مدل تفکیک زدایی برای کمی کردن اثرات ایمنی فعالیت های تعمیر و نگهداری جاده ها در فصل زمستان در سطح عملیاتی

This research presents a disaggregated modeling approach for investigating the link between winter road collision occurrence, weather, road surface conditions, traffic exposure, temporal trends and site-specific effects. This approach is unique as it allows for quantification of the safety effects of different winter road maintenance activities at an operational level. Different collision frequency models are calibrated using hourly data collected from 31 different highway routes across Ontario, Canada. It is found that factors such as visibility, precipitation intensity, air temperature, wind speed, exposure, month of the winter season, and storm hour have statistically significant effects on winter road safety. Most importantly, road surface conditions are identified as one of the major contributing factors, representing the first contribution showing the empirical relationship between safety and road surface conditions at such a disaggregate level. The applicability of the modeling framework is demonstrated using several examples, such as quantification of the benefits of alternative maintenance operations and evaluation of the effects of different service standards using safety as a performance measure.

Winter snow storms have a significant impact on the safety and mobility of highway users. Highway collision rates often increase considerably during snow storms due to slippery road conditions and poor visibility (Andrey et al., 2001, HASTE report, 2002, Knapp et al., 2000, Andrey and Knapper, 2003, Eisenberg and Warner, 2005, Velavan, 2006 and Qiu and Nixon, 2008). Weather related collisions are costly to the society. Andrey et al. (2001) estimated that injury and property damage accidents occurring due to inclement weather cost around $1 billion per year in Canada. Winter storms can cause substantial delay due to reduced traffic speeds and road capability as well as increased collisions. To reduce the negative impacts of winter storms, transportation agencies spend significant resources every year to keep roads and highways clear of snow and ice for safe and efficient travel. Canadian road officials spend around $1 billion each year on winter road maintenance and put around five million tons of salt on Canadian roads (Transport Association of Canada, 2003). This amount excludes other related costs such as damage to the environment, road infrastructure, and vehicles due to salt use (Environment Canada, 2002 and Perchanok et al., 1991). While important for maintaining road safety in Ontario, road salting has also raised significant concerns due to its potential damage to the environment, roadside infrastructure, and vehicles (Perchanok et al., 1991 and Environment Canada, 2002). A recent study by Environment Canada concluded that road salts at high concentrations pose a risk to plants, animals and aquatic systems (Transport Canada, 2001). While there is a consensus that winter road maintenance is beneficial to the nation's economy in general and to the safety and mobility of our highway system in particular, only a few efforts have been devoted to the problem of quantifying the safety and mobility benefits of winter road maintenance (Hanbali, 1992, Norrman et al., 2000, Fu et al., 2006 and Usman et al., 2010). Furthermore, most of the few existing studies have adopted highly aggregated approaches in terms of temporal and spatial levels (e.g. by month, season or year and over a city or region-wide). Usman et al. (2010) was among the first to develop collision models at a disaggregate level with the objective of linking the number of collisions on highways over individual snow storms to the average road weather and surface characteristics and road characteristics. While this level of disaggregation is sufficient for evaluating the average effects of various storm-wise factors, including those of bare pavement policies and standards currently being used in practice, it is not applicable for quantifying the safety effects of specific maintenance treatments deployed over a given storm. This paper describes a disaggregate modeling framework proposed for quantifying the impact of road surface conditions and weather factors on winter collision occurrence controlling for traffic exposure and site-specific characteristics. The significance of this effort is twofold. First, this work fills the knowledge gap on the quantitative understanding of how different road weather and surface conditions and traffic factors influence the road safety. Second, the disaggregate accident frequency model developed in this research is the first in the winter road safety literature, providing a foundation for allowing quantification of the safety effects of individual winter road maintenance operations. Given the limited resources available and the growing concern about negative environmental effects associated with some of the winter road maintenance practices such as salting, the ability to perform such detailed analyses is needed in order to develop outcome oriented performance measurement systems for evaluating winter road maintenance related policies and decisions. The paper illustrates the two potential applications of the developed models, namely, evaluation of the safety benefit of particular maintenance operations and maintenance standards. The developed models are expected to be used by local agencies for assessing different decisions related to winter road maintenance. The paper is organized as follows. The next section provides a literature review of winter maintenance operations, weather and road safety. The proposed methodology, model structure and data, is explained in Section 3. Modeling results, their interpretation and application are given in Section 4. Section 5 highlights the main conclusions and outlines some directions for future research.

This paper presents a disaggregate modeling approach aimed at identifying the factors affecting winter road safety and quantifying the effect of winter road maintenance on road collisions during snow storm events. Detailed hourly data on collision counts along with the corresponding road weather and surface conditions, and traffic on 31 patrol routes across Ontario, Canada, over six winter seasons (2000–2006) were obtained and used for model calibration. Two modelling structures were used – a multilevel Poisson lognormal model (PLN) accounting for within storm correlation and site-specific effects and a single level generalized negative binomial (GNB) model. Four different models were calibrated and it was found that the within storm correlation is relative weak and GNB has a better fit to the data by virtue of its ability to account for the heterogeneity in the data through varying dispersion parameter. Factors such as visibility, precipitation intensity, air temperature, wind speed, exposure, indicator for month, trend within storm, and site specific factors have statistically significant effects on winter road safety. Most importantly, road surface conditions as represented by a comprehensive measure called road surface index (RSI) were found to have a signification contribution to the variation of collisions within and between individual storms and maintenance routes. The practical significances of the developed models have been clearly illustrated in the two example case studies. This research can be extended in several directions. First of all, the modeling approach should be applied to a larger number of study sites from different regions with winter weather of varying severity and duration. This extension is necessary to test the robustness and reliability of the proposed modeling methodology. Secondly, this research has primarily focused on the basic effect of the influencing factors; the potential non-linear effect of these variables as well as their interaction terms should also be investigated. Finally, detailed geometrical features of the highway routes should be evaluated for their effects on collision frequency under adverse winter weather conditions. With this extension, the transferability of the models can be improved substantially.