Remote Sensing Classification of Brownfields in the Phoenix Metropolitan Area

Brownfields, idle or under-utilized urban areas, were identified in three parts of the Phoenix metropolitan area using a supervised classification technique using ERDAS Imagine. The brownfields were first evaluated in an area of south Phoenix used as a control site to establish sample training sites. These sites provided the basis for the supervised classification of an ASTER satellite image, which covered 60 x 60 km of the Phoenix metro area. Thirteen other common urban features were also identified to isolate these spectral signatures from that of brownfields. Using a set of 14 training sites, discrete brownfields were identified throughout the image. The technique was tested using a Kappa coefficient (κ) accuracy assessment and ground-truth observations within the control and two “blind” areas in south Glendale and north Phoenix. The Glendale site, an older urban area, yielded an overall κ accuracy of 64% with a ground-check accuracy of 60%, whereas the north Phoenix site, an urban fringe area, yielded an overall κ accuracy of 77% with a ground-check accuracy of 44%. Small sample sizes and urban reclamation of brownfields between imaging time and site assessment can explain some degree of disparity of the results for the north Phoenix area. A refinement of urban training sites and limiting brownfield search areas to non-fringe areas would improve this technique in future studies.