Landsat urban mapping based on a combined spectral-spatial methodology

Abstract

As part of a program to monitor the evolution of urban growth and to encapsulate aspects of urban sustainability of major Canadian cities, we have developed a new methodology for improved urban delineation. This approach involves the generation of two independent land cover products based on pixel-based and segment-based classifications. These classifications are merged through a rule-based approach to exploit the fundamental advantages of each product. It is shown that a relatively simple rule set can be used both to infer new land cover and land use classes and to simplify the complex land cover class mix that is characteristic of low density residential areas. This paper presents an overview of the methodology and quantitative assessment of its performance. A study has also been conducted to quantify a building density detectability threshold for TM data. Example results are major Canadian cities including Ottawa, Calgary and urban centres in southwestern Ontario that have recently experienced rapid growth.