Variation in spectral shape of urban materials

Abstract

Airborne Hyperspectral Scanner System (HSS) images of 2.7 m spatial resolution and laboratory/field spectroradiometric data were used for reflectance characterization of synthetic and mineral building materials in southeastern Brazil. An impervious surface image was generated from a five-endmember mixture model with the 37 reflective HSS bands to select sampling sites for creation of a spectral library. Principal component analysis was applied to this library and to the pixels identified as impervious surfaces in the HSS image to identify patterns of spectral response and potential discrimination of urban materials. For both aerial and laboratory/field data sets, PC1 showed high within-class brightness variation due to surface aging, degradation and painting as well as illumination-viewing effects. PC2 indicated an important spectral shape contrast between the near-infrared and the shortwave infrared, which has the potential for discriminating urban materials. When tested with the spectral angle mapper, which reduced the brightness influence on classification and allowed spectral shape comparison, near-infrared and shortwave infrared bands differentiated most of the impervious surfaces studied. Synthetic materials with spectra that declined in magnitude at longer wavelengths (e.g. polyvinyl chloride and painted surfaces) were discriminated from mineral materials with spectra that were flat (e.g. asphalt) or ascending (red tile). Despite having brightness that is distinct from steel and red tile roofs, aluminium was misclassified by spectral angle mapper into the two classes with ascending slope curves. The overall classification accuracy was 74%.