Economic, Environmental and Social Impacts of Changes in Maintenance Spend on the Scottish Trunk Road Network
7 Impacts of pavement condition
As part of the study for local roads, related to this study, a review of literature of the wider impacts of changes in maintenance funding was carried out and reported (Transport Scotland, 2012) (Transport Scotland, 2012) (Transport Scotland, 2012) (Transport Scotland, 2012) (Transport Scotland, 2012). To illustrate the extent of that review, Table 7.1 shows the document relevance matrix developed from the literature review undertaken in the study for local roads.
Table 7.1 Document relevance and coverage matrix
Source: (Transport Scotland, 2012) (Transport Scotland, 2012) (Transport Scotland, 2012) (Transport Scotland, 2012) (Transport Scotland, 2012)
7.1.1 General impacts and vehicle operating costs
In general, for countries where roads have historically been built and maintained to deliver smooth travel over the life of the pavement, maintenance decision making has focused on parameters other than road roughness (e.g. travel time impact of roadworks, skid resistance). For countries where rougher roads have been more common, the effects of road roughness have been pivotal in the maintenance decision making framework. Countries where roughness has a more significant role in decision making tend to have networks with a significant proportion of the network with roughness greater than 5 or 6 IRI (International Roughness Index).
The World Bank HDM model (Watanatada, Harral, Paterson, Dhareshwar, Bhandari, & Tsunokawa, 1987) (Watanatada, Harral, Paterson, Dhareshwar, Bhandari, & Tsunokawa, 1987) and other models such as the TRL RTIM (Robinson, Hide, Hodges, Rolt, & Abaynayaka, 1975) were originally developed for roads in developing countries and had a key focus on road roughness impacts on the road user and vehicle operating costs. They have since been upgraded and extended (e.g. the HDM-4 model is used in Eastern Europe and preliminary analyses have been carried out for local roads in England) but the basic conceptual frameworks remain similar. The logic of these models is a link from road condition (predominantly summarised by road roughness) through to road user costs in terms of vehicle operating costs and travel time. These effects are derived albeit at higher levels of roughness than currently experienced on the strategic road networks in the UK using established relationships that, in practice, would be calibrated for the road network under consideration.
Other studies have been carried out elsewhere but the HDM study remains the most widely applicable and reported model for assessing vehicle operating cost changes based on road condition.
Vehicle operating costs are a summation of fuel and engine oil consumption, tyre use, vehicle depreciation and maintenance and repair costs. In terms of modelling impacts of road maintenance, the models depend primarily on road roughness changes as other road conditions (e.g. curvature, rise and fall) will not be affected by changes in maintenance policy, or their impact is second order.
Road roughness does have an effect on vehicle travel speeds in so far as road users will travel at lower speeds on roads which are in a worse condition. The HDM model identified this effect from the studies in the 1970s on experimental road sections in very poor condition and HDM-4 updated the relationships in the 1990s. The model shows variations of between 0.62 and 2.57 km/h reduction in speed per 1 IRI increase in roughness (this is equivalent to changes of around 1.5mm2 in 3m wavelength Longitudinal Profile Variance, LPV, at base ride quality levels of 4 mm2 LPV).
7.1.2 Travel time based on deteriorating road conditions
Little has been reported on how surface conditions affect travel time. Vehicle drivers may choose to drive more slowly over a surface that has deteriorated than they would over a more even surface. However, it has been postulated that with modern vehicles the effects are reduced and vehicle speeds are maintained but with higher operating costs.
An early study (Cooper, Jordan, & Young, 1980) gathered vehicle speed data for three sites in England due to be resurfaced. At two of the three sites the surface unevenness showed little change before and after resurfacing. At the third site a statistically significantly increase in the traffic mean speed levels was seen following reconstruction of the road. The observed increases in the mean speed after resurfacing were 2 km/h for private cars, 2.3 km/h for light goods vehicles, 2 km/h for medium goods vehicles, and 2.6 km/h for heavy goods vehicles.
The World Road Association report, (PIARC, September, 1987), made the following conclusions about the effect of pavement surface condition on vehicle speed:
- An increase in macrotexture and the lower orders of megatexture generally induces the driver to reduce speed; and
- Increases in megatexture and greater roughness, or the incidence of loose gravel or deep snow or mud, frequently have the effect of inducing the driver to reduce speed to below 50 km/h.
Studies in Sweden by (Linderoth, 1981) and (Wretling, 1996) investigated the relationship between road surface condition and travel speed using a sample of resurfaced roads and a control group. They concluded that there was no evidence of reduced speed due to roughness. (Wretling, 1996) described another Swedish study by (Anund, 1992) that investigated the relationship between surface quality (measured in IRI) and vehicle speed. The results showed that there was a statistically significant speed reduction of 1.6 km/h for passenger cars between 3.00 p.m. and 9.00 a.m. if the rut depth increased by 10 mm, and a reduction of 2.2 km/h for an increase of 1 IRI. The corresponding values during 9.00 a.m. and 3.00 p.m. were 1.9 km/h and 3.0 km/h. For trucks with and without trailers, no significant speed reduction with increased roughness or rut depth was found. The results of those studies support a significant reduction in vehicle speed only when road condition deteriorates beyond some critical level that is rougher than the general level of condition of the trunk road network in Scotland.
7.1.3 Road safety impact and accidents
The relationship between skid resistance, site accident risk rating and skidding accident rates is well established in the UK. Many factors influence the rate or risk of accidents, including skid resistance/texture depth, and other road condition factors such as unevenness and ruts (Wilde & Viner, 2001).
An investigation by (Viner, Sinhal, & Parry, 2005) provided comparative friction data over a wide range of surfaces, with a range of skid resistance and texture characteristics. The data also showed that higher risk sites have higher proportions of accidents above a Sideway-Force Coefficient (SFC) of 0.35 than is the case for low risk category sites.
The research also confirmed the necessity of maintaining an adequate level of texture depth to ensure good high-speed friction and the data showed that a texture of at least 0.7mm Sensor Measured Texture Depth (SMTD) was desirable. The results also demonstrated the declining benefits of continuing to increase the texture depth above an adequate level of approximately 1.25mm SMTD.
A large-scale study of the link between skid resistance and personal injury accidents, based on 1000km of road network (Rogers & Gargett, 1991), confirmed the different levels of accident risk for different types of road site and the increase in risk for sites with lower skid resistance.
In general, summarised by (Viner, Sinhal, & Parry, 2005), it has been found that for Motorways, the overall trend with skid resistance is very flat except for the lowest levels of skid resistance. For dual carriageways the results showed there is a statistically significant trend for accident risk to increase at locations with lower skid resistance. For single carriageway non-event sections, the trend was both stronger and more significant and the trend was stronger when considering only wet or skidding accidents. The trend for single carriageway non-event sections showed a continuous increase in accident risk with decreasing skid resistance.
In summary, the literature supports the conclusion that lower skid resistance tends to correlate with an increased accident rates.
7.2 Overview of analysis
Based on the broad experience of the economic impacts of road maintenance noted in Section 7.1 the main impacts to be considered in this study were assessed to be:
- Test any impacts of changes in road surface conditions (summarised by ride quality or road roughness) on vehicle operating costs using the HDM-4 framework;
- Test any impact of changes in road conditions (summarised by ride quality or road roughness) on travel time based on the most relevant UK study (Cooper, Jordan, & Young, 1980);
- Test any impacts of changes in road conditions in terms of skid resistance based on UK models relating skid resistance to accidents.
The following Sections describe each of these analyses.
7.3 Surface conditions and vehicle operating cost
To derive vehicle operating costs from the HDM-4 relationships it is necessary to translate pavement condition (remaining life) into IRI. This was achieved by translating remaining life into 3m LPV and then, a further conversion from 3m LPV to IRI.
7.3.1 Translating remaining life into 3m LPV
Many of the economic impacts (i.e. carbon emissions, vehicle operating costs and travel time) are driven by pavement roughness. As noted earlier, the projected pavement condition data for various pavement budget scenarios was given as the length of network for each road type (Motorway, dual APTR and single APTR) by remaining life: ranging from 0 to 50 years. Figure 7.1 shows the steps taken to translate the projected remaining life data to 3m LPV.
In order to develop a relationship between remaining life and 3m LPV, condition data for 300 sections, 10m in length, of each road type (Motorway, dual APTR, single APTR) were extracted from the Transport Scotland database. Using the methodology for calculating RCI from 10m condition data (Transport Scotland, 2007a) the RCI for each section was calculated. A more recent RCI calculation methodology was available for Transport Scotland but an earlier version used for the Transport Scotland pavement model was considered sufficiently reliable for this study.
From the RCI value for each 10m length a linear regression between RCI and 3m LPV data provided relationships (for each road type independently) suitable for use in this study. It should be noted that since RCI and 3m LPV are not independent variables this methodology was not adopted to prove a statistical correlation, but did yield the simple linear conversion shown in Equation 2 to translate RCI data for each of the different road types to 3m LPV. Some outlier data points were removed from the analysis (i.e. 3m LPV > 40mm2) to improve the correlation.
LPV3m = A (RCI) + B (2)
Table 7.2 shows the parameter values for the relationship, including the correlation coefficient, between 3m LPV and RCI for each road type.
|
Road Type |
A |
B |
R2 |
|---|---|---|---|
|
Motorway |
0.0396 |
0.7506 |
0.6087 |
|
Dual and Single APTR |
0.0397 |
0.8083 |
0.3085 |
Figure 7.1 Outline methodology for converting remaining life to 3m LPV
The remaining life data from the projected network condition was converted to RCI using the relationship shown in Equation 1. The RCI data was further converted to 3m LPV using the relationships shown in Equation 2. Note that the R2 value from the analysis for only single carriageways was particularly poor so the relationship derived for dual carriageway APTRs was used for all APTRs.
7.3.2 Total network travel by condition
To assign the traffic to the network it was assumed that the traffic is distributed evenly over the different lengths of the network in different levels of 3m LPV (i.e. there is no significant avoidance by road users of roads in poor condition or attraction of roads in good condition). This assumption was necessary as the projected condition data does not represent the actual' network (i.e. the projected condition data does not show route information for the lengths of the network in each remaining life condition band). To undertake a more detailed analysis the condition would need to be projected using the actual road network and associated condition data.
Figure 7.2 sets out the steps taken to determine the vehicle km travelled by each vehicle type over the different levels of 3m LPV.
To use IRI in the economic relationships in HDM-4, a further conversion was required to convert the 3m LPV data to IRI. A provisional transformation has been derived in the European FILTER study (Alonso, 2001) and is provided in Appendix D with other default parameters used in this study.
Figure 7.2 Assignment of traffic to network lengths by 3m LPV
7.3.3 HDM-4 analysis
The HDM-4 [5] model includes modules to calculate vehicle operating costs (VOCs) and vehicle emissions and was considered to be an appropriate tool for this analysis. Typically HDM-4 is not used in the UK as the road network is, by international standards, relatively smooth and vehicle operating costs are not sensitive to roughness until the pavement has an IRI of around 4 or 5. Based on this threshold it is only the worst parts, in terms of longitudinal profile variance, of the Scottish trunk road network that will have any impact on vehicle operating costs.
A set of notional 1km road lengths were modelled in HDM-4, using IRI values ranging from 2 to 5.5 in increments of IRI 0.5. All of the modelled road lengths were of asphalt construction, had a width of 7.5m, a Rise and Fall of 10m/km and a curvature of 15 degrees/km. A further set of 3 lengths were modelled with a surface dressing wearing course to investigate the sensitivity of the model outputs to the different material type. No significant changes in the results of the economic analysis were found due to the change in surfacing type and as a result only one surfacing type has been used in this analysis.
Five vehicle types were defined for use in the model (car; light goods vehicle; passenger service vehicle; rigid HGV; and articulated HGV) and appropriate economic data for each vehicle type, based on published data and consultation with TRL experts and the outputs from the literature review. The parameters used for each vehicle type are shown in Table 7.3. For this study, the Rigid 3-axle HGV was assumed to represent OGV1 vehicles and the Articulated HGV to represent OGV2 vehicles.
Note: Costs are economic costs at 2002 prices
HDM-4 was run for each of the vehicles travelling over each of the 1km lengths to evaluate the economic costs associated with each vehicle type.
To take account of vehicle engine efficiency improvements and predicted real growth in the resource cost of fuel, the fuel costs in the HDM-4 outputs were replaced with values calculated from:
- The amount of fuel used given in the outputs from the HDM-4 modelling, modified for each year of the analysis based on the assumed vehicle fuel efficiency improvements given in STAG (Transport Scotland, 2011b) (Transport Scotland, 2011b) and webTAG unit 3.5.6 (Department for Transport, 2011).
- The resource cost of fuel based on the vehicle type, taking into account the proportion of the vehicle type using petrol or diesel and the growth forecast for the resource costs of petrol and diesel given in (Transport Scotland, 2011b) (Transport Scotland, 2011b) and webTAG unit 3.5.6. (Department for Transport, 2011)
The outputs from these analyses for 2010 are shown in Figure 7.3.
Figure 7.3 Variation of vehicle operating costs with road roughness
7.3.4 Analysis results
For each year of the analysis period for each of the three pavement budget scenarios the vehicle km travelled by IRI for each vehicle type and road type, derived using the methodology described in Sections 7.3.1 and 7.3.2, was multiplied by the economic costs per 1000 vehicle km figures for each of the IRI bands used in the HDM-4 analysis described in Section 7.3.3. Parts of the network with an IRI greater than 5.5 were assigned an IRI of 5.5 and parts of the network with an IRI value less than 2 were assigned an IRI of 2. All other parts of the network were rounded to the nearest 0.5 IRI. The results of the analysis are summarised in Table 7.4.
Figure 7.4 shows for the key years in the funding scenarios, that the difference in annual vehicle operating costs between Scenario 1 and the 2 budget reduction Scenarios increases while funding is reduced (i.e. to 2020) and while funding is restored (i.e. to 2025) but the difference then reduces as the funding continues to grow. However, the vehicle operating costs remain higher than for Scenario 1 as the network is in poorer condition for Scenarios 2 and 3.
Figure 7.5 to Figure 7.7 show the variation of annual vehicle operating costs by the different road types for the key years in the funding scenarios. Appendix E contains the results for each year of the analysis.
In summary, these show that for Scenarios 2 and 3, the vehicle operating costs are higher than for Scenario 1 for all 3 road types. By 2030, the vehicle operating costs on Motorways for Scenario 3 are lower than for Scenario 1 and Scenario 2. The increases in vehicle operating costs over the analysis period are similar for all 3 scenarios.
Figure 7.4 Projected annual vehicle operating costs
Figure 7.5 Motorway annual vehicle operating costs
Figure 7.6 Dual APTR annual vehicle operating costs
Figure 7.7 Single APTR annual vehicle operating costs
7.4 Surface conditions and travel time cost
The earlier TRL study (Cooper, Jordan, & Young, 1980) demonstrated that on newly surfaced roads with a good ride quality, expressed by longitudinal profile variance, drivers travel faster than on roads with a comparably poorer longitudinal profile variance (e.g. pavements in need of maintenance). The results of this research were used as the basis of an estimate of the impact on journey time during normal running that would result from the levels of condition predicted for the different pavement budget scenarios. The analysis used the number of vehicle kilometres travelled over the carriageway lengths defined by different levels of 3m LPV condition derived for the analysis in Section 7.3.
The TRL study (Cooper, Jordan, & Young, 1980) showed the average traffic speed increased when a new surface was provided for a road pavement. Conversely, the average traffic speed can be assumed to have reduced between the provision of a new surface and the time for that surface to be replaced.
Table 7.5 shows the assumed average traffic speed changes between new and worn out pavements for the different vehicle types as identified by (Cooper, Jordan, & Young, 1980). For the purposes of this analysis the change in traffic speed for PSV is assumed to be the same as that for OGV1.
The changes in speed shown in Table 7.5 were assumed to vary linearly with changes in 3m LPV (see Table 7.6).
The traffic data available in the asset database did not include a breakdown between cars and Light Goods Vehicles (LGVs). To enable both of these vehicle types to be used in this part of the analysis, the Rigid 2 axle flow was assumed for LGVs.
For the purposes of this analysis, a representative base speed (when the pavements are in new condition) for each vehicle and road type combination was assumed (see Table 7.7).
The travel time for the vehicle kilometres travelled (by vehicle type) on the predicted condition (represented by 3m LPV) less the time travelled on an as new surface was assumed to represent the extra delay caused by the levels of predicted condition. The delay time per vehicle type with the value of time for the vehicle type (see Table 7.8) as given by QUADRO (Highways Agency, 2009) at 2002 prices, was used to give the extra cost of travel time that would be caused by the predicted levels of condition on the network.
Figure 7.8, Figure 7.9 and Figure 7.10 show the effect of the predicted pavement condition on the travel time costs across the network for each of the levels of maintenance budget considered for each road type.
Note: Traffic speed for new pavement surface
Table 7.9 summarises the extra costs of travel time associated with the predicted network condition (compared to travel on a network in new condition) through the analysis period. The costs were calculated using the pavement condition predicted for 2010, 2013, 2017, 2020, 2025 and 2030 with costs interpolated for other years.
The effect of assuming different average traffic speeds to those shown in Table 7.7 was also investigated. Although the changes in speed raised or lowered the level of delay, the pattern of behaviour was unaffected. Appendix E shows the breakdown of the increases in travel time and costs by road type.
Table 7.10 shows the increase in travel time costs from the reduction in pavement budgets compared to the costs expected with the current level of funding for pavements.
Table 7.11 shows the cumulative increases through the analysis period compared with the costs expected from the current level of maintenance funding.
Figure 7.8 Predicted travel time costs (Scenario 1)
Note: Values taken from QUADRO. These do not match directly the values given by STAG and webtag guidance (2002 prices)
Figure 7.9 Predicted travel time costs (Scenario 2)
Figure 7.10 Predicted travel time costs (Scenario 3)
Note: Extra cost = Costs with reduced budget - Costs with current pavement budget
7.5 Skid resistance and accident costs
7.5.1 Methodology
Earlier studies have shown that provision of good skid resistance reduces the risk of accidents on a road network (Parry & Viner, 2005) and (Kennedy & Donbavand, 2008). Using skid resistance data from the asset database, skid resistance over time has been compared with the historic road resurfacing investment (see Figure 3.4).
Based on this analysis, a preliminary assessment of the potential impact of reduced resurfacing budgets on road accidents has been made by:
- Developing a relationship between skid resistance and resurfacing investment;
- Reviewing trends in accidents (total, and wet weather) on the network and assessing the base number of accidents which could be affected by the level of skid resistance;
- Using information from an earlier study (Coyle & Viner, 2009) to identify accident models for the UK trunk road network based on skid resistance.
Section 7.5.2 summarises the results of the above analysis.
7.5.2 Analysis results
7.5.2.1 Step 1: Relationship between skid resistance and surfacing investment
Figure 3.4 shows skid resistance has improved across the network over the last 10 years. It improved most markedly between 2001/02 and 2005/06, since when it has remained relatively constant (if slightly improving). The indexed surfacing budgets during this time, also shown in Figure 3.4 have fluctuated without any strong trend. However, it can be concluded that the average annual budgets between 2001/02 and 2004/05 were higher than during the more recent period. This postulation is demonstrated in Table 7.12.
Notes: 1. The four year period for SCRIM condition data is one year later than the budget data, to make allowance for a lag effect between condition and maintenance work.
2. The indexed summation of the 4 Operating Company spends on pavement surfacing treatments.
From Table 7.12 it has been postulated:
- At an annual budget (Operating Company component) for resurfacing of around £12.7m, the percentage of the network with negative SCRIM deficiency can only be held at the current level of around 20%;
- At an annual budget (Operating Company component) for resurfacing of around £17.8m, the percentage of the network with negative SCRIM deficiency reduced by approximately 16% per year.
From the analysis, it is possible that with further budget reductions, current levels of skid resistance could be maintained. For example, there could be further efficiencies that could be achieved even within the existing budgets. Similarly, it is not certain that improvements in skid resistance between 2002/03 and 2005/06 are attributable only to budget changes. However, based on the available information these statistics have been used to derive a first estimate of potential impacts of the proposed budget cuts.
A linear relationship as shown in Equation 3 has been derived:
APR = 3.2 * OCB - 40.6 (3)
Where APR = Annual percentage reduction in percentage of the network with negative SCRIM deficiency;
OCB = Annual surfacing budget for the 4 Operating Companies (£m)
From the subjective assessment of the funding scenarios (described in Section 6 and Appendix B) the following has been assumed for the funding of surfacing maintenance:
- Scenario 2 (20% budget reduction) will allocate around £13m to the 4 Operating Companies for the first 10 years of the analysis period;
- Scenario 3 (40% budget reduction) will allocate around £9.6m to the 4 Operating Companies for the first 10 years of the analysis period
Using Equation 3, Scenario 2 would lead to skid resistance levels remaining broadly constant, but for Scenario 3 the percentage of the network with negative SCRIM deficiency would increase by 10% of the current percentage per year.
7.5.2.2 Step 2: Accident trends attributable to skid resistance
From a more detailed review of the accident data shown in Section 3.3, a rounded average of the last 5 years of available data (2005 to 2009 inclusive) was calculated. From the number of accidents that occurred only during wet or damp conditions, which are the conditions in which skid resistance becomes a key driver of the number of accidents, the data analysis showed that:
- The number of fatal accidents on all trunk roads in wet/damp conditions has varied from 11 to 21 per year, with an average of around 15; and
- The number of accidents of all severity has varied between 340 and 440 per year, with an average of around 380.
7.5.2.3 Step 3: UK trunk road model application
Using models proposed in the most recent unpublished TRL study (Coyle and Viner, 2009) and a number of related assumptions, an estimate of the increase in the number of accidents due to increases in negative SCRIM deficiency values was made:
- Step 3.1. Network characterised into two populations of positive SCRIM deficiency and negative SCRIM deficiency. Characteristic values of SFC were assigned to the two populations. From the model proposed by Coyle and Viner (2009), the relative change in accident risk was predicted for the two populations;
- Step 3.2. For the current condition of the network, the risk model was applied to the two different populations to generate a total number of accidents equal to the current skid related number of accidents derived in Step 2;
- Step 3.3. The calculated accident risks were used to predict changes in the total number of accidents based on the predicted changes in skid resistance derived in Step 1.
For Step 3.1, the mid-range was assumed for the data described in Section 4 of the study, and was used with equal contributions from the different curves postulated in Figure 6 of the (Coyle & Viner, 2009) report. These assumptions suggest accident risk increases by a factor of 12.5/7.5 = 66% when moving from the population with a positive SCRIM deficiency to the population with a negative SCRIM deficiency.
For Step 3.2, two models for the number of accidents were considered. First, the total number of all wet road related accidents (fatal, severe and slight) and second, the total number of fatal accidents.
The results of the analysis (Step 3.3) show that:
- For Scenario 1 (current budget) and Scenario 2 (20% budget reduction), throughout the analysis period there is no increase in the predicted number of accidents as no change in skid resistance occurs. There is, therefore, no cost impact of the funding reduction in Scenario 2;
- For Scenario 3 (40% budget reduction), the percentage of the network with negative skid deficiency was predicted to increase from the current 20% to around 45% during the first 10 years of the analysis while only the lower budget is available. This would result in a predicted increase in the number of all severity accidents from the current figure of around 380 per year to around 440 in 2019/20. Over the next 5 years, as the budget is restored to the 2010 level, the number of accidents also returns to the 2010 level and is retained at that level for the last 5 years of the analysis period. Similarly, the number of fatal accidents was predicted to increase from the current figure of around 15 per year to around 17 in 2019/20. This would represent an increase in annual undiscounted accident costs from the current figures of around £23m to around £26m by 2019/20 before returning to £23m in 2029/30. The total effect is an increase in the cost of skidding accidents of £29m (undiscounted) over the analysis period.