With the demand for timely, accurate and relevant road weather information increasing across the full range of transportation managers and users, it's important to keep abreast of the many advancements that are being made in this area. This session covers several different projects that explore new developments in road weather data processing, especially geospatial and geostatistical analyses, as well as road weather data uses for improved decision making.
Diving Down the Rabbit Hole: RWIS Data to Develop a SWI
Natalie Villwock-Witte, Western Transportation Institute (WTI)Show Abstract
David Veneziano, Iowa State University
Karalyn Clouser, Western Transportation Institute (WTI)
Laura Fay, Western Transportation Institute (WTI)
Accurately assessing winter maintenance operations is a challenge faced by various transportation agencies, including state departments of transportation (DOTs). One tool that has been successfully used by state DOTs for this purpose is the severe weather index (SWI), which assesses the performance and related costs of winter maintenance operations by considering the relative severity of each weather event. Developing an effective SWI encompasses a modeling process that requires the collection and preparation of quality data. This paper draws from research to develop a SWI for the Maryland Department of Transportation State Highway Administration (MDOT SHA). It focuses on the challenges of preparing data for use in a SWI. The study identified numerous lessons learned and recommendations for data preparation, including: Ensure historical data retention; Ensure long-term data usability through creating and maintaining a data dictionary; Consistent use of road weather information systems (RWISs) site names across the agency and by vendors contracted by the agency (even if the RWIS site names change); Define a consistent rounding scheme for data elements to ensure accuracy; Use consistent units, even if the vendor changes (e.g., miles per hour vs. knots); Keep a record of any RWIS sensor changes that occur over time; Continuously collect data (e.g., do not use a 1-minute data capture to represent 5-minutes); Ground truth and perform quality assurance/quality control of RWIS data; and Ensure a link is established between Maintenance Action Reports (e.g. Emergency Operations & Road Conditions) and RWIS data to allow for ground-truthing and quality assurance.
Spatial Mapping of Winter Road Surface Conditions via Hybrid Geostatistical Techniques
Mingjian Wu, University of AlbertaShow Abstract
Tae J. Kwon, University of Alberta
Liping Fu, University of Waterloo
The monitoring and modelling capabilities of road surface temperature (RST) plays a critical role in optimizing winter road maintenance activities. In recent decades, road weather information systems (RWIS) have gained in popularity with many road maintenance authorities. However, RWIS stations are installed at fixed locations, providing only point measurements that are often unrepresentative of distant surrounding areas. To address such limitations, this study employs a hybrid geostatistical interpolation method, namely, regression kriging (RK), to fill in the large spatial gaps that exist at unmonitored locations. RST data collected by an automated vehicle system along selected Iowa interstate highways between Oct. 2018 and Apr. 2019 was used to model the spatial variation patterns of RST via semivariograms, which were then used to interpolate the conditions in between RWIS stations. Cross-validation results indicated that RK successfully captured the spatial variation of RST along the highway segment with the estimation errors being as low as 5.5%. To improve the generalization potential of the proposed model, the nugget-to-sill ratio obtained from a total of 228 semivariograms was utilized to characterize the weather events, and the results implied that stronger winds and heavier rainfalls were likely to form a stronger spatial dependence within RST. The findings of this research contribute to better understanding of the influences of meteorological factors in RST as well as development of improved models for inferring the road surface conditions between RWIS stations under inclement weather events.
Safety Effects of Road Weather Information System (RWIS) - A Cost-Benefit Analysis
Davesh Sharma, University of AlbertaShow Abstract
Mingjian Wu, University of Alberta
Tae J. Kwon, University of Alberta
Road Weather Information System (RWIS) is used by various transportation departments to improve their winter road maintenance services while reducing weather-related collisions during adverse weather events. However, due to its unclear effects on traffic safety and high financial investment, it is essential to quantify its safety effectiveness and assess whether it is financially feasible for statewide implementation. Several studies have looked into the benefit-to-cost ratio (BCR) of the system considering its safety benefits; however, certain biases are associated with the adopted techniques. This study addresses this issue by implementing a before-and-after study using the Empirical Bayes approach for a case study of two RWIS stations in Iowa, U.S. In addition, this study develops safety performance functions and yearly calibration factors using large scale spatial data and a set of reference locations to quantify the sole effect of an RWIS station. To this end, a detailed economic analysis is conducted to quantify the cost-effectiveness of RWIS. Results show that after the implementation of the two RWIS stations, 41.91% and 26.15% of inclement weather collisions have been reduced. The BCR for these stations is 39.97 and 9.83, respectively, indicating RWIS is an economically viable countermeasure and hence the transportation agencies can be more confident while allocating funds for its implementation.
Regional Road Surface Temperature Estimation Using RWIS Dataset: A Comparison Between Ordinary Kriging, Empirical Bayesian Kriging and Regression Kriging Methods
Branislav Dimitrijevic, New Jersey Institute of TechnologyShow Abstract
Sina Darban Khales (email@example.com), New Jersey Institute of Technology
Steven Chien, New Jersey Institute of Technology
Lazar Spasovic, New Jersey Institute of Technology
The Road Weather Information Systems (RWIS) consist of a network of permanent automated weather stations that collect detailed information on local atmospheric and road weather conditions. They are the primary sources of data used by the transportation agencies in their winter road maintenance operations. Besides providing a situational awareness, the data collected at RWIS stations, such as road surface temperature (RST) and road condition, can be used to optimize the winter maintenance operations including timing and the amount of materials to be used for roadway treatment. For example, this data can inform development of more effective preventive (or preemptive) anti-icing strategies, as opposed to deicing operations. However, the RWIS stations are at permanent fixed locations, usually near bridges and other critical infrastructure, and generally do not provide dense coverage of a highway network. In this study, three different geostatistical interpolation models (ordinary kriging, Bayesian kriging, and regression kriging) are developed and compared in estimating the RST for the regional roadway network based on the RWIS data. The models are implemented in a case study of the State of New Jersey. Cross validation and mobile observations were used for model validation. The results of the validation revealed that empirical Bayesian kriging outperforms the ordinary kriging model, but regression kriging exhibits the highest estimation capability among the three investigated models.
Winter Road Surface Conditions Classification using Convolutional Neural Network (CNN): Visible - Light and Thermal Images Fusion
Ce Zhang (firstname.lastname@example.org), McGill UniversityShow Abstract
Ehsan Nateghinia, McGill University
Luis Miranda-Moreno, McGill University
Lijun Sun, McGill University
Winter road conditions play an important role in traffic flow efficiency and road safety. Icy, wet, and slushy road conditions can reduce tire friction, affect driver's line-of-sight and vehicle stability, which can lead to more and dangerous crashes during winter weather events. To keep traffic operations safe during winter, cold north American cities spend several economic sources on winter maintenances (plowing, salting, and sanding). This paper proposes a methodology for the automatic classification of winter road surface conditions using Convolutional Neural Network (CNN) and the combination of thermal and visible light cameras. First, we perform manual annotation to match the overlapping area of both thermal and visible light cameras. Then, the 4,244 matched images are classified into four classes: snowy, icy, wet, and slushy surface conditions. In addition to two separate single-stream CNN models applied to RGB and thermal image sources respectively, a double stream CNN model was also applied to incorporate the information of both sources simultaneously. The results revealed that the classification of wet, slushy, and snowy images relies more on RGB images. However, the detection of icy conditions is more reliable with thermal images. The double stream CNN model reaches better results than any single stream model where its average F1-score on the original dataset is 0.866 for snowy, 0.935 for icy, 0.985 for wet, and 0.888 for slushy. Keywords: Winter road condition monitoring, Sensor Fusion, Deep Neural Network, Thermal Camera
DISCLAIMER: All information shared in the TRB Annual Meeting Online Program is subject to change without notice. Changes, if necessary, will be updated in the Online Program and this page is the final authority on schedule information.