Development and Evaluation of Geostatistical Methods for Estimating Weather Related Collisions: A Large-Scale Case Study

Abstract

Winter driving conditions pose a real hazard to road users with increased chance of collisions during inclement weather events. As such, road authorities strive to service the hazardous roads or collision hot spots by increasing road safety, mobility, and accessibility. One measure of a hot spot would be winter collision statistics. Using the ratio of winter collisions (WC) to all collisions, roads that show a high ratio of WC should be given a high priority for further diagnosis and countermeasure selection. This study presents a unique methodological framework that is built on one of the least explored yet most powerful geostatistical techniques, namely, regression kriging (RK). Unlike other variants of kriging, RK uses auxiliary variables to gain a deeper understanding of contributing factors while also utilizing the spatial autocorrelation structure for predicting WC ratios. The applicability and validity of RK for a large-scale hot spot analysis is evaluated using the northeast quarter of the State of Iowa, spanning five winter seasons from 2013/14 to 2017/18. The findings of the case study assessed via three different statistical measures (mean squared error, root mean square error, and root mean squared standardized error) suggest that RK is very effective for modeling WC ratios, thereby further supporting its robustness and feasibility for a statewide implementation.

Advances in Regression Kriging-Based Methods for Estimating Statewide Winter Weather Collisions: An Empirical Investigation

“…However, as point measurements they do not provide sufficient information to all roads and areas natively; thus, these values need to be interpolated to ensure complete statewide coverage. As found in previous transportation and environmental studies, the use of kriging is an effective and efficient method for spatially interpolating road surface and environmental data for widespread coverage [12,18,26,27]. Therefore, following their methods, ordinary kriging was used to interpolate these values to ensure statewide coverage for all road segments.…”

“…Table 1 provides the summary statistics for the covariates used. Following the original study, the covariates include the annual average daily traffic volumes (AADT), road surface temperatures (RST), seasonal snowfall averages, daily air temperatures (average, max, and min), and the road surface index surrogate of road warning messages (red, orange, and yellow classifications) [18]. Furthermore, additional road characteristics, such as the posted speed limits and the number of lanes, were also incorporated into the regression analysis.…”

“…Detailed descriptions of each step will follow throughout this section. The majority of the results are obtained following the RK process detailed in our previous study [18] but now expanded to the state and regional study areas and with higher data densities. All relevant collisions were categorized as a winter collision or not and then mapped to the appropriate road segments.…”

“…Our previous study sought to address the aforementioned limitations of previous studies by applying regression kriging to a larger scale, that being a quadrant of Iowa, and using five years' worth of data. Ultimately, it was shown that Regression Kriging (RK) is a viable and effective network screening methodology, though it was still limited in scope by only focusing the method to that one quadrant of Iowa [18]. However, by increasing the spatial scope, the spatial transferability of the variance, known as the second order stationarity assumption (SOSA), must be checked, which was omitted.…”

“…Future Transp. 2021, 1, FOR PEER REVIEW 3 is a viable and effective network screening methodology, though it was still limited in scope by only focusing the method to that one quadrant of Iowa [18]. However, by increasing the spatial scope, the spatial transferability of the variance, known as the second order stationarity assumption (SOSA), must be checked, which was omitted.…”