Landsat-based long_term LUCC mapping in Xinlicheng Reservoir Basin using object-based classification

International Symposium on Resource Exploration and Environmental Science

地球科学;生态环境科学

Su, Wei^1 ; Liang, Dongmei^1 ; Tang, Gula^2 ; Xiao, Zundong^1 ; Li, Jingxin^1 ; Wan, Zhengyu^1 ; Li, Ping^1

Jilin Provincial Academy of Environmental Sciences, Changchun

130012, China^1

Key Laboratory of Agricultural Remote Sensing, Ministry of Agriculture, Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Beijing

100081, China^2

关键词: Accuracy assessment;    Alteration detections;    Hybrid classification;    Land cover datasets;    Multivariate alteration detection transformations;    Object-based classifications;    Overall accuracies;    Rapid urbanizations;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/64/1/012024/pdf
DOI  :  10.1088/1755-1315/64/1/012024

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

Rapid urbanization dramatically changes the local environment around Xinlicheng Reservoir Basin. Landsat images are suitable for the land use change caused by human impact. In order to obtain consistent land cover products, a hybrid classification method combining object-based classification and pre-classification alteration detection method was developed and applied to long-term multi-temporal Landsat images to obtain land cover change information. Object-based classification method was combined with Random forest (RF) classifier to classify the Landsat image in 2008. Then the changed areas in 2000, 2004, 2012, and 2016 were identified by comparing with the images in 2008 via the re-weighted multivariate alteration detection transformation method. The images in 2000, 2004, 2012 and 2016 were classified by RF classifier. Land cover maps for 2000, 2004, 2012, and 2016 were produced by combining the unchanged area in 2008 with the new classes of the changed areas in 2000, 2004, 2012 and 2016. According to the accuracy assessment, the overall accuracy of the land covers of the four periods are all greater than 93%. The accuracy assessment indicates that this hybrid method can produce consistent land cover datasets for a long time period.