MULTIPLE ENDMEMBER SPECTRAL MIXTURE ANALYSIS OF DESIS IMAGE TO IDENTIFY ROOFTOPS IN KIGALI

Abstract. The development and increase of multi and hyperspectral sensors in the recent years have significantly improved urban structure analysis and interpretation. The current study is the first to investigate the potential of DESIS hyperspectral images for the detection or identification of urban roof materials. After field campaigns in 2014, 2015 and 2018 to collect ground truth points and rooftops radiometric properties; a linear spectral mixture, implemented using a non-negative least squares (NNLS) regression based on the sequential coordinate-wise algorithm (SCA) was applied on a DESIS image from 2020 of Kigali city to identify the different rooftops material and color. Although results show that most endmembers were predicted with a very low probability, the study proved that the combination of spectral mixture and hyperspectral data such as DESIS have great potential in the detection of rootops material. The presented study also highlghted a number of challenges resulting from the choice of spectral mixture algorithm and colinearity between materials.