Urban Composition and Surface Temperature at Multiple Scales Using Airborne Spectroscopic and Thermal Imaging

Urban composition maps are key inputs for urban climate models. Spectroscopic imagery, which captures the entire reflected visible, near-infrared, and short wave infrared (VSWIR) range, could improve urban composition maps by discriminating between spectrally similar urban surface materials, such as trees and turfgrass. Several orbital spectroscopic imagers are planned for the next decade, however the imagery produced by these sensors have coarser spatial resolution than is typically used for urban applications. In the summer of 2014, NASA collected airborne spectroscopic VSWIR imagery over Santa Barbara, CA, at 4 m and 18 m resolution, as well as land surface temperatures using the MODIS/ASTER Airborne Simulator (MASTER). We used this unique dataset to explore the effect of spatial resolution on estimates of sub-pixel urban materials. We extracted representative spectra of common urban materials from both scenes, and then used Multiple Endmember Spectral Mixture Analysis to estimate sub-pixel fractions of turfgrass, tree, senesced vegetation, soil, paved surfaces, and roofs at both 4 m and 18 m resolutions. We performed an accuracy assessment of our fractional cover estimates using manually delineated maps of urban surface materials in 66 validation areas randomly distributed across the study domain.

Sub-pixel fractional estimates were highly correlated with the validation dataset at both 4 m and 18 m resolutions. The mean correlations across all 6 cover classes were r² = 0.82 at 4 m and r² = 0.75 at 18 m, indicating that imaging spectroscopy was robust enough to conserve fractional estimates across scales, which could be of significant value for urban analysis. As expected, the mean pixel complexity (the number of sub-pixel materials) was higher at 18 m (1.88) than at 4 m (1.65), however 48.0% of all 4 m pixels were estimated to contain sub-pixel mixtures, indicating the value of spectroscopic imagery for producing sub-pixel estimates even at fine spatial scales in cities. The resulting maps of fractional urban composition were then used to examine the relationships between sub-pixel fractions and land surface temperature, with surface temperatures varying significantly along fractional cover gradients. Overall, these results show imaging spectroscopy can provide robust estimates of urban composition fractions across spatial scales. This suggests that the planned satellite acquisition of spectroscopic imagery could provide significant improvements in urban composition mapping that would enable truly global comparative analyses of urban energy and water balances.