range monitored is very large and the kinds of crops are
so many, it is impossible to investigate the overall fields either
by RS or by ground survey using GPS. Therefore, RSAC
chooses sampling method using RS and GPS.
2. OTHER WAYS OF GETTING CROP AREA DATA
2.1 In European Union (EU)
Monitoring Agriculture with Remote Sensing (MARS) is a
project facing European in order to obtain crop yield
information constituted by European Union Committee (Liu,
1999). It is a kind of three-stage sampling based on
unsupervised classification (Duda, 2002) using multitemporal
RS data (Panigrahy, 1997). The first-stage sampling unit
composed of 60 sites is designed square with the side length of
40 kilometres. There are 16 component parts in each site and
each part has 40 sampling points. Because unsupervised
classification method is used with RS data to cover the third-
stage sampling points 5-6 times every year, all crops in
sampling units are recognized and then the acreage and yield
will be worked out after statistical calculation from the third-
stage sampling points to the first-stage sampling units.
2.2 In America
In America, the prediction of total crop yield is acquired from
crop acreage and crop yield per unit. The crop acreage data had
been gotten by June Agricultural Survey (JAS) (Hu, 2002).
Two different sampling units used by JAS were area frame
covering America and name list frame. The name list was
composed of registered farmers. Every year about 2400
investigators contacted more than 120 thousand farmers in the
first two weeks of June in order to get crops acreage data.
2.3 In China
The operating prediction of crops acreage is mainly provided by
RSAC. RSAC adopts two methods to obtain the acreage of
main crops such as wheat, com, cotton, soybean, rice, etc. One
method is stratified sampling using RS and the other is ground
random sampling using GPS. Stratified Sampling Method (SSM)
using RS is the major one that works as illustrated by the flow
chart below (Chen, etc., 2000).
To use SSM, the first step is to select the sampling units
according to the surveyed crop and to order appropriate RS
images covering all sampling units in the region. The second
step is to discriminate the crop with RS data and get the area
data of the crop distributed in the sampling units. The last step
is calculating the result with SSM.
Figure 1. Flow chart of getting crop area by SSM
3. SAMPLING METHODS
RSAC selects the sampling methods that include stratified
sampling in spatial regions with RS and Ground Random
Sampling (GRS) with GPS.
3.1 Stratified sampling using RS and GIS
The aim of sampling is to estimate the total quantity of the
object such as crop area since the total quantity is too large to
count and it is of no necessity to survey entirely. Stratified
sampling is a kind of methods like random sampling, multi
stage sampling, etc. It is mainly applied to such case as that the
target individuals are very different from one another but have
quantitative attributes
Selecting sampling unit: Sampling unit has two key points:
quantity and accuracy. Meanwhile, it should be convenient for
operating and calculating reasons. RSAC selects two kinds of
sampling units: one is the polygon of administrative counties,
and the other is the quadrangle frame of relief map on which the
scale used is 1:50,000 or 1:25,000, and longitude difference one
quarter degree and latitude difference one sixth degree (Jiao,
2002). The difference between the two sampling units is that the
former is easier for statistics analysis but the quantity is too
small to satisfy the demand of sampling some crops because the
areas of sampling units are apparently different among the
polygons of administrative counties, the latter is just opposite.
3.1.1 Layers and layer amount: In stratified sampling
survey, layer amount has influence on the effect of sampling.
The proper amount of layers is related with the characters of
number of sampling units. The appropriate amount of layers
will minimize the population variance of total layers and the
sampling cost. During a certain threshold interval, adding layer
number could commonly lower population variance, but
increase the workload. In an attempt to balance the effect of
sampling and the expense of survey, the number of six is
confirmed the maximum layer amount in sampling survey of
crops area according to many tests done by RSAC.
The so-called layer is a kind of data set based on sampling units.
There are obvious differences in sampling size among the layers.
In survey of crop samplings, the sampling units that belong to a
certain layer are generally distributed relatively centralized. The
map below shows the distribution of six layers of Chinese late
rice of 2007, on which sampling units is the quadrangle frame
of relief map with the scale 1: 50000. There are totally 5340
sampling units covering 15 provinces of China in this map. It is