GISdevelopment.net ---> AARS ---> ACRS 1999 ---> Agriculture/Soil



The Precision Analysis of Soil Use Investigating in a level of County with the Data of Spot Imagine

Li Bingbai, Yu Chumshen, Chen Yuquan, He Wei and Zhang Jinghong
Jiangsu Academy of Agricultural Sciences, Nanjing 210014 China.


Abstract
Land use classification is very important for economic development in a region. It is very useful to investigate land use change with the technique of remote sensing. The key problem is the precision of investigation in the process of using remote sensing technique. The precision has been discussed with the Spot imagine data in a level of county in this paper. According to the method of " the judging between man and computer" , cultivated land classification was made for five towns of Changsu county, Jiangsu province, using 1/50000 Spot imagine data in 1994 as an information source. The results showed that the accuracy of total land, cultivated land paddy field, house space, roads and water area were 99.2%, 99.5%, 96.7%, 97.2%, 97.5% and 96.6% respectively, comparing with aerial photo survey. While secondary classification was not good enough. For example, in water area, River Lake, fishpond and ditch were about 85%, especially ditch was only 73.4%. The reason was as followings: (1) the time of imagine was not suitable. (2) Little stream and pond could not be selected , may be mixed into cultivated land. (3) Because of change of cropping system, the imagine of vegetable field and lawn my be with that of house space in water.

These problems may be corrected by selecting adoptive time spot imagine and redressing in small and string field . The precision can be achieved at 90-95% for dry lands, roads, water area and villages. The man accuracy of total imagines investigating can be high as 95-98%. In this way, Spot imagine can be use to classify land use in a level of county and monitor land use change, comparing with an old one.

1. Introduction
As the development of reform opening and economic constructing, lots of changes have been made in soil use. In the early 90's, almost 60000 hectares land was reduced from 1992-1995 in south Jiangsu, the reducing rate was as high as -3.2%. Also cropping system was changing in the region some body field may be changed into fish pond or vegetables. So it is very important to clear soil use changing for government management and planning. Usually the land is detected by aerial survey, but it is too expensive and needs a long time. It is very useful to investigate land use change with the technique of remote sensing. The key problem is the precision of investigation in the process of using remote sensing technique. Generally, the accuracy of "TM" imagine data (resolution 30*30) investigating is around 85-90%. That is not good enough for a level of county, especially for the investigation of land use in agriculture.

Spot imagine (resolving power 20*20 m for color, 10*10 m for panchromatic film) has a better resolution. We would like to confirm how the precision can be achieved and how we can do the analysis in a level of county with a spot imagine.

2. Data Acquisition and Processing
2.1 Data Source.
ACD composition Data of 1,2,3 ware band in May 9 1994 was employed.

2.2 Experiment Site:
The experiment was taken in Liantang, Zhangqiao, Yangyuan, XingZhuang and Mocheng 5 counties of Changsu city. Total soil area was 18720 hectare.

2.3 Multi-source data
Available conventional data and maps for this study include:
  1. Land Use - Map of Changsu (1: 50000), Edited by the Changsu soil management Bureau, 1994.
  2. topographic Map of Changsu (1:50000), produced by Survey Bureau of Jiangsu province, 1989. The map includes: specific datasets: county boundary, town boundary, canal geomorphology, landslide development river, main system, swamp, topography, vegetation, soil and water system.
  3. Ground truth data acquired by Changsu Bureau of soil management.
2.4 Land Use classification System
It is well known that there were both scientific needs and social needs for land use data. Some land cover surface properties and land use information should be derived from remote sensing data. In order to meet these needs, a land use classification system was built for Changsu county according to the rule made by the National Soil Management Bureau. The explanation of land use classes is showed in table 1.

Table 1: Explanation of land use classes
Item Code Explanation
Cultivated soil
Paddy
Dry land
Vegetable
Lawn
Winter spare land
Orchard
Forest
House Space
Town
Village
High way
Village road
Water area
River
Lake
Pond
Fish Pond
Ditch
10
11
12
13
14
15
21
31
50
51
52
61
62
70
71
72
73
74
75
All paddy and dry land
Paddy
Cotton or rape field
Vegetable
Lawn
The land is waiting for rice seeding
Orchard
Forest
Town house space
Town
Village
Road is wider in exceeding 6 meters
Width is less than 6 meters
Water area
River
Lake
Pond
Fish pond
Ditch

3. Methodology on Land use Classification
3.1 Outdoor Investigation:
After investigating in every town, a classification standard of land use type was determined. When finishing classification, error was corrected in the field.

3.2 The judging between man and computer
  • Run a software named Coreldraw 8.0
  • Site new map layer was created: rs, tif, xz, poly, label, and line.
RS layer: input Spot imagine data.
Tif layer: the topographic map of Changsu county was input and matched with imagine data.
Xz layer drawing town boundary in imagine data with matched topographic map.
Poly layer: drawing land type boundary according to graph and point with the method of " the judging between man and Computer"
Label layer the code character of every graph was labeled in the layer
Line: Drawing small river and string roads and making code.

3.3 judging accuracy requirement :
Imagine must be enlarged 1/25000 to draw and classify.
Polygon: Minim graph must be 3*3panel. The width of string land must be more than 2 panels.
Road, River: the width is more than 2 panels, they may be considered as polygons.

3.4 Edit and area calculating
The map was edited and area of every type land for every town was calculated with software of Mapgis or ARC/Info.

4 Results and Discussion
4.1 Results and Analysis of primary classification.
Data from spot imagine were classified and distinguished by " the judging between man and computer" method. The accuracy of classification of total land was 99.2%, comparing with theoretical area derived from 1/50000 topographic map, the precision of total area in every town was about 99.80-99.99%, the results were showed in table 2.

Table 2: test results of classification of Total Land
Town Statistical of total land (km2) Theoretical land area (km2) Accuracy (%)
Liantang
Zhangqiao
Xinzhuang
Yangyuan
Mocheng
Total
46.45
34.10
37.38
32.04
37.08
187.17
46.46
34.21
37.42
32.06
37.11
187.30
99.99
99.87
99.93
99.92
99.98
99.92

The primary classification precision of cultivated land, paddy, house space, road and water area were 99.54%, 96.7%, 97.5% and 96.6% respectively, comparing with aerial photo survey finished in 1994. The results were showed in table 3:

Table 3: Test results of classification for five towns
Item Code Statistical of classification (hr) Aerial photo survey (hr) Accuracy (%)
Cultivated land
Paddy
House spa
Roads
Water Area
10
11
50
60
70
106.8
113.2
29.7
5.6
42.5
106.3
102.8
30.6
5.5
44.0
99.54
96.7
97.2
97.5
96.6

4.2 Results and Analysis of Secondary classification
The secondary classification of land use was not good enough, only about 65-85%, For example: the precision of river, village, dry land and lake was more than 86%, but the results of town, vegetable land and fish pond were less than 70%.

Table 4: Test results of secondary classification for five towns
Item Code Statistical (hr) Aerial photo survey (hr) Accuracy (%)
Dry land
Vegetable land
Town
Village
River
Lake
Fish pond
12
13
51
52
71
72
74
3855
979.7
9052
35553
33319.9
13154
8670
3896
1351
13670
32205
38486
14093
10861
89.5
72.5
66.2
89.6
86.6
93.3
79.8

The reasons were as followings: (1) The time of imagine was not suitable. In May, the imagine of rice seeding land, vegetable field were confused with that of house space and dry land. (2) The road of village could not be distinguished. (3) Little stream and pond could not be selected, may be mixed in town and cultivated land. (4) Because of change of cropping system, the imagine of vegetable fields and lawn may be mixed with that of house space in winter.

These problems may be corrected by selecting adaptable time Spot imagine, enlarging imagine into 1/20000 and distinguishing roads, ponds small rivers and other small, string fields. The precision can be 90-95% for dry lands, roads, water area and villages. The mean precision of total Spot imagine investigating may be as high as 95-98%.

5. Conclusion
  • The primary classification of land use with Spot imagine data was proved available. The mean precision can be achieved at about 95%.
  • After considering the areas of small and string field, the results were much better. The accuracy of land, paddy, town, roads and water was 99.2%, 96.4%, 97.5%, 97.3% and 97.3% respectively.
  • The secondary classification of land use was not good enough, only about 65-85%. Apparently this method can't replace of aerial photo survey. Only can be used as the primary classification and measure dynamic change of land use. in this way, Spot imagine can be used to classify land use in a level of county and monitor land use change, comparing with an old one.
References
  • Li Bingbai etc. "Remote Sensing Investigation of Cultivated Land Change in Yangtze Delta Region" China agricultural Resources and Regional Planning Vol. 5, pp. 1-5, 1996.
  • Wu Bingfang and Liu Haiyan, "The Operational Methods for Rice Area Estimation Using Remote Sensing." Journal of Remote Sensing, Vol.1 pp. 58-63, 1997.
  • Chen Jun, "A Application of DTM for classification of Remotely Sensed imagery," Zjournal of Wuhan Institute of Survey and Drawing, Vol. 10, pp.64-72, 1984.
  • Wu Jianping and yang Xingwei, "Accuracy analysis of classification of Remotely Sensed Data" Remote Sensing Technology and Application, Vol. 10, pp.17-24, 1995.
  • Jiang Nan and He Longhua. "Study on Paddy Yield Estimation in Jiangsu Province with Remote Sensing Method", Resources and environment in the Yangtze valley, Vol.5, pp.160-165, 1996.
  • BROWN, J.F., LOVELAND, T.R., MERCHANT, J.W., REED, B.C., and OHLEN, D.O., 1993. Using Multisource Data in Global Land-Cover Characterization Concepts, Requirements, and Methods Photogrammetric Engineering and Remote Sensing, 59, 977-987.
  • RUNNING, S.W., McClelland, T.R., PIERCE, L.L., NEMANI, R.R. and HUNT, E.R. 1995 A Remote Sensing Based Vegetation Classification Logic for Global Land Cover Analysis. Remote Sensing of Environment, 51, 39-48.