融合决策树方法的新疆伊犁地区草地植被分类研究

发布时间：2019-03-09 15:09

【摘要】：草地资源不仅是地球上最大的天然绿色屏障,而且对经济、生态的发展有着重大的影响。随着“3S”(GPS、GIS、RS)的发展,遥感技术为草地资源的调查和评价提供了一种更快捷、准确的技术手段。遥感对草地资源的监测在范围上具有大尺度、时空小尺度的特点,能够实时监测到草地资源的变化状况。遥感技术的广泛应用,为新疆伊犁地区草地资源合理利用与评价提供参考。该研究以伊犁地区为对象,采用MODIS数据和气候数据,对2013年实地调查数据进行预处理之后,在遥感影像的基础上利用计算机的决策树分类对研究区草地资源进行分类,然后将分类结果与目视解译结果相匹配来验证分类的精确度。本文主要从以下方面进行研究:(1)对新疆伊犁地区草地资源进行遥感反演文章以伊犁地区146个草地样地调查数据为基础,通过各因子对生物量及土壤容重影响权重参数分析和加权融合,运用ArcGIS软件,反演分析了新疆伊犁地区植被生物量及土壤容重的空间分布特征。结果表明:生物量及土壤容重反演结果与伊犁地区的地形、地貌、气候特征基本吻合,反映了伊犁地区草地植被的空间分布特征;(2)草地资源目视解译的建立采用人工目视解译方法,将1980年第一次草地调查数据、1:25万土地利用图及TM影像数据根据解译原则进行叠加,最终完成伊犁地区草地资源的目视解译。(3)草地资源自动判别通过分析伊犁地区草地植被的指数植被和植被群落等信息,对不同草地类型特征进行精确描述,将六项指标由栅格转化为点,然后通过多值提取至点将所有指标的属性添加在一个指标上。再根据阈值范围按属性提取出草地类型,然后再将点转化为栅格,输出每个草地类型分布图。决策树分类的总体精度达到68.45%,Kappa系数为0.5205,总的来说决策树分类在草地资源分类中具有一定的借鉴价值。基于专家知识的决策树分类对高寒草甸、温性山地草甸、温性草甸草原、温性荒漠等几类草地类型的分类精度较高,制图精度在97.13%-100%,用户精度在81.57%-100%,说明决策树分类在这几类中的应用具有很高的可靠性。而温性草原、温性荒漠草原、低地盐化草甸的分类精度较低,错分误差达35.19%-85.53%,其原因有待进一步分析。
[Abstract]:Grassland resources are not only the largest natural green barrier on the earth, but also have a great influence on the development of economy and ecology. With the development of "3s" (GPS,GIS,RS), remote sensing technology provides a more rapid and accurate technical means for the investigation and evaluation of grassland resources. The monitoring of grassland resources by remote sensing has the characteristics of large-scale, space-time and small-scale in scope, and can monitor the change of grassland resources in real-time. The extensive application of remote sensing technology provides reference for rational utilization and evaluation of grassland resources in Yili area of Xinjiang. This research takes Yili area as the object, uses the MODIS data and the climate data, carries on the pretreatment to the 2013 field survey data, on the basis of the remote sensing image, uses the decision tree classification of the computer to classify the grassland resources of the study area. Then the classification results are matched with the visual interpretation results to verify the accuracy of the classification. The main contents of this paper are as follows: (1) the remote sensing retrieval of grassland resources in Yili area of Xinjiang is based on the survey data of 146 grassland sample plots in Yili area. The spatial distribution characteristics of vegetation biomass and soil bulk density in Yili region of Xinjiang were inversely analyzed by weight parameter analysis and weighted fusion of each factor on biomass and soil bulk density. Using ArcGIS software, the spatial distribution characteristics of vegetation biomass and soil bulk density in Yili region of Xinjiang were inversely analyzed. The results show that the inversion results of biomass and soil bulk density are basically consistent with the topography, geomorphology and climate characteristics of Yili area, and reflect the spatial distribution characteristics of grassland vegetation in Yili area. (2) Establishment of visual interpretation of grassland resources. The first grassland survey data in 1980, 1: 250000 land use map and TM image data were superimposed according to the interpretation principle by using artificial visual interpretation method. Finally, the visual interpretation of grassland resources in Yili area is completed. (3) by analyzing the index vegetation and vegetation community information of grassland vegetation in Yili area, the characteristics of different grassland types are accurately described by the automatic identification of grassland resources. The six indexes are converted from grid to point, and then the attributes of all indexes are added to one index by multi-value extraction. Then according to the threshold range, the grassland type is extracted according to the attribute, then the point is converted into grid, and the distribution map of each grassland type is output. The overall accuracy of decision tree classification is 68.45%, and the Kappa coefficient is 0.5205. In general, decision tree classification has certain reference value in grassland resource classification. The classification accuracy of decision tree classification based on expert knowledge for alpine meadow, warm mountain meadow, warm meadow grassland, warm desert and other grassland types is higher, the mapping accuracy is 97.13% ~ 100%, and the user precision is 81.57% / 100%. It is shown that the application of decision tree classification in these categories has high reliability. However, the classification accuracy of warm steppe, warm desert steppe and low-land salinized meadow is lower, and the error is 35.19% and 85.53%. The reasons need to be further analyzed.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S812

【相似文献】