One of the first applications of satellite remote sensing imagery was to detect the size and shape of urbanized areas. Under many circumstances the mere identification of what is really urban is not clear. Smaller towns, diffuse development and varying degrees of spatial connectivity combined with the lack of an agreed upon definition of urban complicate the task. The proximity of urban development to spectrally similar fallow agricultural areas is a primary challenge of mapping urban development with satellite imagery.
As sensors on satellites became more sophisticated and technically advanced, urban applications of remote sensing used higher spatial resolution imagery. However, moderate spatial resolution is widely used on non-urban applications because the synoptic spatial and retrospective temporal coverage is superior that offered by high resolution sensors. The most widely available imagery over the longest time period is provided by the Landsat missions. However, the 30 m spatial resolution of Landsat imagery combined with the spectral heterogeneity of urban land cover results in most urban areas being imaged as spectrally mixed pixels. Spectral mixture models may provide a physically based solution to the urban spectral heterogeneity because it is possible to reduce the dimensionality of the multispectral reflectance by converting it to areal fractions of land cover components, thus making interpretation easier. The spectral ambiguity of urban land cover is unavoidable but the challenge of mapping urban extent may be mitigated by using multiple sensors to image different characteristics of the urban environment.
The present analysis was based on a three component linear mixture model incorporating substrate, vegetation and dark targets, directly used for visualization on false colour composites of red, green & blue respectively. Fraction composites suggest the location and extent of urban development - both at the periphery and within Sao Paulo's urban agglomeration and its surroundings - but the spectral ambiguities with non-urban land cover remain a challenge.
Winter and summer image pairs were selected for quality and consistency of solar illumination for two time intervals: 1986-2005 and 2000-2010. Changes over both time periods were quantified in terms of changes in endmember fractions. The results show increases of substrate, greater than 10% of the pixel area, with equivalent reduction of vegetation and/or shadow fractions. From these increases in substrate fraction, together with the presence of night lights of higher intensity and concentration, we infer an increase in urbanized land cover. A quantitative and visual analysis of these changes at different spatial scales is presented.
Wide field synoptic imagery provided by the Defence Meteorological Satellite Program Operational Line Scanner (DMSP-OLS) indicates the presence of urbanized areas by imaging nocturnal lights. This sensor has been used by the Earth Observation group at NOAA NGDC to produce annual global composites of temporally stable nighttime lights since 1992. OLS night light imagery helps to differentiate non-urban substrate (i.e. exposed soil) from urbanized surfaces with different degrees of development, according to the intensity of the emitted light.
EARSeL European Association of Remote Sensing Laboratories, Strasbourg, France