Urban land use classification and modelling using cover-type frequencies

Abstract

A methodology for extracting land use categories from a classification of heterogeneous urban cover types was developed using conventional cluster analysis techniques and discriminant analysis classification functions. Covertype frequencies for a convolved 15 × 15 window were extracted from densitometric data obtained by scanning a NASA high-altitude colour infrared transparency acquired over Champaign-Urbana, Illinois. The initial cover-type frequency classification produced a large number of boundary effects as the convolved window overlapped different areas of land cover patterns. These boundary effects were reduced by relabelling boundary classes that were statistically closest to selected core classes. Homogeneous land use classes such as old, middle-age and new residential areas were identified in the relabelled classification, but considerable confusion still existed between non-infrared reflecting agriculture fields and the urban central business district. Cover-type frequencies obtained for the enumeration tracts of the 1986 census for Edmonton, Alberta were used to examine the relationships between Landsat Thematic Mapper data and housing characteristics (age, value, occurrence). High correlations were found for both the average age and average value of dwellings versus relative (percentage) counts of the cover-type frequencies. A strong relationship was also found between the number of single-detached dwellings and the absolute (raw count) cover-type frequencies. A single cover type in all three models accounted for approximately half of the total variation.