Multiple Endmember Spectral Mixture Analysis By Integrating Topographic And Textural Information

Accurate mapping of land surface parameters using spectral mixture analysis has been recognized as a hotspot and difficult issue in the study of remote sensing. Radiances of images are strongly influenced by topographic effects, especially in extremely complex mountainous terrain characterized by scattered and highly heterogeneous landscape. Because of these topographic effects, there are hardly researches reported on study by using the spectral mixture analysis in high heterogeneous landscapes of mountainous areas.The study site is located in southeastern Yunnan province. It covers an area of 2048*2048 pixels of Landsat-8 images. This study evaluated the performance of seven topographic correction methods (i.e., C correction, CosineC correction, Minnaert correction, Sun canopy sensor correction, SCS+C correction, Teillet regression correction, and Terrain illumination correction) based on multi-source DEMs (i.e., local topographic maps, SRTM DEM Filled-Finished A/B and ASTER GDEM data) and Landsat-8 OLI data, finding the best performing topographic correction method. Based on that topographic correction method, this study compared the performance of four scene results (i.e. topographically uncorrected OLI image, SCS+C topographically corrected OLI image, uncorrected OLI image together with topographical variables and uncorrected OLI image together with topographical variables and texture information) after Multiple Endmember Spectral Mixture Analysis.The results show that:(1) These investigated topographic correction methods removed topographic effects of Landsat-8 OLI data to varied degrees. However, the performance of these methods generally depends on the use of DEMs and evaluation strategies. (2) Among these correction methods, the SCS+C correction performed best and was less sensitive to the use of DEMs. The performance of topographic correction based on the free and open-access DEMs was generally better than or comparable to those with the use of the local topographic maps. (3) The performance of topographic correction was greatly improved using the SRTM Filled-Finished B data with a resampling scheme of the average value within a window of 3×3 pixels. (4) The endmembers of PPI and SMACC together with the field data produce the best result in endmember extraction. And the feature of intra-endmembers is less parallel after SCS+C correction when compared with topographically uncorrected one. (5) Uncorrected OLI image together with topographical variables performed best, and the accuracy of results are not improved after add texture information. For uncorrected OLI image together with topographical variables, MAE of rubber and R2 of forest is 0.227 and 0.522, where 0.234 and 0.496 is the best respectively. R2 of cropland were 0.598 and 0.618 before and after add texture information, while in other situations MAE of forest were 0.256 and 0.275. (6) SCS+C topographically corrected OLI image is not suitable for the Multiple Endmember Spectral Mixture Analysis in rugged terrain area.