Spatial Multiresolution Analysis Approach to Identify Crash Hotspots and Estimate Crash Risk

Abstract

In this paper, we evaluate the performance of a spatial multiresolution analysis (SMA) method, that behaves like a variable bandwidth kernel density estimation (KDE) method, for hazardous road segments identification (HRSI) and crash risk estimation(expected number of crashes). The use of spatial analysis for HRSI is well documented in the literature, especially with KDE methods. The proposed SMA, which is based on the Haar wavelet transform, is similar to the KDE method with the additional benefit of allowing the bandwidth to be different at different road segments depending on how homogenous the segments are. Furthermore, the optimal bandwidth at each road segment is determined solely based on the data by minimizing an unbiased estimate of the mean square error for Poisson data called Poisson’s unbiased risk estimate (PURE). We compare SMA with the state-of-the-practice crash analysis method and the empirical Bayes (EB) method, in terms of their HRSI ability and their ability to predict future crashes. The results indicate that SMA may outperform EB, at least with the crash data of the entire Virginia interstate network used in this paper. The SMA is computationally fast, does not require any data other than crash counts and their location, and is implemented in an Excel spreadsheet freely available for download. Therefore, it can be used for quick large-scale network screening before a more complex analysis that complements crash counts with other crash explanatory variables, such as traffic volume, is used for selected areas of interest.