Prepared by:

Jeff Suggett, M.Sc Manager, Special Projects Synectics Transportation Consultants Inc.

Alireza Hadayeghi, M.A.Sc, Ph.D. (Candidate) Manager, Transportation Safety Systems, Synectics Transportation Consultants Inc.

Dr. Jean Andrey, Ph.D. Associate Professor Faculty of Environmental Studies, University of Waterloo

Brian Mills, B.E.S. Adaptation and Impacts Research Division Atmospheric Science and Technology Directorate, Environment Canada c/o Faculty of Environmental Studies, University of Waterloo

Geoff Leach, P. Eng., Vice-President Integrated Maintenance & Operations Services Inc. (IMOS)

Abstract

This paper discusses the development of two winter severity indicator models that can be used to evaluate the relative harshness of a winter in comparison with a base period. A winter severity index is a measure of the relative impact of winter weather on winter road maintenance operations using historical meteorological or road weather information system data. The primary purpose of the winter severity indicator models is to assess the effectiveness of salt management. Winter road maintenance data were collected from across Canada. Salt usage in tonnes (salt (t)/lane-km/day) was chosen as the dependent variable, standardized to account for differences in road network and the number of days in the observation period.

The first model developed based on meteorological data alone achieved a goodness of fit of 0.54. Explanatory variables were based on snowfall occurrence, air temperature, freezing rain occurrence, and a locational dummy variable. An index was developed based on the predicted values using a scale between 1 and 100. A second model was developed based on meteorological data together with road weather information system data. This achieved a goodness of fit of 0.60, but was based on a significantly smaller sample size. In this model, pavement temperature was substituted for air temperature. The results of the first model were then calibrated to local geographic areas. Local calibration factors were developed using the Bayesian method. Based on the calibration, thirteen of twenty groupings achieved a better goodness of fit compared to the national model results. The model results show a better performance in heavily populated areas and in eastern Canada. Limitations of the models and recommendations for further research are presented in the paper.

For a complete copy of this paper, please contact: jsuggett@synectics-inc.net