An Object-oriented Classification of an Arid Urban Forest

In order to adequately model ecosystems services of the urban environment, it is necessary to accurately inventory urban vegetation abundance and spatial distribution. An object-based, methodological design for estimating urban vegetation structure within the metropolis of Phoenix, Arizona using remote sensing techniques on high-resolution (0.6m) aerial photography was derived utilizing a hybrid of image segmentation and spectral classification. Within the arid urban environment, vegetation is controlled at a very fine scale and is best analyzed at extremely high spatial resolution. Since at a local scale the urban environment is composed of discretely heterogeneous patches, it is necessary to quantify landscape pattern at this scale. An object-oriented approach was taken utilizing a segmentation algorithm that transposes imagery into distinct polygons by incorporating a combination of spectral properties and neighborhood characteristics. This segmentation process was parameterized to isolate vegetation patches with a minimum diameter of 2m, which incorporates typologies from shrubs to large trees. Once segmentation was completed, the image was then analyzed and classified based on the cadastral and topological characteristics. Accuracy assessment of land cover classification was then conducted at 200 random points throughout the metropolis. The intent of this methodology is to allow for regular monitoring of vegetation change at a broad extent with fine resolution in the Phoenix basin.