| Under the dual impact of climate change and human activities,the structure and function of ecosystem in Tibetan Plateau have undergone profound changes,grassland degradation,desertification expansion,glacier retreat and other issues have become increasingly prominent,which has a different degree of impact on the plateau ecological security barrier function.As the entire plateau area is vast,the internal natural conditions and the use of resources vary greatly.Therefore,the main ecological functions carried by each district and the environmental problems and causes are also different.At present,the research on land use / land cover in Tibetan area is mainly concentrated in the urban areas affected by human activities,as well as the important ecological fragile areas,such as the three rivers headstream regions,Qaidam region.While the Changtang Plateau is distributed as the hinterland of the Tibetan Plateau The most common alpine grassland ecosystem,which is the most unique and ecological function,is the typical region with extremely fragile ecological environment,and the research is relatively few.This is related to the natural conditions such as poor accessibility,poor environment and climatic conditions in the Chngtang Plateau.It is difficult,time-consuming and costly to study the land cover by using a large-scale field investigation in the field,which led to the current scientific investigation of the region less.Previous studies have focused on single land cover types,such as grassland,desert,wetland,and so on.However,in the Chngtang Plateau region,there is still a lack of complete land cover research results.At the same time,a lot of land cover classification system is simple or not suitable for the special natural conditions in the region,and its classification system needs to be further studied.Therefore,it is necessary to study the land cover classification and mapping in this area,which is of great scientific significance to fully understand its regional characteristics and regional differences.In this paper,through 5 times(2012 ~2016)field investigation work(obtaining more than 6185 land cover sample data),after being fully familiar with the overall profile of the Changtang Plateau and comprehensive comparative analysis,we chose the remote sensing image of GaiZe county as the typical region of the Changtang Plateau(satellite orbit No.143/37 Landsat 8 OLI image).With reference to the existing study and formulate the appropriate classification system.Based on Landsat 8 remote sensing images,we chosed a variety of methods for land cover classification,including maximum likelihood classification,IsoData classification,neural network classification,based on improved C5.0 decision tree classification and object-oriented classification,and compared and analyzed the classification results.The main conclusions of this study are as follows:(1)Through years of field investigation and review of relevant literatures in this area,based on the national land classification standards,taking full account of the Chngtang Plateau special geographical location,cold and dry climate conditions and fragile ecosystems and other realities,take the constructive plants' life forms and communities ecological appearance as the main basis.Finally,a proper land cover classification system is developed,which consists of 6 level I and 13 level II.(2)By comparing the MODIS-NDVI time series data of different types of 2016 years,we selected the time of the highest spectral difference of each cover type as the remote sensing information.In this paper,the remote sensing image of June 2016 was selected as the final classification image.On the basis of the above image,we made a statistical analysis of the characteristic variables by using the sample points,feature extraction of remote sensing image include 7 original bands,the first 3 parts of principal component analysis and independent component analysis,the first 3 components of the tasseled cap transformation,NDVI,EVI,NDWI,MNDWI,NDSI,SAVI and other vegetation index,land surface temperature data etc.(3)We extracted the land cover remote sensing information in typical region by multi methods,it is found that the improved C5.0 decision tree classification method is the best,the overall accuracy is 77.52%,and the Kappa coefficient is 0.75,which is better than the traditional supervised classification and unsupervised classification method.Through artificial selection of characteristic variable combination and C5.0 algorithm to automatically generate classification rules for training samples,it can quickly select the best combination of feature variables and make the classification rules from the combination of various feature variables,significantly improving the classification efficiency.The classification rules have a certain reference value for the classification of other areas of Chngtang Plateau.However,object oriented classification results in this area was not good,the overall accuracy is 68.92%,kappa coefficient is 0.62.The possible reasons for the differences in the texture of surface types,such as texture,which is caused by the special climatic conditions and sparse vegetation types in the alpine region,the specific reasons for further study.(4)We should choose the classification method flexibly according to the difference of the research purpose or the difference of the type of land cover.When we choose the classification method based on the producer's precision,neural network classification method has the highest precision for alpine grassland,and the support vector machine has the highest precision for extracting bare land,and other types of extraction and selection decision tree classification.Neural network classification has the highest precision for the extraction of Stipa grandis,and the maximum likelihood classification has the highest precision for lakes,rivers and glaciers and perennial snow.The object-oriented classification has the highest precision for the swamp,and other types of extraction selection decisions Tree classification. |
PDF Full Text Request