CA / Freeway Incidents

California has operated High Occupancy Vehicle (HOV) Lane to encourage carpools and to avoid traffic congestion. However, we sometimes feel even HOV lanes are also jammed at peak hours (Degradation). According to a result of the research in this project, we found that HOV lanes in California continue to be congested, with as many as 62% of HOV lane-miles meeting the Federal standards for degradation. The purpose of this project was to identify factors behind degradation and to recommend strategies that mitigate degradation.

I was in charge of data collection/preprocessing and a topic in the project, determining what causal factors, if any, contributed to traffic incidents on HOV facilities. Traffic incidents are events on roads such as car incidents and vehicle breakdowns, traffic hazards, etc. which lead to severe congestion and pose a safety threat to other vehicles on the roadway. Therefore, it is important to understand the relationship between the occurrence of traffic incidents and factors such as geometric features of the physical facility and traffic performance.

California government provides quality traffic data through various resources. I obtained datasets of road shape (Caltrans GIS library), incidents (CHP incidents in PeMS data clearinghouse), and traffic performance (PeMS) from publicly available resources. We then processed the data using satellite imagery from Google Earth as needed (This was the harshest part). Given that each of the data contains locational information, I could merge the different dataset and build a single dataset for the analysis based on their spatial relation.

Due to recording errors inherent in spatial data, places, where frequent incidents occur, does not always imply that incidents occur repeatedly on exactly the same location. It is instructive to recast incident data, which is point data, into a probabilistic measure that is a continuous function over the length of road. There are various kernel functions that allow this transformation. We base our kernel function to reflect the assumptions in Atsuyuki Okabe, Toshiaki Satoh & Kokichi Sugihara (2009), a well-known study that introduces a network-based kernel density estimation method.

As a result of the network-based Kernel density estimation, I produced interactive maps that visualize the density map of traffic incidents on both HOV lanes and Mainline lanes (ML). You can access the maps through the links below:

 (Map control)

  in Desktop environment)  Orbit - Left drag // Zoom - Scroll (or middle botton) // Pan - Right drag
  in Mobile environment)  Orbit - one finger // Zoom - two fingers // Pan - three fingers

 (Legend)

  Blue - Incidents on ML
  Red - Incidents on HOV

  Link - Southern California Traffic Incident Kernel Density

  Link - Northern California Traffic Incident Kernel Density

In the regression work, I explored the relationship between traffic incidents on HOV lanes and structural attributes of the HOV facility. The result shows that incidents on HOV lanes occur more often when the road width and outer shoulder width are larger. we could see more incidents when HOV and ML are separated by barrier than other access types like continuous access or limited access separated by a buffer. We also found the tendency that the more incidents on ML, the more incident on HOV. On the other hand, distance to closest on-ramp behind and off-ramp ahead has a negative association to the frequency of incidents.