Research On Land Cover Classification In Mongolian Plateau Based On MODIS Data

Remote sensing image classification is an important research in remote sensing field. How to access to the information of land cover accurately and rapidly in a large scale, and use this information to the dynamic real-time monitor of land change, is not only our focus but also the basic data of describing the ecosystem. Knowing its status quo and changing has a practical significance. Firstly, it reveals the feature and developmental change of land cover. Then it helps investigating the driving force of changing. Moreover, it is good for analyzing and appraising the regional ecological environment.This text directs many characteristic of spectrum against of MODIS data, choosing MODIS image of the whole year of 2006, calculating NDVI time array datum of 12 month, and analyzing the spectrum relation among the land cover type, then choosing the categorized indexes such as NDWI, NDMI, NDSI, etc. we chose Bayannur, Hulun Buir as test regions from plateau of Mongolia at first, for they having different climate, ground form, vegetation, and classifying the land by minimum from law, most big network law, law of likelihood and nerve of BP making policy, then we chose the good categorized method by categorized precision of the obscurity matrix to classify the whole plateau of Mongolia and get the final categorized result.This text has got the following main conclusion through this job:(1) According to the characteristic of wave band of MODIS data, the text regarding adding, reducing, riding ,excepting etc of wave band as characteristics categorized to classify land, it obvious find that it improves classified result, and has good in choosing the effective categorized characteristic by studying background knowledge of the studying area and professional knowledges.(2) From the MODIS remote sensing data, we draw vegetation index, water body index, humidity index and soil luminance indexes. After studying various indexes on the change law on the type in the land, it can offer basis for setting up of decision tree, thus more quickly and better classify the research district can obtain.(3)The choice of the classifying device has important meanings in improving the categorized precision of the image. In the categorized test of this test, we selected two areas of difference climate, vegetation, ground form as district of studying in order to further explain the classifying device and classify in the area of different research. After using means of Minimum distance, Maximum likelihood, BP Neural Network, and Decision tree to classify the test area, and from the categorized precision of confusion matrix, we found that decision tree means has the highest classification accuracy. Therefore, considering of the specificity of the whole Mongolia plateau, we got high-accuracy land cover.(4) In the classification of this text, the traditional categorized method needs more training samples in classifying, and the tall sample in purity could guarantee the categorized precision, but this text chooses then turns into what ENVI software carries on while choosing to train samples from Arcgis software. We space its position apt to take place displacement and train inaccuracy that sample chooses such conversion, thus influence the categorized precision. Such categorized method is used on area of large yardstick and classified time-consuming and strenuously. It is unsuitable to choose. BP nerve whom network behave at not classifying enough steady, different district has categorized precision differ greatly, and this method is expected relatively much to the precision which trains the sample, and longer reason of training time of this method, do not choose it as one of the reasons why the plateau land in Mongolia covers the categorized method either. But the decision tree not merely demonstrates the categorized characteristic of high precision in classifying, and it is classified fast, can finish setting-up and classification of the decision tree fast.(5)This test has examined the finally classification map by confusion matrix. For this paper is lack of the land use data of Mongolia, we just chose the test sample uniformly in Inner Mongolia to verify the classified result of land cover of the plateau of Mongolia. The conclusion is that the overall classification accuracy is 80.28 percent (reaching the lowest discrimination accuracy that allowed), kappa coefficient is 0.7152(between 0.6 to .97). It is a good result in the sense of quality of classification. Therefore, using decision tree means for the classification of plateau of Mongolia is feasible and worth promoting.