Predicting Parcel-Scale Redevelopment Using Linear and Logistic Regression—the Berkeley Neighborhood Denver, Colorado Case Study

Abstract

Many watershed challenges can be associated with the increased impervious cover that accompanies urban development. This study establishes a methodology of evaluating the spatial and temporal distribution of infill re-development on a parcel scale, using publicly available urban planning data. This was achieved through a combination of linear and logistic regression. First, a “business as usual” linear growth scenario was developed based on available building coverage data. Then, a logistic regression model of historic redevelopment, as a function of various parcel attributes, was used to predict each parcel’s probability of future redevelopment. Finally, the linear growth model forecasts were applied to the parcels with the greatest probability of future redevelopment. Results indicate that building cover change within the study site, from 2004–2014, followed a linear pattern (R2 = 0.98). During this period the total building cover increased by 17%, or 1.7% per year on average. Applying the linear regression model to the 2014 building coverage data resulted in an increase of 820,498 sq. ft. (18.8 acres) in building coverage over a ten-year period, translating to a 14% overall increase in impervious neighborhoods. The parcel and building variables selected for inclusion in the logistic regression model during the model calibration phase were total value, year built, percent difference between current and max building cover, and the current use classifications—rowhome and apartment. The calibrated model was applied to a validation dataset, which predicted redevelopment accuracy at 81%. This method will provide municipalities experiencing infill redevelopment a tool that can be implemented to enhance watershed planning, management, and policy development.