The UK transport sector has been identified as one of six key sectors which will be most affected by climate change (McKenzie Hedger et al., 2000): energy, business, transport, domestic, public services, agriculture, forestry and fisheries. Indeed, the issue of climate change has been identified as ‘…the critical over-arching issue for any transport research agenda’ (Hall, 2010, p. 9). The transport network is a key driver of the economy and in itself is a highly valuable asset. Indeed, the UK road network was valued as the government's single most expensive asset in 2005 with major trunk roads and motorways having an approximate value of £62 billion (Department for Transport, 2005). With this in mind, it is clear that the impacts of climate change on transport networks need careful consideration to allow the networks to continue operating effectively in the future.
Different emission scenarios which may occur in the future as a result of changes in social responsibility, government policies and both government and individual actions, produce a range of predicted changes in climate. As such, there is a range of impacts on transport which have to be assessed. It is widely recognized that weather patterns will change as a result of climate change and it is these changes which will potentially have a big impact on transport. Road networks need to be resilient to changing weather patterns to keep traffic (and the economy) flowing. Transport is negatively affected by heavy rain, snowfall, strong winds, extreme heat and cold, and reduced visibility, which can cause injury, damage and economic loss (Jaroszweski et al., 2010; EWENT, 2011). Hence, it is crucial to understand the potential impacts of future climate change on each transport mode in order to be able to prepare for these effects through adaptation and to provide additional adaptive capacity (Walsh et al., 2007). For example, road surface specifications have already been modified following the hot summer of 1995 to withstand greater temperatures in future and there may be a need to revisit these changes periodically to adapt to a warming future climate (Clarke et al., 2002). Similarly, increased instances of flooding as a result of increasing extreme rainfall events (Alcamo et al., 2007) is another problem which requires adaptation for all transportation modes (McKenzie Hedger et al., 2000). In order to minimize the damage to road infrastructure, adaptation measures, such as improved drainage, will have to be implemented to reduce the impacts of flooding in future. However, in order to project and model impacts via a climate change impact assessment, a clear understanding of impacts in the baseline climate is required (Jaroszweski et al., 2010). This study focuses on the impact of precipitation on the road network, examining how precipitation events affect traffic speeds on sections of the UK motorway network. The aim is to acquire an appreciation of current impacts with a view to understanding how the UK motorway network may respond to future precipitation events within a changing climate.
There has been a number of general studies focusing on the impacts of weather on road transport in a number of countries (Edwards, 1998; Andrey et al., 2003; Golob and Recker, 2003; Al Hassan and Barker, 1999). However, few make the link to climate change. Jaroszweski et al. (2010) provide a summary of climate change impacts on transport, subdividing impacts caused by changes in temperature, precipitation, timing of seasons and extreme events and explains the process involved in conducting a climate change risk assessment. The key starting point in such an assessment is the derivation of relationships or identification of thresholds in the current (baseline) climate. Temperature related impacts have been well studied (London Climate Change Partnership, 2005). However, the impacts of temperature are easy to quantify. Reports such as London's Warming (Clarke et al., 2002) investigate the impacts of a warming climate across a city and examine in depth the potential impacts for the transport sector based on previous work. Many previous studies have shown that friction on road surfaces is reduced in higher temperatures, leading to reduced safety on the roads (Walsh et al., 2007). Temperature is actually a relatively easy parameter to model in climate change risk assessments as clear thresholds of failure exist (e.g. rail buckling; Chapman, 2007) which makes the extrapolation to future climates a relatively simple task. However, such thresholds do not exist for all weather parameters (notably precipitation), which makes the inclusion in climate change impact assessments somewhat difficult.
Of all weather, Koetse and Rietveld (2009) highlight precipitation as the most important variable for impacts on road congestion and traffic safety. Precipitation has several impacts on the road network, causing reduced visibility and friction, as well as increased flooding where drainage is an issue (Walsh et al., 2007; Committee on Climate Change and US Transportation, 2008). Smith et al. (2003) investigated the impacts of precipitation on highway traffic flows in Virginia, US and discovered that both capacity and speed on the highway were affected during a precipitation event, with the possibility of a 5–6.5% reduction in speed during a precipitation event. It is also likely that journey times and delays would increase significantly and traffic congestion would increase along the affected section of road. A similar study by Hranac et al. (2006), also on US roads, identified that free flow speed, capacity and speeds at capacity are all affected during precipitation events of varying intensities. Overall, there is a general consensus that accidents in wet weather are more frequent yet less severe than in fine, dry weather (Edwards, 1998), but the full impact of precipitation on road safety needs further investigation and the relationship needs to be quantified for use in climate change risk assessments. This paper aims to start to address these research gaps by investigating the existence of a failure threshold for precipitation on the UK motorway network.
3.1. Study area
This study analyses the route corridor from London to the England-Scotland border. The corridor starts at Junction 1 of the M1 to Junction 19 (Catthorpe Interchange at Rugby) and then along the M6 to Junction 44, as shown in Figure 1. This study has been undertaken as part of a larger project investigating the impact of climate change on transport using this route corridor. Such an extensive cross country route is important to allow for future analysis of differences between climates and urban and rural sections of motorway and the impacts that may occur in these locations.
3.2. HATRIS data
The Highways Agency Traffic Information System (HATRIS) systematically monitors traffic flow and vehicle speeds across the road network. Information is collected from a number of sources including cameras installed on the motorway network, loops installed in the road surface and vehicles fitted with tracking devices. The main sources of data along the route corridor are MIDAS (Motorway Incident Detection and Automatic Signalling), Trafficmaster, NTCC (National Traffic Control Centre), and ITIS (data from ITIS Holdings plc). Each system records traffic flow and travel time across the motorway links using a different methodology and all are incorporated into the HATRIS data base to provide as much information as possible about the network. Hence, HATRIS is a rich, high resolution, dataset containing flow information for all motorway links across England and Wales. Links are defined as junction to junction sections of the motorway and there are approximately 2500 which make up the entire motorway network. The route corridor in this study comprises 146 operational junction to junction links. Included in the database are details of speed, journey times and traffic flows which can be used for analysis. Significant alterations in the data which correlate with the occurrence of different weather events, in particular extreme weather events will be identified in this study. The variables available for analysis are detailed in Table 1.
Table 1. Variables available for analysis in the HATRIS data set (based on Hardman et al., 2007)
Unique link ID
Date of travel
Day type that the day belongs to, defined by whether the day is a weekday, weekend, bank holiday, school holiday etc
One of 96 15 min periods throughout the day
‘Best of three sources’ Merged-All-Data-Journeys estimate of average journey time (seconds) to travel across the link during the 15 min period
‘Best of three sources’ MADJ estimate of average speed (km h−1) across the link during the 15 min period (calculated using (length/MADJ_avgJT))
‘Best of three sources’ MADJ quality index—high, medium or low
‘Best of three sources’ MADJ data type—indicates whether data is observed or has been in filled from another source
‘Best of three sources’ MADJ vertical row values—average number of records used for vertical infilling (0 indicates that observed data is used)
‘Best of three sources’ MADJ horizontal row values—average number of records used for horizontal infilling (0 indicates that observed data is used)
Source of the travel time data: MD = MIDAS; TM = Trafficmaster; IT = ITIS; FF = freeflow (indicates that there is no observed or infilled data from any of the three sources)
MIDAS estimate of average journey time (seconds) to travel across the link during the 15 min period
MIDAS estimate of average speed (km h−1) across the link during the 15 min time period (calculated using (length/MD_avgJT))
MIDAS quality index—high, medium or low
MIDAS data type—indicates whether data is observed or has been infilled from another source
MIDAS vertical row values—average number of records used for vertical infilling (0 indicates that observed data is used)
MIDAS horizontal row values—average number of records used for horizontal infilling (0 indicates that observed data is used)
Trafficmaster estimate of average journey time (seconds) to travel across the link during the 15 min period
Trafficmaster estimate of average speed (km h−1) across the link during the 15 min time period (calculated using (length/MADJ_avgJT))
Trafficmaster quality index—high, medium or low
Trafficmaster data type—indicates whether data is observed or has been infilled from another source
Trafficmaster vertical row values—average number of records used for vertical infilling (0 indicates that observed data is used)
Trafficmaster horizontal row values—average number of records used for horizontal infilling (0 indicates that observed data is used)
ITIS estimate of average journey time (seconds) to travel across the link during the 15 min period
ITIS estimate of average speed (km h−1) across the link during the 15 min time period (calculated using (length/IT_avgJT)
ITIS quality index—high, medium or low
ITIS data type—indicates whether data is observed or has been infilled from another source
ITIS vertical row values—average number of records used for vertical infilling (0 indicates that observed data is used)
ITIS horizontal row values—average number of records used for horizontal infilling (0 indicates that observed data is used)
Estimate of average journey time (seconds) to travel across the link during the 15 min time period
Estimate of average speed (km h−1) across the link during the 15 min time period (Calculated using (length/MADJ_avgJT)
Quality index—high, medium or low
Data type—indicates whether data is observed or has been infilled from another source
Vertical row values—average number of records used for vertical infilling (0 indicates that observed data is used)
Horizontal row values—average number of records used for horizontal infilling (0 indicates that observed data is used)
NTCC estimate of average journey time (seconds) to travel across the link during the 15 min period
NTCC estimate of average speed (km h−1) across the link during the 15 min time period (Calculated using (length/NT_avgJT)
NTCC quality index—high, medium or low
NTCC data type—indicates whether data is observed or has been infilled from another source
NTCC vertical row values—average number of records used for vertical infilling (0 indicates that observed data is used)
NTCC horizontal row values—average number of records used for horizontal infilling (0 indicates that observed data is used)
Free flow journey time (seconds)
Free flow speed (km h−1) calculated using length/FF_avgJT
Link length (km)
Free flow speed (km h−1)
Flow for the link—number of vehicles over the 15 min time period
Values indicate whether data is observed or infilled
Number of records used for infilling, 0 indicates observed data is used
Flag indicating whether travel time for the link has been unusually high
Field not used at present—set to 0
Adjusted flow (average for count of vehicles (rawflow), time period and day ID) for use when calculating delays
‘Best three source MADJ’ estimate of total delay (seconds) calculated using the formula: totaldDelay = (MADJ_avgJT—(length/freeflowspeed)) × (adjflow)
Recurrent portion of total delay
Non-recurrent portion of total delay
Source from which the data have been obtained
Due to the extensive amount of information available, a data reduction exercise was initially required. Firstly, a single year of data was selected for further investigation: 1 January 2008 to 31 December 2008 for the 146 links (i.e. road sections between junctions) along the route. A single year was selected to enable a preliminary analysis to be performed at the start of the project. Secondly, a series of time slices for each day were chosen for further investigation. The HATRIS data have a high temporal resolution with measurements of journey time and traffic flows taken every 15 min throughout each 24 h period. Journey time is the measured variable across each of the links and from this the average speed across each link is calculated using:
The speed data are then included in the database for each link and it is these data that will be extracted and used for the analysis. Four 15 min time intervals were selected for daily analysis, at 0000, 0800, 1200 and 1800, to allow for analysis of the morning and evening rush hours, plus a daytime and a night time timeslot.
For this study, the Merged All Data for Journeys (MADJ) speed was selected for particular investigation. The MADJ data (covering both journey time and speed) was chosen for its reliability as it is generated using a ‘best of three sources’ approach using Trafficmaster, MIDAS and ITIS data. The recorded value in the database is taken from the source which is considered to be the best available and is used to provide the MADJ journey time. For the motorway network, these data are usually obtained from the MIDAS (although where there are instances of missing data then the ITIS source will be used). Journey time was not selected as the variable to be used in the analysis as each of the links differs in length and so would affect the average time taken to cross each link.
3.3. NIMROD precipitation radar
NIMROD precipitation radar data were obtained from the Met Office archives via the British Atmospheric Data Centre. Around the UK, there is a number of precipitation radars which collect data that are subsequently processed by the NIMROD system at the Met Office. At each site, radar scans are carried out at different elevations to give the best possible estimate of precipitation at the ground surface. NIMROD data has a spatial resolution of 1 km and is available at 5 min intervals, and the units of measurement for precipitation rate are mm h−1 × 32. To correspond with the HATRIS data, NIMROD data for the corresponding time are used to investigate the effects of precipitation both along the entire route and at more local scales where convective precipitation storms are likely to impact significantly on local traffic speeds.
3.4. GIS analysis
GIS was used to overlay the HATRIS and NIMROD datasets to identify any initial relationships between precipitation events and traffic speed. The centroid of each of the 146 operational links along the route corridor was calculated to provide a point where the precipitation rates from the NIMROD data could be extracted for each time slice. GIS maps were then created using the HATRIS and NIMROD data and precipitation values extracted for each of the motorway links (Figure 2).
4. Results and discussion
Linear regression was initially carried out on the data to identify the relationship between traffic speeds and instantaneous precipitation along the route. Although there is an identifiable, but weak, relationship on some days, as shown in Figure 3(a), on others the relationship is almost absent (Figure 3(b)). However, in all cases a clear overall negative trend is present. Based on this preliminary analysis, it is reasonable to assume that a relationship exists, yet it is complex and not clearly defined. Other variables (unavailable in this study), such as traffic volume, road capacity, antecedence and precipitation duration will also need to be taken into account.
At the start of this investigation, it was hypothesized that there were potentially two thresholds of failure on the road network with respect to precipitation:
low to moderate rain, sufficient enough to impact upon visibility and therefore slightly reducing vehicle speeds, and,
heavy rain-inducing flooding on the carriageway and impacting significantly on vehicle speeds.
From the analysis, neither of these thresholds appears to exist in the dataset. However, there is a threshold evident at 0 mm h−1, i.e. between precipitation/no precipitation, with the greatest speeds occurring when there is no precipitation and speeds decreasing when there is precipitation. Previous studies have identified that drivers on UK motorways are aware that there is a need to reduce their speed in wet weather conditions (Edwards, 1999), although most only make a marginal reduction (Edwards, 2002). This awareness of the need to reduce speed may be one factor which contributes to the observed slower speeds during precipitation events over the course of the study period.
In order to test the existence of this threshold, Welch's t-test (assuming unequal variance) was used to determine if there is a significant difference between vehicle speeds with/without precipitation. The results were variable, with some instances showing a significant difference at the ≥ 90% significance level. However, there are also many instances where there is no significant difference between the two data sets. This further reinforces the conclusions drawn from the linear regression analysis that the relationship between precipitation and traffic speeds is complex and requires more investigation to be understood fully. Indeed, over the year of analysis there is little consistency or marked seasonal differences between months (Figure 4). The months with the highest percentage of analyses (all time slices) which are significant at the ≥ 90% level are January (11.34% of analyses significant), May (11.11%) and November (10.78%). In contrast, the months with the lowest percentage of analyses which were significant at the same level were February (1.67%), August (6.6%) and September (6.45%). February 2008 was a dry month, receiving 63% of the average rainfall for the month, based on the 1971–2000 average (Met Office, 2008a, 2008b). December was another drier than average month in 2008 with England again having less than average rainfall at just 65% of the 1971–2000 average (Met Office, 2008c).
The volume of traffic should be an additional significant contributory factor in this relationship. Therefore, to determine whether there is a stronger impact of precipitation on traffic speeds at a certain time of day, analyses at 0000, 0800, 1200 and 1800 were investigated to determine the percentage of instances where the mean speed differed significantly when there was precipitation compared to when there was no precipitation at these time intervals. Table 2 shows the results for each month and time slice. Taking January 2008 as an example, precipitation had a significant impact (at the 90% level or above) on 11.11% of journeys along the route at 0000, 11.54% of journeys at 0800, 4.00% of journeys at 1200 and 15.38% of journeys at 1800. Similarly, the Welch's t-test results for June 2008 show that precipitation had a significant impact on 13.79% of journeys at 0000, 3.70% of journeys at 0800, and 10.34% of journeys at 1800. No journeys at 1200 in June were significantly impacted by precipitation. Figure 5 shows the results obtained indicating the percentage of analyses during each month where the difference in means was significant at the 90% or above level. It was hypothesized that precipitation would have a greater impact on traffic speeds during the busiest times of use on the motorway, where the greatest traffic flows occur. However, from the analysis, there is no clear pattern to suggest that precipitation has a greater impact on traffic speeds at one time of day compared to others. Indeed, taking June as an example, precipitation has the greatest impact at 0000 with 13.79% of analyses showing that speeds when there was precipitation were significant at the ≥ 90% level (Figure 5). As a result of these findings, it is clear that further investigation is needed incorporating traffic flows to develop understanding of this complex relationship between precipitation and traffic speeds further.
Table 2. Percentage of journeys along the route, by month and time, where journey speeds are significantly affected by precipitation
The analysis detailed in this study was carried out initially to determine the impact of precipitation on traffic speeds at the route level, with a view to determining existing thresholds. It is clear that there is no single factor relationship between precipitation rate and speeds across the motorway links, however there is a threshold which often exists between speeds when there is no precipitation and when there is precipitation: the greatest speeds occur when there is no precipitation and speeds immediately decrease when there is precipitation. The relationship between precipitation and traffic speeds is shown to be a complex one which requires further investigation. From the results obtained in this study, it is clear that no universal relationship exists that can be applied at a national level. Instead, further analysis is required at the local level: examining the difference in speeds when there is no precipitation and when precipitation occurs for the same time of day across a whole month (using a Welch's t-test) and also at seasonal time scales. This will allow for identification of those links along the route corridor (or road network), where traffic speeds are most susceptible to precipitation induced changes. These detailed data will then be sufficient as a starting point for a national climate change impact assessment of the impact of precipitation on the road network.
This work is funded by EPSRC ARCC and is part of the FUTURENET project which aims to identify the impact of climate change on the UK transport network (EP/G060762). The authors would like to thank the Highways Agency and the IBI Group for providing the HATRIS traffic data as well as the UKMO (via BADC) for providing the NIMROD precipitation radar data.