Research Of Land Cover Typef ’s Remote Sensing Separability And Scale Effect In Arid Or Semi-arid Region

Because of the especial forming reasons, the land cover structure of arid and semi-arid regions different from other regions, it has own complex and unique spectral characteristics, which makes the remote sensing land cover classification of arid and semi-arid regions much more difficult. Many new developed multi-spectral and different spatial resolution remote sensing images have some limitations to land cover classification of arid and semi-arid regions. Therefore, the study for land cover types’ remote sensing separability and scale effects is one of the inevitable directions of development for remote sensing technology research, especially necessity to remote sensing classification research in the arid and semi-arid region.This thesis conduct land cover classification for multi spatial scales by using combination of multi-features, and then analyze the remote sensing separability and scale effect of land cover types in Darhan Muminggan United Banner as study area for the transition zone of arid and semi-arid region in the middle part of Inner Mongolia. This study include following three parts and the first is multi-scale land cover classification by K-means unsupervised classification algorithm using MODIS data which consists of different variables such as surface albedo Ref1-7, the first three principal components PC1-3, vegetation index NDVI/EVI, moisture index NDMI, water index NDWI and soil brightness index NDSI with three different spatial scales for 250 m, 500 m and 1000 m. the second is accuracy assessment of land cover classification results by using both of Landsat OLI visual interpretation result and field investigation result. The third is analysis of remote sensing separability and scale effects for multi-scale classification results according to its confusion matrix and error of land cover type’s area. This study focuses on the exploration and discovery of remote sensing separability mechanism and characteristics of the scale effect for the land cover types in the arid and semi-arid region.The results show that with the gradual expansion of MODIS image spatial scales, land cover classification accuracy of arid and semi-arid region has gradually changed. Too small and scattered land cover types cannot be classified on 250 m spatial scale; when spatial scales were expanding from 250 m to 500 m and 1000 m, number of misclassified pixels has increased between farmland and meadows, wetlands and shrub, and both of overall accuracy and Kappa coefficient have reduced, and separability of land cover types which also has declined. With the extension of spatial scale, remote sensing separability for land cover types and scale effect is more significant. Therefore, through this research found that with the change of the spatial scale for remote sensing image, Separability and scale effects for land cover types have some variation in arid and semi-arid regions.