Vegetation Classification Research Of Grassland Based On Decision Tree Approach In Xinjiang Yili Area

Grassland resources not only is the world’s largest natural green barrier, but also has a great impact on the development of economic, ecological. With the “3S”(GPS、GIS、RS) the development of remote sensing technology for investigation and evaluation of grassland resources to provide a more efficient and accurate techniques. Remote monitoring of grassland resources in the range of a large-scale, small-scale spatial and temporal characteristics, can real-time monitor the status of grassland resources. Extensive application of remote sensing technology to provide reference for grassland resources in Xinjiang Yili area reasonable utilization and evaluation.Based on the Yili region as the research object, this paper using MODIS data and climate data for 2013 after the survey data preprocessing, on the basis of remote sensing images using the computer for the decision tree classification of grassland resources in the study area are classified, then the classification results matched the visual interpretation to verify the accuracy of classification. This paper mainly study from the following aspects:(1) Inversion the grassland resources in Xinjiang Yili areaBased on the Yili area 146 grass sample plot survey data as the foundation, through various factors of biomass and soil bulk density on weight parameter analysis and weighted fusion, using Arc GIS software, the inversion analysis of vegetation biomass and soil bulk density of space distribution of Xinjiang Yili region. The results showed that: the biomass and soil bulk density inversion results are basically consistent with the Yili area topography, geomorphology, climate characteristics, reflecting the spatial distribution of the Yili region of grassland vegetation characteristics;(2) Visual interpretation of the establishment of grassland resourcesAccording to the interpretation principle, using artificial visual interpretation methods to superimpose the first grassland survey data in 1980, the 1:250000 land-use maps and TM image data, the final completion of visual interpretation of the Yili region of grassland resources.(3) Automatic identification of grassland resourcesBy analyzing the index and vegetation communities of the Yili region and other informations, such as the characteristics of different grassland type accurately described, six indicators from the grid into a point, and then through the multi-value extract add to all indicators point on an index. Extracted of grassland types by attributes according to the range of the threshold value and converted point into a grid, output of each grassland type distribution. The overall accuracy of the decision tree classification reached 68.45%, Kappa coefficient is 0.5205, overall decision tree classification in rangeland resources classification has a certain reference value. Based on expert knowledge of the decision tree classification, alpine meadow, warm mountain meadow, warm meadow steppe, warm desert and other grassland types of classification accuracy is higher, mapping accuracy from 97.13% to 100%, user accuracy from 81.57% to 100%, illustrates the application of decision tree classification in these types has a high reliability. But the warm grasslands, warm desert grasslands, lowland salt meadow classification accuracy is low, misclassification error of up to 35.19%-85.53%, the reason needs further analysis.