Linear Spectral Mixture Analysis Land Cover for Assessment Level Subpixel: A Case Study of Tasikmalaya City Area Based on Landsat Imagery

IGEOS International Geography Seminar 2018

Ridwana, R.^1 ; Danoedoro, P.^2 ; Herumurti, S.^2 ; Himayah, S.^1 ; Ihsan, M.^1 ; Arrasyid, R.^1 ; Urfan, F.^1

Department of Geography Education, Universitas Pendidikan Indonesia, Jl. Dr. Setiabudi 229, Bandung

40154, Indonesia^1

Cartography and Remote Sensing, Geography Faculty, Gadjah Mada University, Yogyakarta

55281, Indonesia^2

关键词: High resolution image;    High resolution imagery;    Impervious surface;    Land cover informations;    Linear combinations;    Linear spectral mixture analysis;    Multispectral classification;    Spatial resolution;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/286/1/012042/pdf
DOI  :  10.1088/1755-1315/286/1/012042

来源: IOP

【 摘 要 】

Land cover in urban areas can be detected through surveys or using high-resolution imagery with a better accuracy. Especially if the need related to land cover information regionally in the period of the nineties that require the availability of data and unavailability of high resolution images. Therefore, images with intermediate spatial resolution are still required. However, the use of medium-resolution images such as Landsat is constrained by the presence of mixed pixels that cause land cover in urban areas to vary. The mixed pixel will be the source of error in the multispectral classification process, so it takes analysis up to the subpixel level. The need for information up to the subpixel level for ground cover detection can be obtained through the Linear Spectral Mixture Analysis method, where one pixel in the Landsat image in this study will be separated into four endmember, ie vegetation, impervious surface, bare soil, and water. These four endmembers are assumed to represent linear combinations of land coverings contained in urban areas in the form of proportions in each pixel. The results show that the endmember can be well separated, whereas the RMS error of 1994th is 0.013 with an accuracy of 94.44%.