A traditional statistical approach to derive a novelty forecast variable – Road surface friction

It is generally known that the frequency of traffic accidents is highly correlated with road surface conditions due to adverse weather. Road surface friction and, accordingly, the grip between the tires of a vehicle and the road surface, decrease rapidly during snowfall and under icy conditions. There are myriads of high-quality instruments and gadgets to measure and observe surface friction (or “skid resistance”) reliably. However, modeling and especially forecasting of road surface friction, or slipperiness, are rare applications in meteorology at least in an operational forecasting environment. This paper illustrates the traditional ‘Perfect Prog' method being tuned for a novelty forecast variable, the road surface friction. The goal is to improve and extend wintertime predictive capabilities in a modern weather forecasting office. This method has already been tested under an operational setting at the road weather service of the Finnish Meteorological Institute (FMI).

The underlying principle of our new friction forecast application is based on (i) the statistical relationship between observed friction and a selection of meteorological variables by applying the regression analysis, and (ii) using this derived relationship in a forecast mode by utilizing output of an operational NWP model which produces forecasts of these given meteorological variables. Hence the methodology mimics the traditional ‘Perfect Prog' approach, but in a totally new context.

To define the statistical association, friction observations were inferred during two winter seasons (2007/08 and 2008/09) from optical Vaisala Remote Road Surface Sensors (DSC111) installed at selected road weather stations in Finland. The device measures the depth of water, snow and ice on the road surface and produces an estimate of surface friction. Such measurements are ideal for this kind of application, because road surface temperature and the thickness of water, snow and ice layers as well as friction are obtained at the same time, at the same location.

In the development of the ‘Perfect Prog' regression equations the following weather parameters were used as potential predictors: road surface temperature (TR), dew point temperature (TD), relative humidity (RH), precipitation intensity (PI), and height (thickness) of snow, ice and water layers (HS, HI, HW, respectively; measured in millimetres of water equivalent). The sum, HS+HI, and the difference, TR-TD, were also estimated, when the latter would provide an indication of potential formation of hoar frost when TR < TD.

Separate distinct regression models were defined (1) for snow and/or ice covered roads, (2) for wet roads, and (3) for cases when all these three forms exist simultaneously:

where A to D are regression coefficients. The "general model" (1) yielded a correlation of 0.84 in the dependent data set explaining c. 70% of the total variance of observed friction. Correlations of more or less the same magnitude were obtained when applying the regression equations on an independent data set during the follow-up winter of 2009/10. More detailed validation results are presented in Juga et al (2012) and Nurmi et al (2010).

The following, forecasting phase of the ‘Perfect Prog' technique substitutes the predictors in the regression equations with deterministic short-range NWP forecasts of the given predictors. This technique, as known, will not attempt to correct for possible NWP model errors, but takes the output for predictor variables at their face value. One fundamental issue lies hence in the quality of the used NWP output, as the approach assumes them to be “perfect”. This, of course, is never the case as accurate prediction of the amount of snow, ice and water on a road surface is highly challenging. Nevertheless, high-quality forecasts of the general meteorological situation by the NWP model are essential, as are specific forecasts of the road conditions by a dedicated road weather model (RWM) (both operational at FMI). The RWM takes into account various complex processes such as the vertical energy transfer at the road surface and the effects of traffic flow. Inaccuracies in the predicted thicknesses of snow and ice are expected to cause large errors in the eventual friction forecasts.

The statistical regression relationships could probably be improved with more extensive data coverage under different weather situations and also by defining the equations for more road weather stations than until now. Investigation of logistic, instead of linear, regression has already been initiated. A major shortcoming hampering further development of the road weather forecast model producing the predictor values is the unavailability of real-time information about road maintenance actions (e.g., salting, ploughing) along the roads, because such data would be valuable to be included in the model. Hence, the friction forecasts should be taken rather as risk forecasts providing an indication of the worst case scenario road conditions that may appear.

Modeling road surface friction is a new method to analyze and predict slipperiness of the roads. The application introduced here has been under operational testing and evaluation in an operational environment during the past two winters (2010-12) and verification results are underway. The application represents a prime example of transition of meteorological research efforts into operations and can, as such, be adapted pretty straightforward for various ITS (Intelligent Transport System) applications, end-user products and warning services. The methodology will be further developed and exported from Finland to new environments during 2012-14 in the new international ITS project CoMoSef (Co-operative Mobility Services of the Future). CoMoSef will culminate in the 2014 Sochi Winter Olympics, where the roads leading to the sports venues, notorious for their challenging winter weather conditions, will serve as our pilot testbed. These activities are associated with a full-scale forecast demonstration project endorsed by the WMO.

References:

Juga I, Nurmi P and Hippi M, 2012. Statistical modelling of wintertime road surface friction. Meteorological Applications. Accepted. DOI: 10.1002/met.1285.

Nurmi P, Hippi M and Juga I, 2010. Evaluation of FMI's new forecast model of road surface friction. In Proceedings of SIRWEC 15th International Road Weather Conference, Quebec City, Canada, 5-7 February 2010. Available from http://www.sirwec.org/conferences/Quebec/full_paper/21_sirwec_2010_paper_nurmi.pdf.