Delineating an exurban development footprint using SPOT imagery and ancillary data

Abstract

Exurban development, characterized by low density residential development, is one of the leading anthropogenic causes of land transformation. A major hindrance to studying this phenomenon is a lack of spatially explicit data. In this paper, we explore a simple method based on NDVI recoding of SPOT 5 imagery (10 m resolution) to delineate exurban built pixels across large spatial extents. The study area is the heterogeneous Peterborough County (Ontario, Canada). While an accuracy assessment of the initial NDVI recoding had a producer’s accuracy of approximately 80%, the user’s accuracy was extremely low (20%) reflecting high commission error. To improve the latter, post-classification structural and contextual processing using readily available data were examined. The structural processing produced slight improvement, but the user’s accuracy was still mostly below 50%. The contextual processing using water, roads, and a dasymetric map also showed only slight improvement. Alternatively, the inclusion of parcel boundary data proved to be the most effective method for exurban mapping in the study area, with the user’s accuracy in most sections of the study area over 65% and the producer’s accuracy approximately 80%. This study highlights how the low density, dispersed nature of the exurban development compounds the normal challenges associated with mapping built cover, making most of the traditional post-classification processing methods ineffective. An exception to this appears to be contextual processing based on parcel size.