The Washington State Department of Transportation (WSDOT) commissioned this research, conducted by the Urban Ecology Research Laboratory (UERL) at the University of Washington, to assist in effectively designing and managing operational, maintenance, and improvement activities within the context of the many growth management and clean water regulations and ordinances in Washington State. The goals of this study were to 1) implement a classification scheme for mapping the percentage of total impervious surfaces due to different types of transportation infrastructure based on satellite imagery, 2) develop and assess a remote sensing methodology for detection of road impervious surface area (RISA) and the fraction of RISA compared to the total impervious surface area (TISA) and 3) make recommendations on the imagery best suited for identifying impervious surfaces related to transportation infrastructure.
The results of this analysis have important implications regarding the use of remote sensing to determine the contribution of impervious surface from transportation infrastructure at regional scales. Higher resolution satellites, while more visually appealing, do not necessarily provide a net benefit in terms of accuracy that may justify their added expense. Our results indicate that, in most cases, Landsat performed as well if not better than the higher resolution SPOT imagery for determining regional scale roadway impervious surface area. The problem with using high resolution data for extracting road footprints at regional scales lies in the difficulty and cost of gathering a comprehensive set of imagery for the entire area of interest. Furthermore, extracting road footprints from high resolution imagery is a difficult proposition.
Our findings recommend using digital imagery with other GIS data that can serve as a proxy for road footprints. Transportation rights-of-ways taken from vector parcel data were highly effective at limiting the area that could be considered as road. Using this in combination with Landsat impervious surface data proved to be an accurate and relatively simple way to estimate road impervious surface area. We recognize that not all areas are covered by the detailed parcel datasets used in this analysis. To fill these gaps, a simple predictive road impervious surface area model was developed using a combination of data developed and gathered for this project. Linear regression was used to build the model and road impervious surface area extracted from test sites was used as the independent, or predicted, variable. The predictors, or independent variables, used in the model were total impervious surface (as measured by Landsat or SPOT), urban area background, and total road length measured using readily available GIS transportation data. All three independent variables were significant with a 95% confidence interval and the model as a whole was significant at the 99% level.