Screening Urban Road Network for Corridors with Promise
StatisticsView Usage Statistics
Both federal and state policy makers increasingly emphasize the need to reduce traffic fatalities and serious injuries. Finding improved methods to enhance roadway safety has become a top priority. In an attempt to reduce traffic crashes, crash-prone locations should be identified for increased law enforcement activities, education programs, and engineering improvements. This dissertation addresses the critical issue in traffic safety research, methods for corridor-based screening for safety improvement. The role of corridor level screening is to periodically examine the entire urban roadway network in order to generate a list of corridors ranked in order of priority by which detailed engineering studies should be conducted. Ongoing debates in regards to corridor level network screening include what should constitute a corridor for the purpose of network screening, and how a local agency should perform a corridor screening. This research provides answers to these questions.Firstly, a comprehensive literature review was conducted to summarize current practices across the nation pertaining to corridor level network screening. No consensus was found in terms of corridor definitions or screening methodologies. Observed traffic crashes are generally used in evaluating the safety of urban facilities in state departments of transportation although model based evaluation is highly recommended because of its exceptional merits. Secondly, this research proposes a spatial clustering based approach to define urban corridor boundaries. The idea is to group signalized intersections along urban arterial roads based on their spatial auto-correlations. Local spatial measurements (local Moran's I and Getis-Ord index) are adopted to cluster multivariate intersection data. The analyses of arterials in the Reno-Sparks area indicate that the proposed approach provides reasonable corridor definitions. The next section of the dissertation proposes a model-based scheme for screening urban corridors. Significant efforts were made to collect crash, traffic and road inventory data at intersection level, segment level and corridor level. Data assembling and processing were conducted in ArcGIS. Statistical models, such as negative binomial regression models, have been widely used in developing crash prediction models over the past decades. This research investigated other models including the Poisson-Inverse Gaussian and Poisson-Lognormal models. Analysis results imply that for a certain data set different model assumptions will generate quite different results. Overall, the Poisson-Inverse Gaussian model and the Poisson-Lognormal model perform better than the negative binomial model in terms of goodness-of-fit statistics. Due to the high flexibility of Inverse Gaussian and Lognormal distributions, such models can be adopted as alternatives to the negative binomial models in developing crash prediction models. Furthermore, this research explores the effect of spatial correlations in crash prediction modeling. Significant spatial correlations were found not only within different intersection data sets and segment data set, but also in the model residuals and fitted values. The spatial eigenvectors are introduced into the developed models to supplement the spatial effects. Different neighboring proximity structures are tested for assembled data sets to establish the configuration that results in the optimal performance of the prediction models. The comparisons of model goodness-of-fit statistics indicate that the spatial correlations contribute significantly to model heterogeneity. Ignoring spatial impacts may result in biased estimates of model parameters and incorrect inferences. Combining safety performance of intersections and segments, a Corridor Safety Measurement (CSM) is proposed as the performance measure for corridor screening. The measurement is used to identify corridors in the Reno-Sparks area that have promise as locations where improvements will result in substantial crash reduction. The findings from this research will assist engineers to proactively identify and analyze high crash locations from a corridor perspective and detect potential problematic locations not identified through the traditional hot spot analysis.