A practical model for estimating the arable land change of
China using remotely sensed imagery
Zhongchao Shi 1 , Liu Haiqi 2 , Ryousuke Shibasaki 1
1 Center for Spatial Information Science and Institute of Industrial Science ,
the University of Tokyo
4-6-1 Komaba, Meguro-ku,Tokyo 153-8558 JAPAN
Tel. +81-3-5452-6413 Fax +81-3-5452-6414
Email : shizc@skl.iis.u-tokyo.ac.jp
2 Dept. of Development Planning, Ministry of Agriculture,
No.11 Nongzhanguan Nanli, Beijing, 100026, P.R.China
Tel/Fax:(010)64192552
E-Mail: Liuhaiqi@chnmail.com
Keywords : Arable land Change, Sampling, Modeling
Abstract A model which can estimate the nationwide arable land change based on number
of sample data is proposed and described in this paper. Since the accuracy of the estimation is
mainly depended on the number of samples and the sampling method, a method for selecting the
optimum sampling number and optimum sampling location is proposed. The accuracy and
efficiency of the proposed model was examined in China. The arable land change in totally 2550
counties were estimated from 238 samples(counties). By comparing the estimated results with the
national statistic data, the mean error of the estimation is below 10 percent which satisfies the
requirements for national arable land change investigation.
Introduction
To master the arable land and its change is very important for making national agricultural planning of
a country. However, because of the limitation of budget, it is usually difficult and sometimes not
necessary to use time series TM images which cover whole country for such a purpose. Generally, it can
be solved by using so-called sampling technique, which can estimate the nationwide arable land change
using just number of samples.
Basically, a sampling data based estimation method should satisfy following four requirements:
- Should be operational and feasible;
-
With low cost;
- With high accuracy and reliability;
- Can estimate dynamically(high efficiency).
Generally, the accuracy, reliability and efficiency of the estimation are mainly depended on the
number and locations of samples as well as the estimation method. That is, if the sampling number is too
big, the estimation will be costly and takes time to fulfill. Conversely, if the sampling number is too
small, it may not get high accuracy. On the other hand, the distribution of samples(location of sample)
has to be take into account for obtaining optimal estimation.
In this research, China was selected for testing. Because the national statistical reports of China are
generally made from county level data, “county” is usually the smallest unit for statistical purpose.
Hence, “county” will be used as the sampling unit in our discussion.
Statistical Arable Land Change Data
Figure 1 shows the distribution of statistical arable land changes of China in 1993 which includes
2550 units with an interval of 100 mu(mu is a widely used unit for present the amount of land, 1mu =
1/15 ha).Following units were not included :
- 75 county-level units in Tibet Autonomous Region because there are no available data from Tibet;
- Some municipal districts such as Beijing’s Eastern District and Western District because they are
not arable lands;
There are 9 units which are called special units in this paper because each of their absolute changes is
more than 100,000 mu.
Fig.1 Distribution of arable land changes of China in 1993(statistical data).
Arable Land Change Estimation Model
Class number
In order to estimate the nation-wide arable land changes from limited sampling data, we have to at
first classify the estimation units into several classes. Because the arable land changes involve quite a
few random factors, the number of classes should be determined in order minimize the sampling rate. In
this research, 6,8,10,12 classes were tested and 6 were found the best class number for estimation.
Class interval
According to figure 1, the accumulating value of squire root (f) of arable land changes can be
calculated. The value is 653.4 which will be classified into 6 classes. Therefore, the interval of
accumulation value of each class is 653.4/6 = 108.9. It is then easy to determine the start and
end(boundary) values of each class as :
class 1: ( -
¥, - 11800]
class 2: (- 11800, - 3800]
class 3: (- 3800, - 900]
class 4: (- 900, + 300]
class 5: (+ 300, + 3600]
class 6: (+ 3600, +
¥)
After the boundary value of each class has been determined, the numbers of units in each class can
be countered easily from the figure 1. Here are the units in each of 6 classes :
class 1: ( -
¥, - 11800] 115
class 2: (- 11800, - 3800] 223
class 3: (- 3800, - 900] 454
class 4: (- 900, + 300] 1187
class 5: (+ 300, + 3600] 419
class 6: (+ 3600, +
¥) 143
sum 2541*
*As mentioned in section 2, 9 special units were removed from the total population of 2550 units
because the change are very large.These 9 units should be investigated by inventory.
