Projecting the Urban Future: Contributions from Remote Sensing

Mapping urban extent accurately with remotely sensed imagery is challenging. In addition to the inherent difficulty of defining urban land use in terms of land cover (LC) at fixed spatial scales, robust identification of LC using remote sensing has its own challenges. Many of these challenges are related to the fact that the physical properties and conditions that influence scattering and emission of radiation vary in time, space, geography, geometry and wavelength. A second set of difficulties arises in the process of discretizing multiple scales of LC that often vary continuously in abundance over the range of spatial scales at which sensors integrate responses into individual pixels. These factors underscore the importance of spatial and spectral resolution for mapping human settlements and urban extent. The evolution of urban remote sensing is discussed in the context of changes in spatial and spectral resolution and in the tools used to classify land cover. A multi-sensor approach to urban extent mapping acknowledges the limitations of individual sensors and uses the intersection of multiple physical criteria to distinguish the built environment from other types of land cover. A simple example of multi-sensor mapping combines decameter-scale reflectance, derived from Landsat imagery, with kilometer-scale emission of night light, derived from DMSP-OLS composites. Current availability of intercalibrated, accurately co-registered Landsat imagery extending back to the 1970s allows for retrospective analyses of a variety of human modified landscapes—including urban. Combined with continuous field representation of land cover at subpixel scales, many of the limitations of single image discrete classification can be avoided. Global availability of Landsat allows for comparative analyses of human settlements in different environments and cultures. To facilitate comparison in terms of physical properties of land cover, we use a standardized spectral mixture model to convert the Landsat surface reflectance spectra to areal abundances of Substrate, Vegetation and Dark fractions. Because many land uses result in time-varying land cover properties we characterize the Landsat response across multiple seasons within a single year in terms of the seasonal mean and variability of each spectral endmember. The results of the multi-sensor characterization quantify both consistencies and inconsistencies in the combined response of modified landscapes in a variety of environments worldwide.