A spatial autocorrelation based approach for identifying hot spots on urban road

Abstract:

An approach based on spatial autocorrelation analysis was presented to identify hot spots on urban road. The application of this methodology was illustrated by five-year’s traffic crash data. Firstly, a non-parametric kernel density estimate sketched a model of two dimensional planar accident point spatial distribution. An estimation of the optimal bandwidth for accident distribution density was 115.1m and subsequently selected to create a spatial statistical unit on road segment. Then, the spatial statistical units derived from a GIS-based region-arc topology model representation of the urban road network, and aggregated attribute values of crash counts and severity indices. Finally, the Globe Moran index was employed to examine spatial distribution clustered pattern of crash data. The Local G-statistic was used to identify the clustering of low and high index values and to generate a crash hot spots map. Results of the research indicate that compared with the negative binomial model in modeling the crash frequency, the spatial analysis under integration of accident attribute data with position can not only identify hot spots on road segments (intersections) but also rank hot spots with spatial correlation according to the attribute values. Traffic management department can use spatial analysis visualized results to locate hot spots and develop further traffic safety decision-making research.