Modeling residential developed land in rural areas: A size-restricted approach using parcel data

In most land cover datasets, the classification of developed land is less accurate in rural areas than in urban areas, due to difficulties in identifying rural developed areas from remote sensing data. This inconsistency makes land cover data less reliable in rural settings, when employed for small area population estimation or for exploring processes such as urbanization. This research addresses this challenge, identifying rural developed land using ancillary variables such as terrain, road density and distance to roads. Predictive models are developed using residential parcel units as a spatial outcome variable. Although parcels are often the most spatially precise indicators of developed land, rural parcels can be very large, leading to high levels of heterogeneity within a parcel. To assess the effect of size on the relationships between the ancillary variables and the locations of rural residential land, parcels are categorized on size and size-restricted statistical models are run. Goodness-of-fit measures and the predictive power of the model improve with decreasing parcel size. A thorough model evaluation quantifies prediction accuracy and highlights rural residential areas with the highest probability of development. A subsequent validation using building footprints as indicators of actual development provides strong evidence that a size-restricted modeling approach improves the predictive power of the statistical model. This type of modeling framework thus has the potential to improve the accuracy of rural developed land classifications in land cover databases such as the U.S. National Land Cover Database (NLCD). (C) 2013 Elsevier Ltd. All rights reserved.