2. Methodology
Geographical information system (GIS) increasingly is being employed for wide range of environmental application, which involve integration and analysis of large amounts of spatial data of different scales. GIS is considered an appropriate technology to perform this task. In addition, large amounts of data available from remotely sensed data are being used successfully in environmental application. Based on the main eco-environment problems in the region around Longyanxia reservoir, using remote sensing integrated with GIS, the environmental detection information system has been established. The system has been successfully implemented in an operation application for eco-environmental change detection.

2.1 Eco-Environmental classification system
In order to derive information of eco-environmental changes from remotely sensed data efficiently, the features of eco-environment in high coldness, arid and semi-arid area is analyzed. Among the eco-environmental problems exited in study area, the sandy desertification is the major one, with which other eco-environmental problems have some relations. According to it and the interpretation capacity of remotely sensed TM imagery, a regional comprehensive environmental classification system is built (Figure 1), which is adapted to extraction of eco-environmental information from remotely sensed data. Based on this multi-level eco-environment environmental information, the multi types of environment such as sandy desertification, grassland cover and landuse can be derived easily. Using this system, the effect on classification results affected by spectral confusion of eco-environmental types can be reduced efficiently.


Figure 1 The eco-environmental classification system
2.2 Data acquisition
The multi-temporal and multi-spectral Landsat TM data are applied for detecting changes for eco-environment dynamically. The TM data are obtained under good atmospheric conditions. Bands 1, 2, 3, 4, 5, and 7 are used for the classification. The size of Landsat TM pixels is approximately 30 by 30 meters. The acquisition data for the study area are 1986 and 1996 respectively.

The data for regional physical environment and so economic are also acquired. Some of them are thematic maps in large scale, others are data derived from GIS. These data are used for the object classification.

2.3 Detection procedure
The detection procedure of the eco-environmental changes is designed as follow (figure 2):


Figure 2 The detection procedure
The image are geometrically corrected by identifying ground control points in the original imagery and on the reference topographical maps and then quadratic polynomial transformation equations are applied. Subsequently, the images are resampled by applying the nearest neighbor algorithm. The root mean square (RMS) error for this transformation is less than 0.3 pixel (9m).

The supervised classification is executed using the maximum likelihood classification algorithm, supported by visual interpretation, the application of post-processing techniques and the image information analysis. The TM original bands 3, 4, and 5 are selected as classifications bands.

Classification results are affected by spectral confusion of eco-environment types and mixed pixels. Some types are mixed. This is because the supervised maximum likelihood classification is based solely on spectral observation and generally insufficient to meet the needs of regional study. Therefore, a post-classification technique with GIS is used to improve the classification accuracy. Thematic information that is mixed in the pre-classification results is extracted from remotely sensed data. Then the comprehensive various information is merged in GIS. Through knowledge-based post-classification and modification with GIS, the accurate of pre-classification results is improved efficiently.

Following the classification, the spatula analysis with multi-temporal information is applied to detect the dynamical changes of eco-environment in the region around Longyanxia reservoir.