Theor Appl Climatol (2011) 104:551–559 DOI 10.1007/s00704-011-0433-9

ORIGINAL PAPER

What spatial resolution do we need for a route-based road weather decision support system? L. Chapman & J. E. Thornes

Received: 13 August 2009 / Accepted: 23 February 2011 / Published online: 8 April 2011 # Springer-Verlag 2011

Abstract Since their implementation, road weather information systems have mostly relied on point measurements from outstations to initiate and verify daily forecasts. Initially, spatial extrapolation was achieved by thermal mapping, but this is gradually being replaced by routebased forecasting techniques. Both techniques are similar in the sense that they use a point measurement, often taken from an outstation, to provide a spatial forecast of road surface temperatures around the road network at varying resolutions. A substantial research effort has been undertaken to understand and model the complex environmental conditions and mechanisms responsible for the variation in road surface temperatures around the road network. In particular, the interaction of varying geographical parameters around the road network (e.g. altitude, land use, road construction, topography, etc.) has been used to develop local climatological models and route-based forecasting products. By considering the needs of winter maintenance engineers, this paper reviews the current state of the art and takes a critical look at the embedding of forecast products into decision support systems. This is achieved by considering a case study of how road surface temperature and condition vary across the width of a road profile, instead of just lengthways along a road. It is shown that temperature and condition both vary significantly across the profile, which immediately raises questions about the validity of current surveying and modelling practices. This has implications for both the resolution of route-based forecasting products as well as user confidence in automated decision support systems. L. Chapman (*) : J. E. Thornes University of Birmingham, Birmingham, UK e-mail: [email protected]

1 Introduction 1.1 A history of road weather information systems Traditionally, winter maintenance engineers base their nightly decision making by consulting a road weather information system (RWIS) which combines weather forecast data with road temperature and condition data. The first generation of RWIS, still frequently used in many countries, relies on methods and tools developed in the 1980s. Early ‘Ice Detection Systems’ simply comprised of a number of outstations which detected when ice had been formed upon the integral road sensor. This approach was limited for two reasons. Firstly, as the ice had already formed, it was too late to pre-salt the network. Secondly, the sensor only provided ‘spot’ measurements of slipperiness (road surface condition (RSC) and road surface temperature (RST) at a single location). It was generally unknown how representative this location was in comparison to the remainder of the network. This lack of spatial information was the motivation behind the development of thermal mapping techniques (e.g. Gustavsson 1999). Thermal mapping utilises an infrared thermometer to conduct a thermal survey from a moving platform. RSTs were typically taken at a 20-m resolution around the road network to provide the highway engineer with the local knowledge of thermal variations. It was assumed that if this was done over a number of nights of varying atmospheric stability, the variations in RST and RSC could be interpolated between outstations. Throughout the 1980s, these technologies matured into an ‘Ice Prediction System’ which, via an energy balance model, enabled RST and RSC to be forecast for each outstation before being interpolated with a thermal map. This system was quickly implemented in many countries

552

L. Chapman, J.E. Thornes

worldwide and has been considerably refined as computer processing power and communications improved. However, as technology has moved on, many of the components of the existing system were starting to look dated, and the whole system is now being superseded by route-based forecasting techniques, e.g. XRWIS: the neXt generation RWIS (Chapman and Thornes 2006; Thornes et al. 2005). Instead of relying on forecast interpolations made by thermal mapping, complete with its inherent limitations (Chapman and Thornes 2005), XRWIS models RST by considering the influence of the local geography on the road surface (Chapman and Thornes 2006). The forecast is displayed in a GIS environment and disseminated directly to the highway engineer via the Internet (Fig. 1). 1.2 Decision support systems In addition to high-resolution RST and RSC forecasts, routebased forecasting also includes an automated decision-making algorithm where salting routes are colour coded depending on the required action (Fig. 1c). This is an important step and ultimately transfers the decision making away from the highway engineer to the forecaster; RWIS essentially becomes a decision support system. This step towards automated and coordinated decision making is a key research goal of the winter maintenance community (e.g. Nixon 2010; Chapman et al. 2010). The rationale behind this approach is straightforward. With ever increasing forecast resolution, there is a need to summarise the forecasts to provide information and directions to the winter maintenance engineer instead of just a baffling amount of data (Bogren and Gustavsson 2007). Significant advances have been made along these lines by the US Federal Highway Administration with respect to its Maintenance Decision Support System (MDSS). This provides information to the engineer concerning appropriate strategies pertaining to treatment types, timing, rates and locations (Petty and Mahoney 2008; Nixon 2010; Chapman et al. 2010). Embedded into MDSS is the Model of the

a)

b)

Environment and Temperature of Roads (METRo model) developed by Environment Canada (Crevier and Delage 2001; Linden 2010). However, METRo presently does not have a spatial modelling component which limits the scale of potential savings of MDSS (although there is a clear potential to combine the systems). The inclusion of a spatial forecasting component (i.e. a route-based forecast) would provides enormous potential for savings to be made by leaving warmer routes untreated (i.e. selective salting) or ultimately enable dynamic routing practices (Handa et al. 2005; 2006). Therefore, the key difference between traditional RWIS and route-based forecasting is the change from a reliance on measurements to an increasingly high-resolution modelling approach. Instead of running a model for a handful of outstations across a network, the model can now be run for thousands of sites just metres apart. This is how the maps are produced in Fig. 1—these consist of point data and not line data, which is how the visualisation appears. This new approach has been facilitated by the recent proliferation of geomatics technology and increased computer power. No longer are surveys limited to just thermal measurements, it is now possible to measure many geographical parameters (e.g. altitude, sky-view factors, screening, aspect, slope, etc.) during a single geomatic survey. It is this increase in geographical data that has ultimately enabled the development of route-based forecasting techniques. However, is the current resolution of route-based forecasts sufficient? For the highway engineers to fully trust an automated decision support system, they need to be confident that a forecast fully captures the variation in conditions across the road network, particularly the coldest locations on which decisions ultimately need to be based. Although route-based forecasting is now used in many countries across northern Europe, the resolution of forecasts varies. XRWIS was the original route-based forecast and was developed for use in the UK (Chapman and Thornes 2006). It provides forecasts at a 50-m resolution, which is similar to the standard adopted by the Netherlands and

c)

Fig. 1 Visualisation of the new XRWIS paradigm showing a road surface temperature, b road surface condition and c ‘traffic light’ salting routes

Spatial resolution for a road weather decision support system

Spain (Smeding-Zuurendonk et al. 2010). An alternative approach is ‘stretch’ forecasting, which utilises a much coarser resolution and works by dividing the road network into sections of similar characteristics. This is the approach used in Denmark (Pederson et al. 2010) and in some parts of the UK (Brown et al. 2008). However, as RST can vary by as much as 5°C over a distance of several hundred metres (Smeding-Zuurendonk et al. 2010), it could be argued that road stretch forecasts presently lack the resolution for inclusion in a decision support system. However, even with increased resolution forecasts (e.g. 50 m), thermal singularities (i.e. cold spots; Saarikivi et al. 2008) caused by katabatic drainage or bridge decks may fail to be captured in the survey (Hammond et al. 2011). Theoretically, this can be accommodated in the new paradigm as surveys can be increased to cover every 5 m of road, or better, if needed. However, it does raise the question of how fine does the resolution need to be? When conducting any survey (whether thermal mapping or a XRWIS geomatic survey), there is always the factor of repeatability. It is impossible to survey the exact same point twice (Chapman and Thornes 2004), and therefore, it is very difficult to systematically survey a road to cover the full geographical variation encountered around a network. To highlight this point, this paper begins by presenting variations in RST and RSC at an understudied scale, the cross profile.

2 Cross profile road surface temperature differences It is not unusual to find variations in excess of 10°C in RST around a road network (Shao et al. 1996). This variation can be mostly explained by variations in geography and other parameters (Table 1). The influence of all these parameters is modelled in a route-based forecast. When dealing with the cross road profile, many of these parameters can be assumed constant. For example, there will be negligible changes in altitude and road construction across the road profile. However, some parameters will vary markedly. Notably, these are sky-view factors and screening (particularly in urban environments) as well as variations caused due to traffic. Although this reduces the magnitude of Table 1 Parameters controlling road surface temperature (Thornes and Shao 1991)

The parameters in italics are the factors controlling variations widthways across the road profile

553

the variations, it is not impossible to find variations in surface temperatures of the order of 2°C across the carriageway profile (Gustavsson et al. 2001; Shao 1990). 2.1 Sky-view factors and screening The sky-view factor (ψs) is a dimensionless parameterisation of the quantity of visible sky hemisphere at a location. When ψs is unity, the site is very exposed and subject to increased radiative cooling at night. Conversely, the same sites receive increased levels of direct beam solar radiation in daylight hours as there are no shading/ screening effects. As a result, ψs is the dominant control on RST (Chapman et al. 2001), and its inclusion as a parameter derived from either visible fisheye imagery (Fig. 2a), infrared hemispherical photographs (Chapman et al. 2007) or GPS proxy techniques (Chapman et al. 2002; Chapman and Thornes 2004), forms the basis of routebased forecasts (Chapman and Thornes 2006). Unlike many other geographical parameters, ψs is highly spatially variable. For example, in a rural environment where there are trees encroaching into the sky hemisphere directly over the road, the ψs (and thus RST) will vary significantly over even the smallest of areas (Fig. 2b). A similar effect is noted in urban canyons, where the ψs will vary depending on the location in the profile of the canyon (Fig. 2c). In the example provided, ψs can cause RST to vary by nearly 3°C. This has significant implications for how to conduct thermal or geomatic surveys. Ideally, these should be conducted in the centre of the road in urban canyons (to measure the maximum ψs and therefore, the lowest RST), but it is not always practical to do this. The thermal image shown in Fig. 2b was taken at night under stable conditions and underneath a tree-shaded section of a road. Just beyond this screened section, cold sections can be clearly identified, which correspond to clearings directly overhead in the tree canopy. Care must be taken when interpreting this image as the coldest pixels (i.e. −10.2°C) shown on the legend are actually measurements of the cold sky hemisphere in gaps in the tree canopy. Also, the image is not corrected for the effects of varying anisotropy or emissivity, so there will be variations of temperature apparent,

Meteorological

Geographical parameters

Road parameters

Solar radiation Terrestrial radiation Air temperature Cloud cover and type Wind speed Humidity/dew point Precipitation

Latitude Altitude Topography Screening Sky-view factor Land use Topographic exposure

Depth of construction Thermal conductivity Thermal diffusivity Emissivity Albedo Traffic

554

a)

L. Chapman, J.E. Thornes

c)

b)

Fig. 2 a Sample fisheye image from which sky-view factors and screening effects can be derived, b thermal image showing the variation of RST in relation to shading from trees (February 17, 2008)

and c variation of SVF in an urban canyon and resultant effect on road surface temperatures in stable conditions (adapted from Steyn and Lyons 1985)

which are due to the heterogeneous nature of materials and surfaces in the image (this is true for all thermal images used in this paper).

conducted, but the general trend is that differences of between 1°C and 2°C are not uncommon between inside and outside lanes (Farmer and Tonkinson 1989; Shao 1990). Thermal mapping techniques have also been used to systematically survey each lane of a major road for comparison with actual traffic count data (Chapman and Thornes 2005). However, these studies all have the same sampling limitations; they all use point measurements of data. Even thermal mapping studies are limited in this respect by the resolution of the thermal mapping technology (i.e. 20 m). The sampling will typically have been completed in the centre of each lane, so there will be no evidence of the influence of tyre tracks or lesser trafficked areas to the edge of the profile. An alternative technique to provide a snapshot of the variations is to use a thermal imaging camera. This can be easily mounted on a motorway gantry from which data can be readily collected of the thermal variation of the road profile. Data were collected under extreme conditions (high atmospheric stability) in the early hours of February 14, 2008. Two camera angles were used—a wide photo showing both carriageways (Fig. 4a) and a close-up of the southbound carriageway (Fig. 4b). The influence of traffic,

2.2 Traffic The most important and widely researched parameter concerning cross profile RST variations is traffic. Traffic modifies RST via a number of processes which are highlighted in Fig. 3. Of these processes, the most important are the addition of heat to the road surface via sensible heat and moisture fluxes from the engine as well as frictional heat dissipation from the tyres (Chapman and Thornes 2005; Prusa et al. 2002). Although traffic effects are difficult to model, they have a general cumulative effect, reducing the difference between air and surface temperatures (Crevier and Delage 2001), therefore promoting increased RST in heavily trafficked areas. For example, during the early morning peak commuting period, RSTs in Stockholm were 2°C warmer than in the suburbs (Gustavsson et al. 2001). This study was unusual as it is much more common to isolate traffic effects by studying multi-laned roads. A number of studies have been

Spatial resolution for a road weather decision support system

555

Fig. 3 A schematic illustration of the impact of traffic on RST. Taken from Chapman and Thornes (2005). Copyright Royal Meteorological Society. Reproduced with permission. Permission is granted by John Wiley & Sons Ltd on behalf of RMETS

particularly the presence of tyre tracks, can be clearly determined in each image. Spot temperatures of various pixels can be easily extracted from the thermal images. By using this approach, a quick and easy approximation of the surface temperature in an area can be obtained. Several spot temperatures have been used on Fig. 4a to provide an indication of how the temperature varies across the three lanes of the highway as well as the emergency hard shoulder (far left). Although temperature differences in the image can be easily identified (e.g. the hard shoulder is 1°C colder than the centreline of the inside lane and 1.4°C colder than the tyre track on the centre lane), simple spot measurements provide an inadequate sample of measurements to draw any valid conclusions regarding the cross road profile. Instead, an alternative methodology is used which samples sub-sectional areas across the carriageway. In Fig. 4b, two areas have been used for analysis. AR01 covers the entire carriageway (including the hard shoulder) whereas AR02 excludes the hard shoulder. The summary statistics are shown in Table 2. Interestingly, the magnitude of the results for AR02 (without the hard shoulder) are comparable to previous studies (Chapman and Thornes 2005; Farmer and Tonkinson Fig. 4 a Thermal image of the M5 motorway between junctions 3 and 4, UK showing spot temperatures on the northbound carriageway and b close-up of the southbound carriageway from which tyre tracks are clearly evident (February 14, 2008). AR01 and AR02 are the reference areas used for statistics

a)

1989; Gustavsson et al. 2001; Shao 1990]. However, the difference is slightly higher in magnitude than the results demonstrated by Chapman and Thornes (2005), which were taken from the same stretch of highway under similar extreme conditions. This can be explained by the more inclusive sampling strategy which will ensure all variations (e.g. tyre tracks) are sampled. These are difficult to take into account of when using thermal mapping techniques. However, the major difference in this study is the inclusion of the hard shoulder. This increases the temperature difference experienced across the road cross profile to 2.1°C, which is a significant variation. The hard shoulder theoretically has zero traffic flow as it is only used in an emergency, and therefore, the temperatures measured there are representative of what the temperatures would typically be across the carriageway without traffic effects.

3 Cross profile road surface condition differences The winter maintenance engineer is not exclusively interested in RST as the presence of moisture also needs to be forecast to determine whether or not ice will form on

b)

556

L. Chapman, J.E. Thornes

Table 2 Summary statistics of the analysis areas defined in Fig. 4b

AR01 AR02

No of pixels

Minimum

Maximum

Difference

3,750 3,480

−2.2°C −1.4°C

−0.1°C −0.1°C

2.1°C 1.3°C

the road surface. RST and RSC are inherently related, and the presence of traffic here can also have quite an effect. Figure 5a shows the differential drying of a road surface on a multi-lane road in southeast Sweden. The drying evident on the inside lane of this image is due to the effect of traffic processes shown in Fig. 3. In the same way as increased surface temperatures are evident on the more heavily trafficked inside lanes, it is here where the effects are more apparent, where the additional heat provided by the traffic is sufficient to promote drying on the inside lane (Chapman 2007). Figure 5b provides an alternative example of variations in RSC across the road surface. This is the opposite of the situation in Fig. 5a caused by traffic travelling over a wet section of the road (i.e. seepage) and carrying this moisture down the carriageway for several hundred metres. Such variations in RSC are difficult to quantify but can pose a considerable problem for winter maintenance engineers. Although roads may be forecast to remain dry throughout the night, there will be seepage points that require treatment. Furthermore, wet road surfaces are often the coldest sections of a road due to the increased latent heat loss from the wet road surface. This leaves these sections of a road prone to increased slipperiness under marginal conditions.

4 Discussion 4.1 Implications of varying road conditions in the cross profile The case studies provided in this paper have shown how RST and RSC can vary considerably across the road profile. The most obvious implication of this is that a different Fig. 5 a Differential drying on the E4 highway north of Gävle, Sweden; a shows the effect of how heat fluxes from traffic dry the road surface on the heavily trafficked inside lane and b seepage across a minor road in the UK

a)

forecast will be generated depending on where the forecast point is located on the cross profile (typically the centreline). However, this finding also has direct repercussions for the motorist. Take for example, a simple bend on a country road. There are an infinite number of ways of negotiating this corner. As this paper has shown, each route will potentially present the driver with a different combination of RST and RSC and therefore, slipperiness. If roadweather forecasts are based on centreline forecasts/measurements, then there is a danger that the highway engineer will effectively be using an ‘optimistic’ forecast. For example, Fig. 4a clearly shows that the hard shoulder is of the order of 1°C colder than the main carriageway. Should this be taken into account in forecasts? On a marginal night, the forecast may indicate that the main carriageway remains above freezing and so the decision is made not to treat the network. However, whilst the heavily trafficked lanes are above freezing, just 2 metres away on the hard shoulder, the road has fallen below freezing and is now slippery. Under these circumstances, the driver of a vehicle that deviates onto the hard shoulder or negotiates a corner away from the centreline is being put at risk and could be subject to an accident. Does the winter maintenance engineer have a duty of care to protect that motorist? In an environment of increasing litigation, the answer to this question is probably yes. The downside to this are the financial and environmental burdens of overtreating the network, and it is for this reason why maintenance engineers still express an interest in average/median RST and RSC rather than minimums. Overall, a case can be presented to take measurements and to locate forecast points to produce a forecast for the worst case scenario encountered on the cross road profile. Only then can all motorists be protected. This approach has implications for both surveying techniques and for road weather models. The influence of traffic is often cited as an important parameter to include in route-based forecasting (Chapman and Thornes 2005, 2006); however, due to the difficulties of modelling such effects, the general approach to including traffic in road weather models is by means of a simple parameterisation (e.g. Chapman et al. 2001; b)

Spatial resolution for a road weather decision support system

Hammond et al. 2010). There is now a growing research effort to model these effects more closely (Fujimoto et al. 2008; Prusa et al. 2002). However, is this the correct approach? Can a case be made where traffic parameterisations are not included in route-based forecasts? By using this methodology, the worst case scenario found on the cross road profile will be accounted for. Admittedly, this will lead to increasingly pessimistic forecasts and more expensive winter maintenance budgets, but are these justified by increasing numbers of lawsuits? It is these same concerns that will prevent highway engineers from fully utilising the benefits of decision support systems. 4.2 What resolution is needed for a route-based road weather forecast? Overall, route-based road weather decision support systems need to have the ability to resolve the thermal characteristics of the road network to the extent that a highway engineer has the confidence to use the decision support system. However, as this paper has demonstrated, significant differences in RST and RSC exist even at the smallest of scales. Whilst the original RWIS point outstation network was clearly inadequate, route-based forecasting has started to address the problem. Although 50 m between survey points is presently the finest resolution available, this is only just sufficient to cover most thermal singularities, and hence, there is some justification in increasing this resolution to 10 m or better. Even then, it is unlikely to capture the full variation of RST and RSC experienced across the network. Furthermore, how can the differences in the cross road profile be taken into account? Indeed, are these variations just as important? These ideas can be taken a stage further, what about variations at the sub-metre scale? These can be quite considerable. Figure 6 would indicate that it is not impossible for RST to vary by several degrees even across just a few centimetres. By taking a representative area of the pave-

Fig. 6 Close-up thermal image of approximately 30 cm2 of road surface (loose chipping, top right, provides scale)

557

ment (AR01, approximately 5 cm by 20 cm), which ignores the loose chipping used for scale to the top right of the image, variations of the order of 1.6°C are apparent. This is comparable to the results found of RST variation on multilaned roads (e.g. Chapman and Thornes 2005; Shao 1990). So, what scale is really necessary? Modelling to the submetre scale is clearly impractical, but the thermal differences encountered are not insignificant. However, perhaps a scale of 1 to 5 m is not unrealistic in the future as this will also take into account any cross road profile differences, although this will be limited by computer processing. Decision support systems provide the framework to allow the forecaster and maintenance engineer to make sense of all the additional data generated by higher-resolution forecasts. Forecast verification is the key element required in order to provide the winter maintenance engineer with the confidence to use an automated decision support system. The verification of route-based forecasts has always been a challenge, although new techniques are under development to improve this situation. For example, Hammond et al. (2010) have investigated the use of advanced statistical techniques to cluster road sections together for highresolution monitoring by infrared sensors. A separate approach under investigation is the collection of real-time weather data by proxy directly from motorists using information pertaining to road slipperiness and temperature by measuring ABS, wiper usage and stopping distances (Drobot et al. 2010; Bogren 2010). However, as it stands, the existing measurement network based on weather outstations is clearly insufficient to verify a route-based forecast at the level demanded by winter maintenance engineers. Thermal mapping is the traditional solution to interpolate between outstations. Unfortunately, both thermal mapping and route-based forecasts depend on a survey vehicle taking point measurements around the road network. The repeating of surveys can give confidence in the accuracy of measurements, but there are many errors involved in spatial joining surveys for comparative analysis (Chapman and Thornes 2004). Furthermore, it is impossible for the same survey vehicle to survey exactly the same route on two occasions. There will be instances where the vehicle needs to change lanes to overtake or will corner at a slightly different angle. These will produce erroneous measurements as the vehicle will deviate from the warm centreline onto the colder sections between lanes, perhaps indicating the false existence of a thermal singularity. One way in which this could be overcome would be by the development of next-generation thermal mapping techniques utilising video thermal imaging. This is not impossible and has indeed already been used on the railway network (Chapman et al. 2006). Furthermore, additional remote sensing techniques for the measurement of slipperiness, and

558

therefore RSC, are being piloted (e.g. Saarikivi et al. 2008; Nurmi et al. 2010). Hence, it is only in making consistently more detailed measurements (including the cross profile) can the appropriate position of forecast points be fully determined.

5 Conclusions This paper has highlighted the problems faced when making point forecasts. The increased resolution of forecast points offered by route-based products provides added information to inform decision making by winter maintenance engineers. In this regard, the resolution of the forecast is irrelevant, as a ‘stretch’ forecast is equally useful as a higher-resolution forecast provided that it resolves the key thermal characteristics of the road network. However, based on the concept of ‘information, not data’ (Bogren and Gustavsson 2007), there is a need to be able to summarise all the additional information for the highway engineer. This is where a decision support system is used— effectively transferring the decision making away from the engineer towards the weather forecaster. Such a step requires the highway engineer to have full confidence in the quality and accuracy of the forecast. Essentially, the decision support system replaces the ‘local knowledge’ of the winter maintenance personnel. Hence, the required resolution of a route-based forecast needs to be comparable to what is presently used in decision making. Therefore, a resolution of 50 m is clearly adequate for decision making, if not superior, provided that the forecast points are located in the coldest section of the cross profile. This will be informed by advanced thermal mapping using infrared imagery such as the case studies shown in this paper, but could also be modelled quite simply by excluding traffic parameterisations presently included in route-based forecasts. Although this approach will add a negative bias to the forecast, it will ensure that the highway remains safe and secure for all users.

References Bogren J (2010) Traffic data and road weather. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 22 Bogren J, Gustavsson T (2007) Information, not data: future development of road weather information systems. Geogr Ann Ser A Phys Geogr 89:263–271 Brown A, Jackson S, Murkin P, Sheriden P, Skea A, Smith S, Veal A, Vosper S (2008) New techniques for route based forecasting. Proceedings of the 14th International Road Weather Conference, Prague, May 2008. SIRWEC, p 5 Chapman L (2007) Differential drying of a multi-laned road surface. Weather 62:200

L. Chapman, J.E. Thornes Chapman L, Thornes JE (2004) Real time sky-view factor calculation and approximation. J Atmos Oceanic Technol 21:730–741 Chapman L, Thornes JE (2005) The influence of traffic on road surface temperatures: implications for thermal mapping studies. Meteorol Appl 12:371–380 Chapman L, Thornes JE (2006) A geomatics based road surface temperature prediction model. Sci Total Environ 360:68–80 Chapman L, Thornes JE, Bradley AV (2001) Modelling of road surface temperatures from a geographical parameter database. Part 2: numerical. Meteorol Appl 8:421–436 Chapman L, Thornes JE, Bradley AV (2002) Sky-view factor approximation from GPS receivers. Int J Climatol 22:615–621 Chapman L, Thornes JE, White SP (2006) Thermal imaging of railways to identify track sections prone to buckling. Proc Inst Mech Eng, F: J Rail Rapid Transit 220:317–327 Chapman L, Thornes JE, Muller JP, McMuldroch S (2007) Potential applications of thermal fisheye imagery in urban environments. IEEE Geosci Remote S 4(1):56–59 Chapman MB, Drobot S, Linden SK, Cowie J, Mahoney WP (2010) A decision support system for winter weather maintenance of roads, bridges and runways. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 8 Crevier L, Delage Y (2001) METRo: a new model for road-condition forecasting in Canada. J Appl Meteorol 40:2026–2037 Drobot S, Chapman M, Pisano P, McKeever B (2010) Overview of the vehicle data translator. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 16 Farmer SF, Tonkinson PJ (1989) Road surface temperature model verification using input data from airfields, roadside sites and the mesoscale model. UK Meteorological Office, Exeter Fujimoto A, Watanabe H, Fukuhara T (2008) Effects of vehicle heat on road surface temperature of dry condition. Proceedings of the 14th International Road Weather Conference, Prague, May 2008. SIRWEC, p 37 Gustavsson T (1999) Thermal mapping—a technique for road climatological studies. Meteorol Appl 6:385–394 Gustavsson T, Bogren J, Green C (2001) Road climate in cities: a study of the Stockholm area, south-east Sweden. Meteorol Appl 8:481–490 Hammond D, Chapman L, Thornes JE, White SP (2010) Verification of route-based winter road maintenance weather forecasts. Theor Appl Climatol 100:371–384 Hammond D, Chapman L, Thornes JE (2011) Parameterising road construction in route-based road weather models: can ground penetrating radar provide any answers? Meas Sci Technol 22. doi:10.1088/0957-0233/22/5/057001 Handa H, Chapman L, Yao X (2005) Dynamic salting route optimisation using evolutionary computation. IEEE Congress on Evolutionary Computation. IEEE CEC 2005(1):158–165 Handa H, Chapman L, Yao X (2006) Robust route optimisation for gritting/salting trucks: a CERCIA experience. IEEE Comput Intell M 1(1):6–9 Linden SK (2010) The evolution of METRo in a roadway decision support system. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 15 Nixon W (2010) Operational resilience in maintenance decision support systems. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 13 Nurmi P, Hippi M, Juga I (2010) Evaluation of FMI’ new forecast model of road surface friction. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 24 Pederson TS, Peterson C, Sattler K, Mahura A (2010) Physiographic data for road stretch forecasting. Proceedings of the 15th

Spatial resolution for a road weather decision support system International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 48 Petty KR, Mahoney WP (2008) The U.S. Federal Highway Administration winter road Maintenance Decision Support System (MDSS): recent enhancements & refinements. Proceedings of the 14th International Road Weather Conference, Prague, May 2008. SIRWEC, p 14 Prusa JM, Segal M, Temeyer BR, Gallus WA, Takle ES (2002) Conceptual and scaling evaluation of vehicle traffic thermal effects on snow/ice covered roads. J Appl Meteorol 41:1225–1240 Saarikivi P, Hippi M, Nurmi P, Sipilä J (2008) Observing the variability of road and weather conditions with hybrid mobile and fixed sensors. Proceedings of the 14th International Road Weather Conference, Prague, May 2008. SIRWEC, p 24 Shao J (1990) A winter road surface temperature model with comparison to others. Unpublished PhD Thesis, Department of Geography, University of Birmingham, UK

559 Shao J, Lister PJ, Hart GD, Pearson HB (1996) Thermal mapping— reliability and repeatability. Proceedings of the 6th International Road Weather Conference, Birmingham, April 1996. SIRWEC, pp 235–242 Smeding-Zuurendonk I, Wokke M, Wisse J (2010) Road surface temperature forecasting for gritting routes. Proceedings of the 15th International Road Weather Conference, Québec City, Canada, February 2010. SIRWEC, p 46 Steyn DG, Lyons TJ (1985) Comments on “The determination of view-factors in urban canyons”. J Climate Appl Climate Meteor 24:383–385 Thornes JE, Shao J (1991) Spectral analysis and sensitivity tests for a numerical road surface temperature prediction model. Meteorol Mag 120:117–123 Thornes JE, Chapman L, White SP (2005) The X-Factor. Surveyor Magazine 192(5813):14–16