Discussion On The Interpreting Of Rangeland Resource Types In The Arid Mountain With Vertical Belt

The rangeland resource is one of the largest natural green barrier on earth, which has moreeffect in ecology and economy. As a high-tech, RS technology provides a convenient and anaccurate way to investigate and appraisal the rangeland resources with the development of thetechnology of “3S”(GPS, GIS, RS). This studying area seted up in the northern slope of TianshanMountains in Urumqi, this paper studied on6rangeland types (alpine meadow, meadow, meadowsteppe, rangeland, mountain desert steppe, desert) and3non-rangeland types (glacier, forest,farmland). The research mainly used decision tree classifier to classify the rangeland resources instudying area by combing terrain factor (DEM, aspect, slope), NDVI, the spectral characteristicsof plants and remote sensing image. Then put the classification result to verify the accuracy withexperts visual interpretation. The results showed:1. Overall Accuracy was71.6859%, and Kappa Coefficient was0.6690. In short, the decisiontree classifier had value to classify the rangeland resources.2. The decision tree classifier had higher classification accuracy in desert, mountain desert,meadow and alpine meadow by researching the evaluation indicators. The Prod. Acc was56.72%-75.67%, the User. Acc was60.39%-94.45%, therefore the decision tree classifier had ahigh reliability. However, it had lower accuracy in rangeland and meadow steppe, the Commissionreach39.77%-59.82%, which need further analysis.3. The decision tree classifier had higher classification accuracy in farmland, glacier, the Prod.Acc was75.67%and72.16%; the User. Acc was71.69%and93.59%. But the Prod. Acc was56.28%in the forest, the User. Acc is65.40%.4. This study existed errors because of the HJ bad phase, more clouds, and the low resolutionof image. From now on, we could try our best to select higher resolution, better phase andmulti-source remote sensing data to improve the classification accuracy.5. The grassland and meadow is difficult to distinguish, on the basis of the above research,we can distinguish them with the help of hyperspectral image.