Sub-pixel Estimation of Urban Land Cover Using QuickBird and Landsat Imagery

Canada Centre for Remote Sensing
Earth Science Sector, Natural Resources Canada
588 Booth Street, Ottawa, ON, K1A 0Y7

Abstract

The spatial distribution of man-made surface and treed cover in metropolitan areas are key land cover factors influencing not only both urban environment, but also the urban micro-climate. Sub-pixel mapping of the two land cover components based on Quickbird and Landsat data is currently underway for Canadian urban areas. This involves estimation of the percent contribution of these land cover components within each 30-metre pixel of the CUrLUS (Canadian Land Use Survey). High resolution multi-spectral images from the Quickbird satellite plays a key role in this process. First, a thematic classification is derived from these data that includes the following land categories; man-made surface, trees, grass, dry grass and disturbed land (i.e. land in the process of conversion from rural to urban use). Quickbird land cover classifications are then used to ‘calibrate’ a variety of Landsat image transforms. The a priori 30-metre land cover land use classification of CUrLUS allows us to derive unique calibration curves for each CUrLUS urban class, thereby minimizing problems such as confusion between true impervious surfaces and other non-vegetated surfaces such as fallow fields and ‘greenness’ ambiguities between forest and herbaceous land.

The results are presented for the Greater Toronto Area and Ottawa-Gatineau area. Multiple Quickbird scenes have been acquired for the areas thereby allowing us to undertake an in-depth study of absolute and relative accuracies. Because of cost and availability considerations, ‘signature extension’ techniques are being developed to assess the feasibility of inferring sub-pixel land cover over areas where high-resolution data is not available. This process involves a combination of the calibration procedure described above in conjunction with rigorous inter-scene radiometric balancing of multiple Landsat scenes.