Spectral Unmixing vegetation cover of a desert ecotone with Landsat 8 OLI data using MESMA

Keywords: spectral Unmixing analysis, endmember, desert vegetation, vertex component analysis
Abstract Type: Paper Abstract

Abstract

Vegetation cover is essential to the ecologic and biogeochemical dynamics of drylands. These systems are marked by spatiotemporal variations in vegetation cover and biomass. Regular monitoring of vegetation cover in drylands is critical for their conservation and management. Spectral unmixing of spaceborne remotely sensed data have provided a promising tool to estimate sparse vegetation in dryland environment. In this study, we explored multiple endmembers spectral mixture analysis (MESMA) applied to Landsat 8 OLI data using three different sources of endmembers: spectral library, selection form image scene, and spatial interpolation based on field survey. The study area on north hillslope of Tian Shan in Manas river watershed, Xinjiang, China, includes three plant communities: desert steppe, dry steppe, and meadow steppe. A photo-based grid point intercept sampling method were used to quantifying vegetation cover over this desert ecotone. The quantification transects plots provided reference data for retrieving endmembers and assessing unmixing models. Two different spectral libraries: USGS spectral library and field spectra measured using ASD FieldSpec, were tested for unmixing respectively. Vertex component analysis (VCA) algorithm was used to retrieve endmember automatedly from image scene. A least square regression was applied to interpolate vegetation endmember from image scene based on reference data collected from the field survey. The assessment with reference data indicated that MESMA with endmember interpolated from scene achieved the highest accuracy for unmixing of vegetation cover among the different sources of endmembers.