Book Description

This dissertation consists of five published or presented papers in which addresses different gaps in the knowledge by presenting innovative methods to identify and analyze weather-related naturalistic driving data to better understand driver behavior and performance in adverse weather conditions. An innovative methodology introduced in Chapter 4 helped to effectively identify weather-related trips in real-time using vehicle wiper status and other complementary methodologies introduced in chapter 5 helped to identify naturalistic driving weather-related trips using external weather data sources. In addition, a semi-automated data reduction procedure was developed and introduced in chapter 5 to process raw trip data files into a format that further analyses and modeling techniques could be easily applied. The novel approaches developed in this dissertation for NDS trip acquisition and reduction could be extended to other naturalistic driving studies worldwide. In addition to the contributions in data extraction and reduction, preliminary analysis as well as advanced modeling techniques were utilized in this study. These analyses were used to explain the relationship between different levels of speed selection/lane keeping behaviors and a set of contributing factors including roadway characteristics, environmental and traffic conditions and driver demographics on a trajectory level. These modeling techniques ranged from common parametric approaches such as binary logistic regression and ordinal logistic/probit regression models to a more advanced non-parametric/data mining modeling techniques such as Classification and Regression Trees (CART) and Multivariate Adaptive Regression Splines (MARS). The results from this study suggest that both parametric and non-parametric modeling approaches are important to analyze driver behavior and performance. In fact, this study attempted to maximize the benefits out of the advantages of parametric models, such as the ability of interpreting the marginal effects of various risk factors, as well as the advantages of using non-parametric models, including but not limited to the ability of providing high prediction accuracy, handling of missing values automatically, and their capability of handling large number of explanatory variables in a timely manner, which might be extremely beneficial specifically for assessing traffic operations and safety in real-time considering weather and traffic data to be directly fed into the model. The results of the developed speed selection models revealed that among various adverse weather conditions, drivers were more likely to reduce their speed in snowy weather conditions compared to other adverse weather conditions. Specifically, the odds of drivers reducing their speed were 9.29 times higher in snowy weather conditions, followed by rain and fog with 1.55 and 1.29 times, respectively (compared to clear conditions). In addition, variable importance analysis using CART method revealed that weather conditions, traffic conditions, and posted speed limit are the three most important variables affecting driver speed selection behavior. In addition, the results of the developed lane-keeping models revealed that drivers in heavy rain conditions were 3.95 times more likely to have a worse lane-keeping performance compared to clear weather conditions. The developed speed selection model is a key example of a derived mechanism by which the SHRP2 database can be leveraged to improve Weather Responsive Traffic Management (WRTM) strategies directly. Moreover, the results may shed some light on driver lane keeping behavior at a trajectory level. Moreover, a better understanding of driver lane-keeping behavior might help in developing better Lane Departure Warning (LDW) systems. Evaluating driver behavior and performance under the influence of reduced visibility due to adverse weather conditions is extremely important to develop safe driving strategies, including Variable Speed Limits (VSL). Many roadways across the U.S. currently have weather-based VSL systems to ensure safe driving environments during adverse weather. Current VSL systems mainly collect traffic information from external sources, including inductive loop detector, overhead radars and Closed Circuit Television (CCTV). However, human factors especially driver behavior and performance such as selection of speed and acceleration/deceleration behaviors during adverse weather are neglected due to the lack of appropriate driver data. The findings from this study indicated that the SHRP2NDS data could be effectively utilized to identify trips in adverse weather conditions and to assess the impacts of adverse weather on driver behavior and performance. With the evolution of connected vehicles, Machine Vision and other real-time weather social crowd sources such as WeatherCloud®, more accurate real-time data similar to the NDS data will be available in the near future. This study provided early insights into using similar data collected from NDS.