United States Department of Transportation
i
Two level approach to safety planning incorporating the Highway Safety Manual (HSM) network screening.
-
2014-04-01
-
[PDF-9.59 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Corporate Publisher:
-
Abstract:Compared to microscopic safety studies, macroscopic-focused research is more efficient at integrating zonal-level features into crash prediction models and identifying hot zones. However, macroscopic screening has accuracy limitations. Thus, this study developed a new integrated screening approach to overcome the above-mentioned shortcomings of current screening techniques and to achieve a balance between efforts towards accuracy and efficiency.
For conducting macro level safety analyses, the research team faced several challenges. First, using current Traffic Analysis Zones (TAZs) as basic geographic units caused a high percentage of boundary crashes. The research team used regionalization to develop a new study unit: Traffic Safety Analysis Zones (TSAZs) systems. Approximately 10% of boundary crashes have been integrated in new zones after the regionalization but more than 60% of crashes still occur on the boundary of TSAZs. Hence, a nested structure was proposed to estimate safety performance models separately for boundary and interior crashes. This nested structure allows different contributing factors for different crash types, so this model can provide more accurate and predictable results than a single model. In addition, a Bayesian Poisson Lognormal Spatial Error Model (BPLSEM) was adopted for the SPF analysis. The BPLSEM contains a spatial error term that control for the spatial autocorrelation of crash data. As for the micro level analysis, the research team developed SPFs based on the major function classes of roads in our study area. The research team still used the Full Bayesian Poisson Lognormal models to predict crash frequency but tried four different variable combinations to identify the best model.
After identifying hot spot areas at the macro- and microscopic levels, the research team integrated these macroscopic and microscopic screening results. However, this integration task was challenging because we needed to (1) combine various SPFs from different scales, areas, and roadway types; (2) determine an appropriate weight for each group; and (3) choose a measurement for our final results.
In order to solve the above mentioned problems, this study then developed a new criterion to identify whether a zone has safety issues at the macro- and/or microscopic levels. All TSAZs were classified into twelve categories that include two scale groups (macro or micro) and four safety levels (hot, normal, cold, or no data). Then, the research team defined weights for different scales and roadway types. At the macroscopic level, TSAZs were ranked by their zonal PSIs (Potential for Safety Improvements); at the microscopic level, the calculation of average PSI was more complicated because each TSAZ had several intersections and segments. Both the intersection and segment PSI ranks were averaged. The PSI is used in the HSM for network screening but it is the first time that is used for zonal screening. TSAZs with top 10% PSIs were categorized as “Hot” zones. Finally, the percentile ranks of the PSIs were used in the integration (instead of the original PSIs) because the units of PSI intersections and PSI segments were different. In summary, this study presents an integrated screening method that can be used to overcome the shortcomings of macro- and micro-level approaches. In particular, our results provide a comprehensive perspective on appropriate safety treatments by balancing the accuracy and efficiency of screening. Also, it is recommended that different strategies for each hot zone classification be developed because each category has distinctive traffic safety risks at each of the different levels.
-
Format:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
