1. Introduction
The road surface temperature is traditionally monitored by sensors installed in the pavement. The data retrieved from these sensors are used in models determining the road status conditions. This requires an accurate monitoring of the topmost road temperature, combined with meteorological measurements and forecasts, to correctly predict if and when ice will form on the pavement and to determine how severe the snowy and icy conditions will become. The location of monitoring stations is crucial and guidelines for the installation of monitoring stations have been developed (Manfredi et al. 2008). The performance of the sensor is also important. The Aurora Consortium has written a report on sensor performance regarding surface temperature measurements. The study shows that pavement sensors, that is, ground temperature sensors installed at a depth of a few millimeters, generally provide a good level of accuracy and offer low error indications. It was also noted that more research is required in relation to the different sensor constructions, materials, compositions, and road installations to understand the sensors’ response to temperature changes caused by strong heat radiation in or out of the road body (Scott et al. 2005). Efforts have also been made to overcome the limitations of spot measurement in relation to ground temperature sensors by using vehicle-mounted sensors (Sakai and Stenson 1990).
It is essential to have knowledge of the temperature distribution in the pavement to estimate the extent of road slipperiness at the point when the fluid on the pavement freezes and to initiate preventive anti-icing actions. Knowing the temperature distribution in the pavement provides important information to the warning systems regarding slippery road conditions and assists in optimizing the spread of de-icing agents (SIRWEC 2007).
Current methods for measuring road surface temperature are very coarse since the measurements are performed at small individual points. However, it is known that the temperature varies both across and along the road and that these variations will affect the state of the road surface. Furthermore, it is known that the temperature of the wheel tracks is normally slightly warmer than the rest of the pavement due to the friction heat from the tires of moving vehicles. A detailed examination regarding how the traffic affects the pavement, especially during heavy traffic conditions, can be found in Prusa et al. (2002). Prusa et al. showed how the rolling friction could be estimated for specific vehicles under different weather conditions. The estimation formula formulated by Prusa et al. could not, however, be validated within this project because this study focuses on practical measurements and, for these purposes, knowledge of the vehicles is not registered. The traffic conditions on the road used in our research are light to medium and, in addition to traffic, the local climate, such as winds and clouds, influences the condition of the pavement. It has also been proved that the amount of sunlight that reaches the road surface will affect the pavement condition, but that the wind speed does not affect it to the same extent (Bogren et al. 2000). Nevertheless, it is possible to predict the road temperature and the possible buildup of ice by using a surface energy balance equation (Bogren et al. 2000) for which the road temperature is one of the most important equation parameters. This justifies this study, which aims to discover whether the IR temperature measurements can increase the knowledge of the road status.
Research has been conducted with regard to predicting the road temperature using the heat flux as the parameter and for which the data used are from the Road Weather Information System (RWIS) (Sass 1992). Further research has been conducted to determine the road condition by performing a multivariate analysis on the RWIS data together with the data from camera images (Jonsson 2011). In both of these studies, surface-mounted sensors were used to determine the surface temperatures. This study indicates that infrared temperature sensors provide additional information for these models.
The models, developed in previous research, demand accurate and reliable sensor data for the numeric calculations. An example of a model for road forecasting has been conducted by Crevier and (Delage 2001). The difficulties involved in measuring the topmost surface temperature by using surface-mounted sensors and the possibilities of using IR techniques are discussed by Lindqvist (1987). Many of the obstacles identified by Lindqvist in relation to the implementation of IR technology, such as sensor complexity and price, have been solved by the technical developments with regard to both sensors and microcomputers. The evolving IR technology has led to the development of new pavement temperature sensors and pavement condition sensors, such as DST111 and DSC111 from Vaisala (http://www.vaisala.com/weather/products.html), and promising results have been obtained from the evaluation of these sensors. One key finding was that the temperature readings between the IR sensor and the sensors installed in the pavement were comparable, even allowing for the fact that minor differences existed (Feng and Fu 2008; Tilley and Johanneck 2008).
In the future, it is possible that infrared camera sensor measurements could be further developed to complement road-installed sensors so as to provide more information about the road surface temperature and state. One advantage of using an IR camera instead of an IR sensor is that the camera provides a detailed picture of the temperature distribution of a given section of a road instead of only a spot temperature measurement. Both an IR camera and an IR sensor were used in this study.
2. Theory
The integrand in (1) is known as Planck’s distribution, Eλ,b, and as we are dealing with real, nonblackbody objects in infrared thermometry, the spectral emissive power described by Planck’s distribution is more suited to our purposes.
Equation (5) shows that the sensitivity ΔEλ/ΔT increases as the wavelength decreases and that errors in measuring the spectral emission produce larger temperature errors at longer wavelengths. The conclusion based on this is that the shortest possible wavelength for infrared thermometry should be chosen with respect to the atmospheric absorption (Childs 2001). This absorption is mainly due to water vapor and carbon dioxide (Mikhailenko et al. 2005; Rothman et al. 2009); see Fig. 1. The majority of infrared cameras, intended for thermometry, measure within the 8–14-μm wavelength window.
Absorption of IR light by common atmospheric gases. The atmospheric absorbance is low in some wavelength bands, for example, between 8 and 14 μm.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
3. Methodology
a. Infrared thermometry
An infrared temperature camera was installed at an RWIS test site in Myggsjön, which is located in the middle of Sweden, approximately 600 km south of the Arctic Circle. The test site has been made available for research purposes by the Swedish Transport Administration. The road section at the test site has an annual daily traffic flow of approximately 4000 vehicles according to the Swedish Transport Administration. The standard RWIS parameters recorded were air temperature, relative humidity, wind speed, wind direction, precipitation type, precipitation amount, and ground temperature at 1-min intervals. Images of the road section were recorded using a standard color camera, which was capable of retrieving black-and-white night images using an IR searchlight. The test site layout is described in Fig. 2. The mast is 6 m high, and at the top the wind speed, the wind direction, and the precipitation are measured. The cameras and searchlight are installed at a height of 5 m, and the air temperature and humidity sensors are installed at a height of approximately 2 m.
Sensor installation layout at the test site located in the middle of Sweden. The mast is 6 m high, and at the top wind speed wind direction and precipitation are measured. At 5-m height the cameras and searchlight are installed, and at approximately 2-m height the air temperature and humidity sensors are installed. Because of the large amount of data retrieved during this research, data were stored locally in a ruggedized personal computer.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
The temperature of the pavement and the ground temperature at a depth of 0.3 m were both measured using Pt100 DIN class A temperature probes. The surface temperature was measured by means of an A320 IR camera from FLIR systems (www.flir.com). The IR camera operates within the 8–12-μm range, for which there is a low atmospheric absorption; see Fig. 1. The sensitivity of the camera is less than 0.070°C at 30°C and the accuracy is better than ±2°C at 20°C. The accuracy can be calibrated down to the sensitivity of the camera by using reference temperature probes. In this case, the camera was calibrated against the ground-mounted Pt100 probes at 2-mm depth available at the site. The Pt100 probes have an accuracy of better than ±0.3°C. The IR camera has a resolution of 320 × 240 pixels and was installed together with the road camera at approximately 5 m above the pavement and 4 m from the road surface edge. The IR camera lens was chosen so that the measurements would cover the entire width of the road section.
In addition to the IR camera, an IR thermometer of brand Optris CT laser (http://www.optris.com) was installed at approximately 2 m above the road and 0.5 m from the pavement edge. The diameter of the measurement spot size was 8 cm in diameter for the IR thermometer, thus it was measuring over a much smaller area than that of the IR camera. The temperature reading from the IR thermometers was recorded every second. The IR thermometer had a spectral range of 8–14 μm, an accuracy at 20°C of ±1% or ±1°C, whichever is the greater, and has a resolution of 0.1°C.
The emissivity ɛ for the road surface is between 0.8 and 0.99, depending on the precipitation building up on the surface (Brewster 1992); see Table 1. According to Eq. (3), this emissivity variance, due to precipitation buildup, will only affect the temperature reading by 0.1%, the implication of which is that it is possible to monitor the temperature without performing emissivity adjustments during precipitation and still be able to maintain an acceptable accuracy.
Emissivity at 300 K.
The aim was to collect the data as frequently as possible from all the sensors, but when the limitations of the local computer computational power and storage space were taken into consideration, the comprise that had to be made was that the recording of data from the majority of the sensors and the IR camera had to be at 1-min intervals, the IR thermometer at 1-s intervals, and the recording of color camera images at 10-min intervals. In relation to the following evaluation of the retrieved data, the 1-min data were the primary source used. Regular backups from the database and from the IR camera images were performed to ensure the availability of the data.
The road condition, as described in Figs. 3–5, was determined by examining the road status data and the weather data together with the camera images and information from the road maintenance personnel visiting the site. Information concerning salting and snow removal was also taken into consideration when determining the road status. The resulting road condition information for the wheel tracks and in between the wheel tracks was used when the decision was made as to whether the road conditions were dry and stable, icy and snowy, or wet and snowy.
Data from a dry period with stable, dry weather conditions. This period is used as reference. The small precipitation amount detected on 16 Mar can be neglected as it could not be observed on the road surface. Please refer to the text in section 4 (Results) for detailed information about these plots.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
Data from a period with snow precipitation when the road surface was covered with snow or wet fluid. Please refer to the text in section 4 (Results) for detailed information about these plots.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
Data from a cold period with snow precipitation. The road condition was snowy or icy and sometimes dry in the wheel tracks. Please refer to the text in section 4 (Results) for detailed information about these plots.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
The readings from the IR camera and the IR thermometer were compared with the Pt100 ground sensors and the road surface state to discover any correlations and connections among the data. The data used in this study were collected during the winter period from February 2010 to April 2010. Three different periods, with varying road conditions, during 2010 were used to produce the data. One reference period, which had dry road conditions, started on 14 March and ended on 16 March. A second evaluation period involved road conditions that varied between snow- and ice-covered lanes to having dry wheel tracks and this started on 21 February and ended on 25 March. The third evaluation period, with road conditions varying from wet to snowy, started on 25 February and ended on 3 March.
b. Detection of outliers
An investigation of the difference between yj,max and yj,min should make it possible to discover those outliers caused by vehicles, as their values differ significantly from the rest of the data. If an individual reading exceeds three standard deviations from the n = 10-min yj,avg, then the reading could be considered as being an outlier, which is most likely to be represented by a passing vehicle. The use of three standard deviations as the limit for the outliers is chosen based on the fact that 4000 vehicles are passing during a 24-h period with an assumed approximate speed of 10 m s−1 and a vehicle length of approximately 5 m. The implication based on this is that the time passage for a vehicle within the IR sensor field of view is 0.5 s. Thus, the total time that vehicles will fall within the IR sensor field during each 24-h period becomes 4000 × 0.5 s = 2000 s. The number of IR sensor records affected by vehicles should then be 2000/(86 400) ~ 2.3%, which corresponds to 2.2 standard deviations. There is the possibility that the outliers could occur for reasons other than for cars; thus, it was felt that the use of three standard deviations was probably an acceptable choice.
4. Results
Figures 3–5 show the data retrieved from the periods mentioned above. These figures have multiple curves on one page to enable readers to follow events at a specific time for different parameters.
a. Summarized IR thermometry results
According to the Swedish Transport Administration, the best practice for road maintenance personnel states that the temperature of the pavement should be quite uniform over the driving area, along the road alignment and road width, and that it is generally the case that the temperature rise in the wheel tracks can be ignored most of the time. Even though only minor temperature differences were observed during this research, namely, up to 1.0°C, it is possible for them to cause very different road conditions. This small temperature difference for the wheel tracks and in between the tracks can cause severe road conditions on parts of the road at which the road temperature is close to the road surface fluid freezing point. Figure 6 shows the IR camera detection areas used for the analysis. An image that shows a car is also supplied to provide an idea of the location of the traffic in the picture.
(top),(bottom) IR camera detection areas are to the left in the picture. From these areas it was possible to determine the temperature in wheel tracks (AR01) and in between wheel tracks (AR02). AR03 defines the driving area. (bottom) A car caught on the IR camera image is shown and these cars could be detected as outliers.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
The road temperature distribution was very uniform after a snowfall, to the point where the road was completely covered with snow. Even a very thin layer of snow appears to provide a good insulation for the pavement.
Wet, slushy, snowy, or icy road conditions caused larger differences in the road temperature, both in the wheel tracks and in between the wheel tracks. The individual readings that were close to each other also tended to differ more than was the case for a dry pavement. During these weather situations, the spatial pavement temperature distribution was less homogenous.
b. IR temperature correlation coefficient
The practices of the road weather community states that a ground temperature at a depth of 2 mm describes the road surface temperature. If this is true, then the ground temperature at 2 mm should correlate well with the IR temperature measurements of the surface. This correlation is high during dry weather conditions (see Table 2), but the correlation of temperatures during weather changes shows a better correlation between the IR and air temperature; see Tables 3 and 4. It is noticeable that the correlation between the temperatures monitored using a ground temperature sensor at a depth of 2 mm and an IR surface temperature sensor is poor during wet, icy, and snowy road conditions. This poor correlation is caused by the coverage of the ground temperature sensor at 2 mm by water, ice, or snow. These materials will act as insulators in relation to the road surface. The ground temperature sensor at a depth of 2 mm will, under these circumstances, behave more like the ground temperature sensor at a depth of 0.3 m, which can be clearly seen in the topmost plot of Fig. 5. All of the correlation coefficients in Tables 2–4 had a p value of zero, which means that the correlations are significant.
Temperature correlation coefficients during a dry reference period.
Temperature correlation coefficients during a winter period with partly snow- or ice-covered road surface.
Temperature correlation coefficients during period with wet and sometimes snowy road surface.
The ground temperature sensor installed at the depth of 2 mm is affected by the surrounding asphalt and does not respond to temperature changes as quickly as the IR camera and IR thermometer that measures the “skin” temperature. This difference has been studied by means of a time series analysis of the two temperatures; see Fig. 7. The time series analysis involves investigating the correlation change as the two temperature readings are time shifted, or lagged, against each other. From Fig. 7 it can be seen that the time difference, or lag, for the highest correlation coefficient of 0.96 is less than 1 min during dry and stable road conditions. For snowy and icy road conditions, the correlation reaches the highest value of approximately 0.3 after 3 h, and for wet and snowy road conditions the highest correlation coefficient is 0.5 after approximately 17 min. These correlations are, however, quite low due to the insulation effects of water, ice, and snow. Further causes for the low correlations are the latent heat caused by water freezing or melting and the heat transfer due to fluid flow.
Cross correlation of surface temperatures retrieved from IR camera and surface temperature from ground temperature sensor at 2-mm depth. The highest correlation occurs approximately at (top) 1 min, (middle) 3 h, and (bottom) 17 min for the different road conditions.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
c. Heat flux
From (8) it is seen that for asphalt with κ approximately 10−7, the time scale for thermal energy to penetrate from the surface to the ground temperature sensor at a depth of 2 mm is approximately minutes, while the time for thermal energy to penetrate to the ground sensor at a 0.3-m depth is approximately one day, as can be seen in Fig. 3. The implication of this is that the ground sensor at 0.3 m may only be used as a measure of a stable road body temperature. The time scale for the 2-mm ground temperature can be verified from the time series analysis plot for dry weather conditions in Fig. 7.
The direction and magnitude of the heat flux from the road surface can be described as the IR surface temperature subtracted from the ground temperature at a 2-mm or a 0.3-m depth.
From Figs. 3–5 it can be seen that the difference between the IR temperature and ground temperature at a 2-mm depth offers a good metric for rapidly changing road conditions, which is not possible from observing the difference between the IR temperature and the ground temperature at a 0.3-m depth.
d. Identifying road conditions by thermal imaging
The temperature difference between wheel tracks and in between the wheel tracks is, according to the reference period, mostly ≥0, which originates from the vehicle tires heating the road surface; see Fig. 3. This is not the case during a period with wet, icy, and snowy road conditions (see Figs. 4 and 5), and this is probably because traffic runs in other wheel tracks during these weather conditions. The conclusion is that it could, potentially, be possible to detect changes in the road condition by investigating differences in temperature, as a negative temperature difference tends to indicate changes from a normal dry road condition.
e. Traffic detection
An IR thermometer, as described in the materials and methods, was used instead of an IR camera to detect traffic. The temperature measurements from the thermometer were stored every second as compared to the majority of the other data, which were stored every minute. This high measurement frequency of the IR thermometer made it possible to detect individual vehicles. The main difference between the camera and the probe is that the latter only measures the temperature in a single and very small spot; hence, the field of view of the thermometer will readily become fully blocked by a passing vehicle. The noise in the temperature signal was processed by searching for outliers using the method described in the methodology (section 3b). The outliers were identified and were assumed to be passing vehicles obstructing the field of view to the road surface, which resulted in a traffic intensity measurement, as shown in Fig. 8. The method will provide a good estimation regarding whether there is more or less traffic on the road, which will have an influence on the road surface temperature. However, the method is not suitable for measuring the exact number of passing vehicles and neither the speed nor type of vehicle.
(top) Temperature outliers corresponding to traffic are represented by a “+.” (bottom) Traffic intensity is integrated over time and given in the unit vehicles per hour.
Citation: Journal of Atmospheric and Oceanic Technology 29, 6; 10.1175/JTECH-D-11-00071.1
f. IR sensing attentions and pitfalls
The sources contributing to the result of the IR sensor are the object radiation, reflected radiation, and atmosphere radiation. Depending on the atmospheric conditions, the temperature readings will be more or less reliable—one of the most difficult situations being under clear-sky conditions, where the low sky temperature will be partly reflected by the measurement surface. The traffic was the immediate observed cause of error in relation to pavement monitoring. On roads with light traffic conditions, this is not a severe problem. Instead, the thermal effects from the vehicles can be used to measure the traffic intensity. However, if the road is under heavy traffic conditions or is congested, then it would be difficult, or even impossible, for the infrared instruments to measure the road surface temperature correctly. Another source that affects the results of the IR measurements is the radiation from sunlight, which can cause sudden temperature rises due to direct radiation from the sun or from reflections. Other causes of IR measurement errors could be the relation between the visible sky area and the area covered by obstacles, a discrepancy in the pavement, local surface repairs, and variations in the road construction (Lindqvist 1987).
5. Discussion
IR thermometry has led to new insights about road surface temperature measurements. In relation to winter road maintenance practice, the temperature distribution of the pavement is often assumed to be uniform along and across a road area for the purpose of simplicity. These measurements have shown that this assumption is true during periods of dry weather conditions, but differences can be observed during changes in the weather. Although there are only minor temperature differences in a road section, they may cause varying road conditions, especially when the road is wet and close to the pavement fluid freezing point.
Currently, forecast models often determine the road condition based on surface temperature monitored by sensors installed in the road surface. This study has revealed that the temperature readings from IR sensors and surface-mounted sensors differ significantly, especially during snowy and icy road conditions. From the temperature curves and heat flow curves, it can be concluded that both the absolute temperature readings and the temperature derivatives differ from those of the surface-mounted probes. An infrared thermometer will be able to detect a sudden reduction in road surface temperature sooner than that of a road-installed sensor and will thus be able to provide an early warning in relation to a critical icing situation. This means that utilizing IR temperature measurements for surface temperature measurements will provide a more precise indication of heat flow compared to that involving the use of traditional surface-mounted temperature probes.
The expectation was that there would be occasions when a wet road surface freezes and that this could be investigated by looking at the temperature rise during freezing, when water releases its latent heat (Jonsson 2009). Unfortunately, this was not observed as the road maintenance obligation is to keep the road at a high friction level at all times. If there is any risk of water freezing to ice, then this will trigger a maintenance action involving salting, which was the case during this study. Further research is required to find occasions when water freezes to ice on the road surface.
It was possible, to some extent, to perform road condition classification by comparing temperatures between the wheel tracks and in between the wheel tracks. It became obvious that the expected rise in temperature in the wheel tracks was suppressed during periods when the road surface was covered with fluid, ice, or snow. This is probably due to the insulation of the road surface from the covering fluid, ice, or snow, which is the point at which the ice and snow is supposed to have the largest insulation capabilities. It was also discovered that the difference between the IR temperature and ground temperature at a 2-mm depth is a good indicator of rapidly changing road conditions.
From the detection of outliers, it was discovered that the traffic intensity could be determined by IR thermometry, as individual vehicles could be observed. Further research is required to verify the IR thermometry vehicle detection accuracy by using traditional traffic monitoring equipment. The use of this traditional traffic monitoring equipment means that it would be possible to register different vehicles and their corresponding axes. This would also allow for there to be a validation of the more theoretical heat flow models described in Prusa et al. (2002).
Another sensor that should be included in future research is a solar radiation sensor. This sensor could assist in describing some of the fluctuations of the IR temperature readings as compared to the ground temperature sensor located at a 2-mm depth.
This paper shows how infrared thermometry together with new and updated models could improve knowledge in relation to the road condition. Current practice using a ground temperature probe for determining the road surface temperature has only been shown to be correct under conditions of dry and stable roadways. To improve the knowledge in relation to road conditions, our research has shown that it is possible to use infrared thermometry together with existing surface- and ground-mounted sensors. Our suggestion is that infrared thermometry should be used in addition to existing sensors, not to replace them. However, a great deal of work still exists to develop commercial products that are able to support winter road maintenance using these principles. This paper may be used as an input for further research and for the development of RWIS.
Acknowledgments
This study was partially funded by the Swedish Transport Administration and the company Combitech AB.
REFERENCES
Bogren, J., Gustavsson T. , and Postgård U. , 2000: Local temperature differences in relation to weather parameters. Int. J. Climatol., 20, 151–170.
Brewster, M. Q., 1992: Thermal Radiative Transfer and Properties. Wiley, 543 pp.
Childs, P. R. N., 2001: Practical Temperature Measurement. Butterworth-Heinemann, 372 pp.
Crevier, L.-P., and Delage Y. , 2001: METRo: A new model for road-condition forecasting in Canada. J. Appl. Meteor., 40, 2026–2037.
Feng, F., and Fu L. , 2008: Evaluation of two new Vaisala sensors for road surface conditions monitoring. Ontario Ministry of Transportation Rep. HIIFP-054, 40 pp.
Jonsson, P., 2009: Road status sensors: A comparison of active and passive sensors. Proc. 16th ITS World Congress and Exhibition on Intelligent Transport Systems and Services, Stockholm, Sweden, ITS World Congress, 14 pp. [Available online at http://miun.diva-portal.org/smash/get/diva2:278005/FULLTEXT01.]
Jonsson, P., 2011: Classification of road conditions from camera images and weather data. 2011 IEEE International Conference on Computational Intelligence for Measurement Systems and Applications (CIMSA) Proceedings, IEEE, 91–96.
Lindqvist, S., 1987: Sensing road surface temperatures with infrared techniques. Göteborgs Universitet Nationalekonomiska Institutionen GUNI Rapport 23, 22 pp.
Manfredi, J., and Coauthors, 2008: Road weather information system environmental sensor station siting guidelines, version 2.0. U.S. Department of Transportation FHWA-HOP-05-026, FHWA-JPO-09-012, 83 pp.
Mikhailenko, S. N., Babikov Yu. L. , and Golovko V. F. , 2005: Information-calculating system spectroscopy of atmospheric gases: The structure and main functions. Atmos. Oceanic Opt., 18, 685–695.
Prusa, J. M., Segal M. , Temeyer B. R. , Gallus W. A. Jr., and Takle E. S. , 2002: Conceptual and scaling evaluation of vehicle traffic thermal effects on snow/ice-covered roads. J. Appl. Meteor., 41, 1225–1240.
Rothman, L. S., and Coauthors, 2009: The HITRAN 2008 molecular spectroscopic database. J. Quant. Spectrosc. Radiat. Transfer, 110, 533–572.
Sakai, I., and Stenson C. , 1990: Optical sensors for inferring road adhesion. Fibre Optics ′90, P. McGeehin, Ed., International Society for Optical Engineering (SPIE Proceedings, Vol. 1314, 236–243.
Sass, B. H., 1992: A numerical model for prediction of road temperature and ice. J. Appl. Meteor., 31, 1499–1506.
Scott, B., Minge E. , and Peterson S. , 2005: The Aurora Consortium: Laboratory and field studies of pavement temperature sensors. Minnesota Department of Transportation Rep. MN/RC–2005-44, 110 pp.
SIRWEC, 2007: A guide to road weather systems. Standing International Road Weather Commission Rep., 83 pp. [Available online at http://www.sirwec.org/en/rwis_web_guide.pdf.]
Tilley, J. and Johanneck L. , 2008: Evaluation of Vaisala Spectro pavement sensor. Aurora Project Summary 2006-04, 53 pp.
