Abstract for presentation at The 13th Australasian Remote Sensing and Photogrammetry Conference

Accurate maps of urban green space and impervious surfaces are important for planning and monitoring urban development. Detailed thematic maps can be derived from high resolution satellite imagery such as Ikonos or Quickbird, but we aimed to map urban state on dates prior to availability of these image sources. Thus we used medium resolution satellite imagery, with a sub-pixel mapping technique, in a project with Christchurch City Council to map Christchurch City (New Zealand).
Initial work with summer satellite imagery (Landsat, February 2000) showed that, in a drought-prone city, grass is dry for much of the summer unless irrigated, and this dead vegetation was not highly separable from impervious surfaces. This reduced the accuracy of our impervious surfaces quantification, as assessed against high resolution aerial photography. We then moved to use of spring Landsat imagery (October 2001) when most vegetation is green and separation from non-vegetative surfaces is much stronger.
Using spectral unmixing, we assessed percent cover of green vegetation vs. impervious surfaces within each pixel. We will present maps of the full city area, showing variations of impervious surface areas between suburbs, industrial areas, and the CBD. Percent impervious surfaces ranged from 47% – 55% for urban suburbs, 85% for the CBD, 71% for an industrial area, and 18% – 31% for residential areas on the urban fringe. Invalid areas (e.g., water, shadow, dead vegetation, gravel) were masked out.
For accuracy assessment, ten 750 x 500m tiles of aerial photography were used, representing a range of urban land cover types. On a whole-tile basis, the average absolute error between the aerial photography and satellite image assessment of impervious cover was just 4%. Of the 10 tiles, 9 had an error of 6% or less. This is well within accuracy levels required for operational use.