Using HyspIRI Campaign Data for Sub-pixel Classification of the Urban Land Surface

Abstract:

Quantifying urban ecosystem characteristics and dynamics requires accurate characterization of urban surfaces, which is complicated by the spectral similarity and surface heterogeneity of urban materials. In the late summer of 2014, visible near infrared/shortwave infrared imagery was acquired over the Santa Barbara, CA, urban area by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) (18 m resolution, 224 bands, 0.35 - 2.5 µm) and AVIRIS-Next Generation (4 m resolution, 432 bands, 0.35 - 2.5 µm). We developed spectral libraries of urban image endmembers at both 4 m and 18 m, and then applied them to the 18 m imagery using Multiple Endmember Spectral Mixture Analysis (MESMA) to produce a classified map and sub-pixel fractions of impervious ground surfaces, roofs, irrigated grass, trees, senesced vegetation, and soil. Final products were validated using 1m resolution aerial imagery.

Sub-pixel fraction results show that broad, ecologically important classes were assessed at a relatively high degree of accuracy. These classes include modeled green vegetation and impervious surface fractions (r² > 0.85) and a combined senesced vegetation and soil (r² > 0.70). Sub-classes of vegetation (trees, irrigated grass, and senesced vegetation) were also detectable, indicating the possibility for refined inputs for future urban ecological studies. MESMA results showed that the majority (> 85%) of the 18 m scale image was mapped with one or two endmembers (plus shade). Furthermore, endmembers extracted from higher spatial resolution imagery (4 m) did not yield a detectable increase in sub-pixel accuracy. Overall, these results indicate that the 18 m spatial resolution imagery was adequate to capture these broad urban classes at the sub-pixel level.