3. Realization of High Medium and Low Yield Farmland Classification

3.1 sampling in training areas and Initial Classification

3.1.1 Selection of information source
Landsat TM data of August 12,1987 and August 26,1994, and 340 pieces of color infrared aerial photographs (23cm*23cm) of September 8, 1987 with a scale of 1:19000 scale were used. 160 pieces of photographs were scanned.

3.1.2 Materials
Materials such as maps of land use, soil survey, hydro geology and water conservation were collected. Data on natural environmental characters and social economical circum-stances were collected. The landform maps of Xinding Basin area at scale of 1:100000 and of Dingxiang county at 1:50000.

3.2 Classification System of High Medium and Low Yield Farmlands
High yield farmlands:300kg/mu and more (wheat)
Medium yield farmlands:150-300kg/mu(wheat)
Low yield farmlands:150kg/and lower(wheat)

3.3 collection and Introduction of Special Introduction
A. Introduction sources incluce:7 band TMCCT data and images of Aug.12,1989 and Aug.26,1994, 340 pieces of color infrared air photographs of Sep. 8,1987 and 1:50000 land use map Dingxiagn county. A set of 30 agricultural special maps for Dingxiag county, color infrared air photograph and field investigation provide with hierarchical inquiry of agricultural information.

3.3.1 Irrigation:
Irrigation land with full water supply, high yield farmland
Irrigated land with full water supply (one watering / year), medium yield farmland
Irrigation system but No water supply, low yield farmland

3.3.2 Soil salinity
Non saline soil, 0.1% and lower
Light saline soil, 0.2-0.4%
Medium saline soil, 0.4-0.6%
Severe saline soil,0.6-0.8%,low yield farmland
Salinity soil, more than 0.8% low yield farmland
3.3.3 Slope: four types of hill farmland are recognized based on slope :
0.60 : plan
60-150 degree: gentle steep
150-250 degree : steep, low yield farmland
and >250 : very steep, low yield farmland


3.4 Method of compound Hierarchical classification
In the practical operation, firstly, according to landform characters of Dingxing county sub-regions were divided before classification to special factor of irrigation, slope and soil salinity.

Secondly, composting TM image with the sub- region map, a color image sub-region was got. Six sub regions were divided. In this way class numbers for every sub-region were much less than the whole region map. The phenomenon of same spectra for different object for different spectra was decreased.

Because numbers of classes were decreased and area of sub-region was reduced, time used for computer classification was cut down greatly. It is benefit to computer training of sample area and modification of the classification results. Flowchart of classification is as follows. this method integrated remote scenting information for interpreting high middle and low yields farmlands with non-remote sensing information which effect on yield. ?The factors of irrigation, soil salinity and slope were analyzed comprehensively. Classes tend to simple because of sub region divided. As a result, accuracy of classification was increased greatly.

In this study, Landsat TM data were the main information source, corroborated with color infrared aerial photographs and field inspection. A farmland monitoring system was established, which is supported by RS, GIS and multi-media technology.

Considering multi-factors and complexity in classification of high-medium-low yield farmland, composite hierarchical classification with multiple information sources, assisted by GIS and based on statistical decision was used. The non-remote sensing factors, such as topography, elevation ,irrigation soil salinization etc, were selected to complex with TM data image. Image were interpreted by comprehensive analysis and compared with all kinds of scientific knowledge and ability of visual interpreters. Actually, at first the regions were divided followed classification of the lastly images, at last composite hierarchical divisions were combined.

Thus, the maps of the sub-region for high-medium-low-yield farmland in Dingxiang county were made according to topographic characters, irrigation, soil salty and soil slope. Six sun-regions divided

Sub-division resulted in less number of crop types in each sub-region. Interpretation errors resulting from same targets with different spectra or different targets with same spectra con be overcome. Flowchart of classification is as follows:


Flowchart of classification of composite hierarchy
In the final stage, combining the classification result of six sub-regions, the classification map of high-medium-low yield farmland of Dingxiang county was made. Accuracy of classification was increased greatly by using GIS.