Past experience with crash prediction modeling has confirmed the importance of traffic volume not only as exposure, but also as a predictive variable. For intersection-related collisions, for example, angle collisions or any involving turning vehicles, traffic volumes on both intersecting roadsare necessary for sufficient prediction of accident count. These collisions occur not only at intersections but any place where vehicles turn on oroff the roadway, such as driveways. Intersecting traffic volumes at such locations are either not available or very labor intensive to acquire. Theobjective of this study was to investigate the use of Geographic Information System (GIS) land use inventories to estimate exposure measures for estimating models for predicting segment-intersection crashes, defined as collisions occurring on road segments involving one or more turning or crossing vehicles. Model results for rural two-lane and urban two- and four-lane undivided roads show that either the number of trips generated or the extent of surrounding land development itself act as excellent predictors for segment-intersection crashes, and in fact work better than models using the number of access points. This is likely because these variables betterdescribe the intensity of the traffic accessing the major artery. This isa very valuable finding, as counting access points along a road segment cannot be done automatically, but many jurisdictions have GIS land use inventories available for all sorts of planning purposes. This development will permit better accounting of exposure to segment-intersection crashes in crash prediction modeling.
Samenvatting