Accuracy Assessment of Land Cover Maps of Forests within an Urban and Rural Environment

Abstract

Land cover maps of forests within an urban and rural environment derived from high spatial resolution multispectral data (QuickBird) and medium spatial resolution multispectral data (Landsat ETM+ and SPOT 4) were compared to ascertain whether increased spatial resolution increases map accuracy of forests and whether map accuracy varies across land cover classification schemes. It is commonly assumed that increased spatial resolution would probably increase land cover map accuracy regardless of land cover classification methodology. This study assessed whether that assumption is correct within a rural and an urban environment. Map accuracy for modified National Land Cover Data (NLCD) 2001 Level II, Level I, and Unique (a modified NLCD 2001 Level II and Level I combination) shows that 30-m Landsat ETM+ data had the highest overall map accuracy for rural, urban, and combined rural/urban land cover maps. Analysis of user's and producer's accuracies shows that Landsat ETM+ data had higher levels of producer's accuracy of >90.0% for the coniferous cover type for modified NLCD 2001 Level II and Unique, excluding one instance for which SPOT 4 had a user's accuracy of 98.5% for the rural coniferous cover type. Modified NLCD 2001 Level I Landsat ETM+ data had user's and producer's accuracies for a homogeneous forest cover type of 98.4 and 90.6%, respectively. Landsat ETM+ data also outperformed SPOT 4 and QuickBird within an urban environment, creating the only map products with forest cover type user's and producer's accuracies of >90.0%.