Land Cover Mapping in an Urban Setting: the integration of Ikonos imagery, image segmentation, and decision trees.

Tom Miewald, Canaan Valley Institute, tom.miewald@canaanvi.org

Drannon Buskirk, Paxton Creek Watershed Association, PaxtonCreek@hotmail.com

Janette Bennett, Canaan Valley Institute, janette.bennett@canaanvi.org

Land cover mapping in an urban setting poses several problems due to the high degree of spatial heterogeneity and spectral similarities of various features. The use of high-resolution imagery often tends to compound these problems. In this presentation, image segmentation is used to spatially clump similar pixels into polygons. We then developed a methodology which looks at multiple dimensions to classify features (segmented polygons) in an urban setting. Spectral information is incorporated with shape, texture, and ancillary information derived from remote sensing. Classification and Decision Trees (CART) is used to classify the polygons. This is a robust and repeatable process for urban areas in the Mid-Atlantic Highlands. The advantages of this methodology are: it reduces the amount of speckle in urban area, incorporates measures of shape and texture, CART is non-parametric, and this method can be easily implemented. In an urban setting, this method has been proven useful for separating critical urban features such roads, rooftops, parking lots, and residential zones.