Cross-country applicability of evaluation methods. A pilot study in Portugal and Germany.

Transcription

1 Road Infrastructure Safety Management Evaluation Tools (RISMET) Cross-country applicability of evaluation methods. A pilot study in Portugal and Germany. Project Coordinator: Deliverable Nr. 6.3 December 2011 SWOV Institute for Road Safety Research Project Partner 2: TUD - Technische Universität Dresden Project Partner 3: LNEC - National Laboratory for Civil Engineering Project Partner 4: TØI - Transportøkonomisk institutt Stiftelsen Norsk senter for samterdselsforskning Project Partner 5: TRL - Transport Research Laboratory Project Partner 6: KfV - Kuratorium für Verkehrssicherheit Page 1 of 88

2 Project acronym: RISMET Project title: Road Infrastructure Safety Management Evaluation Tools RISMET Road Infrastructure Safety Management Evaluation Tools Deliverable Nr 6.3 Cross-country applicability of evaluation methods. A pilot study in Portugal and Germany. Due date of deliverable: Submission date Draft 1: Start date of project: End date of project: (revised 31/12/2011) Author(s) this deliverable: João Lourenço Cardoso, LNEC, Portugal Version: Final Page 2 of 88

3 Executive summary The traffic system and cultural dissimilarities are believed to contribute significantly to regional and country differences in road safety performance. Therefore, caution is required when transferring safety management and intervention tools from one region to another. This report deals with the application of two safety evaluation tools, developed at the Laboratório Nacional de Engenharia Civil in Portugal and at the Technische Universitaet Dresden in Germany, to a set of road stretches in both countries. The procedures analysed are intended for the detection of inconsistent horizontal curves and dangerous nonintersection sites. The main questions investigated are related to the direct applicability of both methods outside the region where they were developed. This was investigated by means of a comparison of the detected danger and non-danger zones identified with each method and the corresponding accident rates and by direct comparison of the danger classifications obtained with both methods. Geometric and traffic data on 42 km of Portuguese roads and 190 km of Brandenburg roads were analysed. Data on traffic and registered accidents refer to a four year period in Portuguese roads (1147 accidents) and a three year period in Brandenburg (126 accidents). The main conclusions are that both methods need further recalibration to local conditions, in order to fully take advantage of their potential. When properly used, both methods effectively assist road designers in detecting high accident risk sites at the design stage; however, they are not so successful at discarding low accident rate sites from further safety analysis. Despite incorporating variables intended to represent driver behaviour, there is still a considerable percentage of high accident rate sites not being identified as deserving further study and safety improvements in both methods, indicating that their effectiveness may be improved. Page 3 of 88

4 List of Tables Table 1 Classification of horizontal curve consistency Table 2 Range of values of explanatory variables Table 3 Horizontal alignment geometric characteristics. IP4 from km to km Table 4 Horizontal alignment geometric characteristics. Analysed subset of Brandenburg road itineraries Table 5 IP4. Consistency class distribution according to the Portuguese method Table 6 IP4. Accident and killed and serious injuries rates per consistency class, in the Portuguese method Table 7 IP4. Ratio between an elements accident rate and the average accident rate of the itineraries, per class of element Table 8 IP4. Sequence classes according to the TU Dresden method Table 9 IP4. Ratio between a sequence s accident rate and the average accident rate of the itineraries, per sequence class (TU Dresden method) Table 10 IP4. Distribution of road elements per class (both methods) Table 11 Brandenburg itineraries. Consistency class distribution of road elements according to the Portuguese method Table 12 Brandenburg itineraries. Accident and killed and serious injury rates per consistency class Table 13 Brandenburg itineraries. Ratio between an element s accident rate and the average accident rate of the itineraries, per class of element Table 14 Brandenburg itineraries. Sequence danger classes according to the TU Dresden method Table 15 Brandenburg itineraries. Ratio between a sequence s accident rate and the average accident rate of the itineraries, per sequence type and class (TU Dresden method) Table 16 Brandenburg itineraries. Distribution of road elements per class (both methods)30 Page 4 of 88

5 List of Figures Figure 1 Inconsistency factor on curves on single carriageway roads with non-paved shoulders Figure 2 Average injury accident rates per type of road element Figure 3 Overall plan view of IP4 from km to km Figure 4 Overall plan view of the Brandenburg road links selected as pilot itineraries Figure 5 Brandenburg road stretch elements Figure 6 Brandenburg itineraries. Danger classes per sequence type Page 5 of 88

6 Table of content Executive summary... 3 List of Tables... 4 List of Figures... 5 Table of content Introduction Portuguese tool for safety evaluation of horizontal alignment Background Geometric consistency Method for evaluation of geometric consistency Basic principles Detection of road elements Estimation of unimpeded speed profiles Injury accident prediction models Inconsistency factors Consistency class identification of horizontal curves Follow-up to the safety evaluation Accident risk and consistency classification The tool for safety evaluation of horizontal alignment developed at the Technische Universitaet Dresden Data collected Pilot road stretch in Portugal Pilot road stretches in Germany Comparison of results obtained with the Portuguese and German pilot road stretches Pilot road stretch in Portugal Pilot road stretches in Brandenburg, Germany Concluding remarks and recommendations Sources Annex 1 Plan view of IP4, between km and km Annex 2 Plan view of Brandenburg pilot road stretches Annex 3 IP4 sequences of road elements Annex 4 Brandenburg sequences of road elements Page 6 of 88

7 1 Introduction ERA-NET ROAD Coordination and Implementation of Road Research in Europe was a Coordination Action funded by the 6th Framework Programme of the EC. The partners in ERA-NET ROAD (ENR) were United Kingdom, Finland, Netherlands, Sweden, Germany, Norway, Switzerland, Austria, Poland, Slovenia and Denmark ( Within the framework of ENR this joint research project was initiated. The funding National Road Administrations (NRA) in this joint research project are the United Kingdom, Germany, the Netherlands, Norway and Portugal. The "ROAD INFRASTRUCTURE SAFETY MANAGEMENT EVALUATION TOOLS (RISMET)" project targets objective A (Development of evaluation tools) of the Joint Call for Proposals for Safety at the Heart of Road Design ("The Call"). This project aims at developing suitable road safety engineering evaluation tools that will support the aims of the Call as described in the Guide for Applicants (GfA) and furthermore those of the Directive for Road Infrastructure Safety Management (2008). These evaluation tools allow the easy identification of both unsafe (from accidents or related indicators) and potentially unsafe (from design and other criteria) locations in a road network. With such evaluation tools estimates of potential benefits at the local and the network level can be calculated and potential effects on aspects such as driver behaviour can be estimated. Such tools empower road authorities to improve their decision making and to implement (ameliorative) measures to improve the road safety situation on the roads. Since evaluation tools rely on good quality data, RISMET aims at reviewing available data sources for effective road infrastructure safety management in EU-countries, linked to a quick scan and assessment of current practices. RISMET also aimed at exploiting results related to the development and use of Accident Prediction Models (APMs) in road safety management. One important issue in infrastructure road safety management deals with the transferability of lessons learned in one geographical area to another, different, one. Regional and country differences in road safety performance, given traffic system and cultural dissimilarities, justify precautionary measures when safety tools are transferred from one region to another. According to some authors, at the strategic level the importance of issues at stake may have been, however, overestimated. Current knowledge on the relations between accident frequency and infrastructure related risk factors is mainly based in cross sectional statistical accident models (safety performance functions, or accident prediction models APM) and in results from before-after studies and in informed expert judgement. Most of these results are obtained through studies in one restricted state or region. Despite this geographic limitation, the associated data collection and processing work involves important investment in both specialized human resources and time. Theoretically, APM resulting from cross sectional studies and from before-after studies aim at incorporating the most relevant accident factors and should represent the underlying mechanisms leading to accident occurrence. Therefore, the case for their general applicability would seem strong. However, research studies have found that different equations can be fitted to safety related data, depending on the geographical data [eg. Reurings et al (2006)]. Several factors may explain this variability: APM do not incorporate all contributing factors, but just a sub-set of the factors analysed in the model fitting study (which in itself doesn t incorporate the universe of contributing factors); the underlying mechanisms assumed while framing the model fitting study are broad simplifications of a very complex traffic process. Also, traffic safety performance varies geographically, both in magnitude and in characteristics. All these issues strengthen the case against the application of APM outside the scope of the geographical Page 7 of 88

8 area to which they were fitted. The development of safety evaluation procedures and tools, other than APM, usually aims at dealing with specific issues, being by character constrained in their scope. With this framework, their general applicability would not be a major concern; however, the success of some tools may generate the wish for their application in other geographical settings. The application of APM and safety evaluation tools outside the geographical areas and traffic systems in which they were developed raises both scientific issues and questions concerning the efficiency of the adaptation results obtained. Cross country APM were developed successfully in two projects within the 4th Research and Development Framework Program, MASTER and SAFESTAR. In the former case accident frequencies in interurban road stretches were estimated for the UK, Sweden, The Netherlands and Portugal, using AADT, unimpeded speed distribution parameters and lane width as explanatory variables and a country dependent scaling factor (Baruya, 1998). In the case of SAFESTAR, APM were fitted to accident data from interurban road curves in two countries (France and Portugal): AADT, speed reduction and maximum speed were the most important explanatory variables in the APM using speed consistency measures; AADT, workload, maximum approach speed and lane width were the most important explanatory variables in APM using workload consistency measures (Cardoso, 1998b). Furthermore, in SAFESTAR a common speed behaviour model in curves was developed, and fitted to Finnish, Greek, French and Portuguese speed and road geometry data on 72 curves; however, efforts to fit a common model for predicting speeds in tangents were not successful (Cardoso, 1998a). Transferability issues are not specific to European road safety tools. Within the scope of the development of the recently published US Highway Safety Manual considerable efforts were made to present a procedure to calibrate basic APM to local conditions (AASHTO, 2010). In this report a description is made of the results arising from the application of the consistency safety evaluation tools developed by LNEC and the Technische Universitaet Dresden to a common set of German and Portuguese road stretches. A description of the TU Dresden method and results from its application are presented in another WP4 RISMET report [Dietze, et al (2011)]. In the next two chapters descriptions of the methods developed at LNEC and TU Dresden are provided; chapter 4 is dedicated to describe the road characteristics, traffic and accidents data used; in chapter 5 results from the application of both methods to a set of pilot roads in Germany and Portugal are presented; and in chapter 6 a comparison of results from both methods is provided, with a discussion on their practical implications. In this final chapter conclusions and recommendations are presented, with a mention to research needs in this area. 2 Portuguese tool for safety evaluation of horizontal alignment 2.1 Background The method to evaluate the safety of Portuguese single carriageway rural roads was developed as a result of research carried out within both the Research Programme of the Laboratory for Civil Engineering (LNEC) and the SAFESTAR project (which was part of the European Research Programme TRANSPORT of EU s Fourth Framework Programme), as described by Cardoso (2001). It was developed by LNEC for the National Road Administration and it may be applied both the maintenance stage of existing roads and the design stage of new and redesigned roads. Page 8 of 88

9 This method involves the determination of the consistency class for each horizontal curve on a road, based on speed profiles, geometric characteristics and estimated variations in the expected number of accidents along the analyzed road stretch. Depending on the stage, two different systematic treatments are defined for each consistency class: changes in the geometric design; and, in the case of existing roads, the application of a systematic treatment - involving such low cost engineering measures as marking, signing and shoulder improvement - to the curves of each class. 2.2 Geometric consistency Geometric consistency may be defined as the agreement between the characteristics of the geometric design of a road and the unfamiliar driver s expectations [Fitzpatrick, et al (1999)]. Expectancy is the tendency of a driver to react to a situation, an event or a set of information in a systematic way, based on his/her past experience. Driver expectancy and geometric consistency are important concepts in safety and road design, because inconsistencies on a road can surprise drivers and lead to errors that increase accident risk. When driver s expectancies are violated, the probability that a situation will be correctly identified is significantly reduced. The incorrect identification of a situation greatly reduces the time available for executing the manoeuvres needed to successfully deal with it. Several methods for representing driver expectancy and for evaluating the design consistency of a road were developed [Cardoso et al (1997)]. Some were derived from geometric indices directly related to design characteristics of the road layout [Polus and Dagan (1987) and DfT(1984)]; Fitzpatrick, et al (1999), Krammes et al (1994) and Messer et al (1979) used methods requiring a subjective or objective estimation of driver workload; and the most used methods were based on selected parameters of the unimpeded speed distribution (mainly the average and the 85th percentile) and on their variation along the road alignment [Fitzpatrick, et al (1999), Leisch et al (1976), Lamm et al (1988)]. A procedure for estimating the unimpeded speed profile along the road is needed to apply this last type of method. Unimpeded speeds are observed under very low traffic volumes and are assumed as a good proxy for driver behaviour. Several methods for evaluating design consistency were directly related to accident risk using statistical models [Krammes et al (1992), Wooldridge (1994), Ottesen and Krammes (1994) and Cardoso (1996)]. It was concluded that it is possible to enhance the goodness of fit of accident frequency estimates as a function of road characteristics, by incorporating in the empirical models explanatory variables related to driver behaviour (such as speed reduction or the average speed) or to driver workload. However, the quality of these predictions is strongly dependent on the accuracy of the calculated speed and workload parameters, calling for considerable care in the adequacy of the models used for their estimation. 2.3 Method for evaluation of geometric consistency Basic principles The method developed for consistency evaluation on Portuguese roads of the National Road Network (NRN) is based on four principles: a) The observed driver behaviour on Portuguese roads, drivers expectations and the difficulty they have in the execution of the required manoeuvres are explicitly represented by suitable variables. b) Correlations between the injury accident risk in Portuguese roads and explanatory variables related to driver behaviour and the road geometry (APM fitted to Page 9 of 88

10 Portuguese data), are taken into consideration. c) The deceleration rate required to reduce speed from its value on the approach tangent to its value on the curve should be less than 2 ms -2. d) The variation in kinetic energy required to decelerate from the approach speed to the speed on the curve shall be weighted in the consistency evaluation, as a measure of the severity of the difficulty created a possible accident. The practical application of the method involves the following steps: a) the division of the road in curved and straight elements (curves and tangents); b) the calculation of the unimpeded speed profiles of the road (one for each travel direction); c) the estimation of the increase in accident risk on each curve (as related to the expected accident risk if it was a tangent); d) the calculation of the required deceleration rate on the approach to each curve (in cases where the normal 0.8 ms -2 is not applicable). The method may be applied using a computer program developed at LNEC [Cardoso (2001) and (1997)] Detection of road elements The safety evaluation starts with the identification of curve elements and tangents in the road stretch. According to the Portuguese road design standards, horizontal curves consist of a circular arc (with constant radius) in between transition arcs, at each end. Transition curves are defined by a clothoid. Transition arcs are not needed if the radius of the circular arc is such that there is no need for superelevation (2500 meters on single carriageway roads and 5000 meters on dual carriageway roads). Curve elements include the full circular arc of horizontal curves and 2/3 of the clothoid arc at each end of the circular arc; tangents contain the full straight road stretch and 1/3 of each adjoining clothoid arc Estimation of unimpeded speed profiles Unimpeded speed profiles are calculated using a procedure in line with the methods proposed by Leisch (1976), Lamm (1988) and Krammes (1994). The method is fully explained in Cardoso (1997), for roads with non-paved shoulders, and Cardoso (1998c), for roads with paved shoulders. The following hypotheses are assumed in the speed profile calculation algorithm [Cardoso (2001)]: a) Unimpeded speed choice depends on the horizontal alignment characteristics (sequence of horizontal curves and tangents); b) Unimpeded speed choice is well represented by empirically derived speed models, having road horizontal and vertical alignment characteristics as explanatory variables; c) Unimpeded speed effectively adopted in a selected element depends solely on that element s characteristics, on the speed adopted in the previous road element and on the speed that will be adopted in the following road element; d) Speed is constant in curve elements, except in the case where the entrance speed is lower than the corresponding unimpeded speed; Page 10 of 88

11 e) Speed changes are carried out in tangents only (except for the cases mentioned in the preceding point); f) Speed changes are executed with a constant acceleration/deceleration of +/ 0.8 ms -2 ; however, in some sequences of curve elements on existing roads, tangent lengths may be too short to accommodate a speed reduction with that deceleration rate; in those cases, the needed speed reduction is assumed to take place along the existing tangent length, and the corresponding deceleration is assumed to have been adopted; g) Speed profiles are calculated for each travel direction, dependent on the sequence of road elements in that direction. The following equations are used for estimating the unimpeded speed on tangents (V T ): V V On roads with non-paved shoulders 1 : = S LT DECL LF +1.7 LS (1) T On roads with paved shoulders: = S LF RPC (2) T Where: S L T DECL L F L S R PC - Average road bendiness along 500 m preceding the tangent (degrees/km); - Tangent length (m); - Average hilliness along 500 m preceding the tangent (m/km); - Carriageway width (m); - Shoulder widths (average of both shoulders) (m); - Radius of the curve preceding the tangent (m). Average bendiness and hilliness represent the influence of geometric characteristics previously encountered along the road in drivers speed choice at a given section. Average road bendiness is defined as the sum of the curve deflection angles along 500 metres preceding the tangent; average hilliness is the sum of the vertical height changes (uphill plus downhill) along 500 metres preceding the tangent. Unimpeded speeds on curves (V C ) are estimated using the following equations: Where: V V On roads with non-paved shoulders: = LC LF V MR (3) RC C On roads with paved shoulders: = LF V MR (4) RC C 1 Note: on some Portuguese roads (especially the older ones) stabilized soil shoulders are provided, which may be used to regain control of errant vehicles and to park broken vehicles outside of the carriageway. Nevertheless, these shoulders are not sealed with a bituminous or cemented. Page 11 of 88

12 L C - Curve length (m); R C - Curve radius (m); V MR - Average unimpeded speed on the approach tangent (km/h). The fitting process for equations 1 to 4 is described in Cardoso (1996, 1997 and 1998a). Models for tangents explain over 75% of the observed variation in unimpeded speed (pseudo-r 2 equal to 0.76 on roads with non-paved shoulders and 0.81 on roads with paved shoulders); models for speeds on curves have a slightly better fit (pseudo-r 2 equal to 0.92 on roads with non-paved shoulders and 0.88 on roads with paved shoulders). The models were fitted to unimpeded speeds values between 64 km/h and 149 km/h on tangents and 39 km/h and 142 km/h on curves. The models are valid for the following range of values: curve radii between 20 and 2000 meters; element length between 30 and 5500 meters: carriageway width between 5.1 and 8.9 meters; shoulder width between 0.0 and 5.4 meters; average bendiness between 0.0 and degrees/km; and average hilliness between 2.3 and 61.0 m/km. Calculation algorithms use a maximum value of 2000 m, for an element s length and radii. Recently, studies on unimpeded speed characteristics in Portuguese roads were carried out by LNEC in 2000, 2002 and 2004, and summarized in Cardoso and Andrade (2005). Measurements were made in several tangent sections, selected to be representative of the whole country. The results showed that, under current levels of enforcement, the average unimpeded speed on tangents is 105 km/h on roads with paved shoulders and 97 km/h on roads with non-paved shoulders. In the calculation of speed profiles these values are used as maximum for the corresponding road type. The basic methods proposed by Leisch and others for the estimation of speed profiles are well adapted to roads designed according to modern standards. However, their use to calculate unimpeded speed profiles on existing roads required the introduction of a modification to the original procedure. In fact, on some existing roads there are cases where the length of the tangent between two curves is not enough to accommodate the required speed reduction at the assumed standard deceleration, as referred in Cardoso (1997). In these cases, it is assumed that the deceleration rate will be higher than the standard value. The exact value is calculated by the program PERVEL, developed at LNEC for automatic consistency evaluation. When the value of this deceleration rate is lower than -2 ms -2 the curve is marked as an inconsistent curve (class D ) Injury accident prediction models APM were developed for estimating expected injury accident frequencies on curves and on tangents on roads with paved and non-paved shoulders, as reported in Cardoso (1998c and 2001). The resulting equations for curves and tangents were combined in equations for calculating the increase in accident risk at a curve, as related to the expected accident risk in a tangent with the same geometric characteristics (except for curvature). The following equations were defined: For roads with non-paved shoulders: TAc VRAC = TAc Curve Tangent = e For roads with paved shoulders: ( V ) S LF V MR (5) AADT LC Page 12 of 88

13 VRAC = TAc TAc Curve Tangent = e ( V ) V MR (6) LC AADT L F Where: VRAC - Variation in the injury accident rate (risk) due to the horizontal curvature; TAc Curve - Accident rate on curve (injury accidents per million vehicle-km); TAc Tangent- Accident rate on tangent (injury accidents per million vehicle-km); V L F AADT V MR S L C - Maximum reduction in the unimpeded speed at the beginning of the curve (both directions), negative values being taken as null values (km/h); - Carriageway width (m); - Average annual daily traffic - (vehicles); - Unimpeded speed on the preceding (approach) tangent (km/h); - Average road bendiness in the 500 m preceding the initial section of the curve (degree/km); - Curve length (m). These equations are not intended as cause-effect relations, especially in what concerns the variables representing geometric characteristics. For instance, AADT being in the denominator may simply indicate that there is a relation between the overall quality of the road layout and AADT. Therefore, attempting to use them as a guide for setting corrective actions on Portuguese NRN roads is strongly discouraged. Similarly, it is believed that, in the equation for roads with non-paved shoulders, carriageway width (L F ) is in the numerator because wider roads are more frequent in flat areas or as new roads. On the other hand, narrower roads are more frequent in hilly areas or as old roads. Consequently, curves are more unexpected in wider (and less bendy) roads than in the narrower ones. Moreover, speeds are higher on wider roads (see equations 1 and 2), making it more difficult to avoid dangerous roadside obstacles in case of loss of vehicle control. This is especially true if obstacle free zones are not wide enough and if the space between the carriageway and the obstacles has low skidding resistance, as is the case for roads with non-paved shoulders Inconsistency factors It is well established that the severity of the consequences of an accident is not linearly related to the speed of the crashing vehicles; also, the relation between the kinetic energy of an object and its speed is quadratic. An attempt to include these issues in the consistency rating was made, with the definition of a factor weighting the expected increase in accident risk (VRAC) with a measure of variation in kinetic energy required from the approach tangent to the curve. This factor, designated inconsistency factor (FH), is calculated as a function of VRAC and the kinetic energy at the approach and in the curve, and standardized in order to obtain FH equal to one when speed variation (usually a reduction) is zero. As an example, Figure 1 presents the variation of FH as a function of the average unimpeded approach speed and of the maximum reduction on average unimpeded speed, on a curve 200 m long with average approach bendiness of 32 º/km. The curve is on a single carriageway road with a AADT of 3000 vehicles, 7.50 m wide carriageway and 5.0 m wide non-paved shoulders (2.5 m wide shoulder on each side of the road). Solid lines are used to represent FH and wide broken lines are used to mark the borders of the consistency classes defined in Table 1. Page 13 of 88

14 9 Inconsistency Factor (FH) Maximum reduction of average speed (km/h) 70 km/h 80 km/h 90 km/h 100 km/h 110 km/h Average approach speed (km/h) Figure 1 Inconsistency factor on curves on single carriageway roads with non-paved shoulders Consistency class identification of horizontal curves Curves are divided in five consistency classes, depending on their FH and on the values of speed reduction on their approach and the expected deceleration rate, according to the criteria summarized in Table 1. The highest consistency class (class O ) corresponds to the simultaneous verification of a speed reduction not greater than 5 km/h and FH not greater than 2.5 on roads with paved shoulders or 1.5 on roads with non-paved shoulders and a deceleration rate greater than -2 ms -2. Consistency Class * Table 1 Classification of horizontal curve consistency Speed Reduction Deceleration Inconsistency Factor (FH) Type of Road Paved shoulders Non-paved shoulders O 5 km/h A > - 2 ms 2 B > 5 km/h C D - 2 ms 2 > 8.0 > 6.0 * For classes O to C, all three criteria ( Speed reduction, Deceleration and FH ) must be fulfilled. For class D, Speed reduction and only one other criterion ( Deceleration or FH ) have to be satisfied. Page 14 of 88

15 Curves in the lowest consistency class (class D ) present a speed reduction greater than 5 km/h and match one of two conditions: a deceleration rate below -2 ms -2 ; or an FH greater than 8.0 on roads with paved shoulders (6.0 on roads with non-paved shoulders). For intermediate consistency classes ( A, B, and C ) only FH values are taken in consideration; in these classes speed reduction is above 5 km/h and deceleration is above -2 ms -2 (gentle deceleration). Speed reduction and difficulty of the driving task ratings for each consistency class are described in the next Section, below. Using the computer program PERVEL, it is possible to automatically evaluate the consistency class of each curve on a road. 2.4 Follow-up to the safety evaluation Recommendations regarding the consistency classification of horizontal curves were set. These relate to the design of new highways and the redesign of existing highways to meet current standards and safety requirements. At the design stage of new roads, curves of consistency classes B and C are only accepted on low order roads, based on space constraints or budget (economic) considerations; curves of consistency class D are not acceptable at this stage. In the redesign of existing roads, consistency class D curves must be corrected to the safest level (within budget restrictions), by means of changes in their geometric alignment (for instance, by increasing the curve radius) or in the road alignment at their vicinity to achieve a reduction in the approach speeds. Road alignment at the vicinity of consistency class B and C curves should (but don t have to) be corrected; in these cases, if redesigning is not cost-efficient appropriate signing may be applied. Road signing is intended to create a predictable and efficient traffic operation. Therefore, signing should inform drivers of possible risky situations ahead, guide them through the less risky or easier paths, and induce them to adopt adequate driving behaviour and to raise their attention to a level that is appropriate to the manoeuvres and to the complexity of the driving tasks to be performed immediately ahead. To help drivers develop an adequate set of a priori expectancies related to road curves on the NRN, the Portuguese system for signing road curves was defined, on the basis of the consistency classification described in the previous section. The system is quite similar to the one proposed in the Workpackage 6 - Signing of road curves of SAFESTAR Project, as described by Nielsen et al (1999). It comprises four different sets of signing devices (delineators, vertical signs, markers and road marks) to be applied systematically on the curves of each consistency class, as presented in Cardoso and Roque (2000). 2.5 Accident risk and consistency classification The described system was developed using road, traffic and accident data corresponding to the periods of and Following the start of the implementation of the signing system, new data on road characteristics and accidents became available, raising the possibility for testing the classification criteria with a new set of data. The method of safety evaluation was applied on 1060 km of single carriageway highways (4151 road elements), with data for the period (3896 injury accidents, resulting in 392 fatalities). Figure 2 presents the average accident rate for tangents and for each curve consistency class. Page 15 of 88

16 Accidents per million vehicle.km Tangent Class 'O' Class 'A' Class 'B' Class 'C' Class ' D' TYPE and CLASS of ROAD ELEMENT Figure 2 Average injury accident rates per type of road element Overall, at curves the average accident rate is 25% higher than the average accident rate on tangents (0.35 accidents per million vehicle-km on curves and 0.28 accidents per million vehicle-km on tangents). The accident rate on curves increases significantly with the decrease in their consistency class: the average accident rate on class D curves is almost four times the average accident rate on tangents. Class O curves present an average accident rate that is slightly lower (-10%) than the average rate for tangents. This may be due to the fact that the percentage of highway length where passing manoeuvres are not allowed is higher on curves than on tangents. Also, curves in this class only require minor changes in driver behaviour, as compared to the one adopted in the preceding tangent, which may lead to greater attention on curves, for the same level of driving task difficulty. Generally, the classification obtained seems to reflect rather well the increase in accident risk. 3 The tool for safety evaluation of horizontal alignment developed at the Technische Universitaet Dresden In this chapter a brief description of the method developed by the TU Dresden is presented; a more detailed description is provided in another RISMET report (Dietze et al, 2011). The method developed at the TU Dresden is intended for the detection of safety design issues related to driving behaviour as described by speed choice and its variation along a route. Thus, it is directed for the evaluation the risk of two specific types of accident: single vehicle accidents and overtaking accidents. Application of the method involves several steps for detecting sequences of alignment elements, applying APM for estimating injury accident frequencies and accident cost rates, and classifying each sequence according to both the number of injury accidents and the accident cost rates, by comparison with reference values. Alignment sequences are detected on the basis of speed profiles, each driving direction being analysed separately. Speed profiles are preliminarily calculated using a speed model developed by Lippold (1997): Page 16 of 88

17 V = W R R R (7) R 85 Where: V 85 W R - 85th percentile of speed distribution (km/h); - Carriageway width (m); - Curve radius (m). Afterwards, the preliminary speed profiles are adjusted to reflect speed variations along each travel direction, considering that the required decelerations and accelerations are performed with a constant value of 0.8 m -2. Sequences are detected using the following procedure (Dietze et al, 2011): 1) Each element that causes a speed reduction of at least 10 km/h and is longer than 40 m is set to the status Single Curve. Elements shorter than 40 m are excluded from the analysis because they are too short to be perceived as an element at all. 2) Each single element detected in step one is further analysed to see if there are any other curves within a distance less than 200 m. If so, these elements are aggregated and are characterised by the minimum radius and the curvature change rate. They are set temporarily as Connected Single Elements. 3) In the last step of sequence detection connected single elements are analysed to check if they assemble a sequence that is longer than 250 m. If so, their status is changed to Curved Sequence, which is characterised by the curvature change rate (CCR). Otherwise they keep the status Single Curve and are characterised by the minimum radius. Detected single elements already detected in prior steps keep their status. 4) All other elements that were not analysed so far because their impact on driving behaviour is insignificant are classified as Straight section, characterised by CCR. They consist of tangents and less curvy sequences. As a result of the mentioned procedure, each travel direction of the road is described as a set of sequences of the following types (Diettze et al, 2011): Single curve (type 1 in annexes 3 and 4): a single curve that causes a minimum 10 km/h speed reduction; Curved sequence (type 2): a sequence longer than 250 m consisting of two or more single elements that require a minimum speed reduction of 10 km/h, Straight section (type 3): a series of tangents and curves which have no significant impact on speed. Injury accident frequencies and accident cost rates for each detected sequence of a road stretch are estimated using APM developed in RISMET project (see Dietze et al, 2011). To estimate injury accident frequencies, four equations were fitted, as presented below. For single curves: AF AF SE SE = AADT = AADT ( dv Lprior ) L e and L prior 600m (8) L e and L prior > 600m (9) ( dv ) Page 17 of 88

18 For curved sequences: AF SES = AADT L e (10) ( CCR ) For straight sections: AF ES = AADT L e (11) ( CCR) Where: AF SE AF SES AF ES AADT dv CCR L L prior Injury accident frequency of a single curve; Injury accident frequency of a curved sequence; Injury accident frequency of a straight section; Average annual daily traffic (vehicles); Speed reduction (km/h); Curvature change ratio (gon/km); Length of the sequence (m); Length of the prior sequence (m). Accident cost rates are estimated using the equations presented below. For single curves: 2 ACR = dv dv and L prior < 500m (12) 2 ACR = dv dv and L prior 500m (13) For curved sequences: ACR = CCR (14) For straight sections: ACR = ln( CCR) (15) Where: ACR Accident cost rate of a sequence (in Euros per 1000 vehicle-km). Equations 8 to 15 were developed for a set of element/sequences characteristics described in Table 2. Page 18 of 88

19 Table 2 Range of values of explanatory variables Equations 8 and 12 9 and and and 15 Value AADT (vehicles) dv (km/h) CCR (gon/km) L (m) L prior (m) minimum maximum minimum maximum minimum maximum minimum maximum In a final step each single sequence is classified by comparing its injury accident frequency (the number of injury accidents) and accident cost rate with corresponding reference values for the road category. The result is calculated as a percentage of the relevant reference value. Three safety classes are considered: not critical, semi-critical and critical (Dietze et al, 2011). 4 Data collected 4.1 Pilot road stretch in Portugal The IP4 road was selected as the pilot stretch in Portugal. It is a main trunk road of the Portuguese National Road Network (NRN) located in the north and establishing the main link between the inland populations and the coastal cities. From km 0 to km 58.8, the route is a motorway; the studied stretch, from km 58.8 to km 101.9, is a single carriageway road. The studied stretch of road is located in a mountainous area (Figure 3 and Annex 1). It is a two lane single carriageway road with an additional lane for slow vehicles on almost all its ascending gradients. The generic cross section is formed by a 3.5 m lane in each direction, an additional 3.0 m lane in alternate directions for slow vehicles, and paved shoulders of 2.5 m (one lane per direction) or 0.5 m (dual lane per direction). Figure 3 Overall plan view of IP4 from km to km Page 19 of 88

20 The horizontal layout consists of 147 curves and 135 tangents (Table 3). Horizontal curves account for 69% of the stretch s length, and a considerable number of those have small radius. Approximately 66% of the curve radii are smaller than 420 m and 40% smaller than 240 m. These values are specified as the Minimum Absolute Radius for 100 km/h (420 m) and 80 km/h (240 m) design speeds in the Portuguese Road Design Standards (JAE, 1994). Furthermore, most elements (both curves and tangents) are short, making the horizontal layout sinuous. All horizontal curves have spiral curve transitions. Regarding the IP4 longitudinal profile, 65% of the road length has gradients steeper than 5%; and a particular section of 11.5 km has 95% of its length with a continuous gradient higher than 6%. Additionally, the majority of the vertical curves have radii smaller than the minimum stipulated by the Portuguese road design standards. Table 3 Horizontal alignment geometric characteristics. IP4 from km to km Type of element Number of Length (m) elements Total Minimum Average Maximum Average radius (m) Tangent R> <R <R <R R The studied stretch has a flexible pavement along its entire length. To apply both safety evaluation methods, the following data on geometric characteristics were collected: Start and end kilometre of each tangent, circular curve and transition curve; Curve radius or clothoid parameter; Start and end of each ramp and longitudinal curve; Ramp grade; Curve radius for each longitudinal curve; Carriageway and shoulder widths; Average road bendiness in the 500 m preceding the initial section of each element (per direction); Average hilliness in the 500 m preceding the initial section of each element (per direction). Traffic and accident data were collected for the period beginning in January 2002 and ending in December The Average Annual Daily Traffic (AADT) at km 86.6 was vehicles in 2005 (similar values were observed at km 67.6); at km 96.8, near a major city and the node connecting to the IP3 (another trunk road), it was vehicles. According to Cardoso et al (2006), the average value of vehicles per day may be considered for the analysed stretch. Heavy Page 20 of 88

21 goods vehicles (HGV) accounted for a considerable percentage of traffic: the yearly average was 8.8%. From August to December this ratio was lower, due to a higher number of cars travelling for holidays. The HGV percentage also varied considerably during the week: 10.5% during business days and 4.5% on weekends. During the period 2002 to 2005, a total of 1147 accidents occurred in the analysed stretch, resulting in 52 fatalities, 75 persons being seriously injured and 616 persons slightly injured. A total of 753 property damage only accidents were registered. Only accidents outside of intersections and interchanges were considered. According to Portuguese official definitions, during the period analysed road fatalities were not registered using the international definition death within 30 days of the road accident, excluding confirmed suicides and natural deaths (EC, 2011). In fact, only deaths at the accident scene or during transport to hospital were registered as resulting from road accidents. However, as regards the definition of serious injury, the current EU definition a non-fatally injured victim hospitalized during at least 24 hours was already used by Portuguese officials. By 2004, the whole IP4 road was infamous for its bad safety performance. In itself, the analysed road stretch (15% of the total road length) accounted for 67% of all IP4 registered accidents. The majority of accidents (72%) occurred at horizontal curves or were related to them. Road design, traffic characteristics (speeds) and prevailing weather played a role in the IP4 safety performance. Most of the registered accidents (75%) happened when the pavement was wet; the injury accident rate in rainy months was twice the corresponding figure for the rest of the year. Head-on collisions were the most serious accident type, representing 57% of all fatal and serious injury accidents. This can be explained by driver behaviour on the IP4 road, especially regarding speed choice, as a significant proportion of head-on collisions was preceded by loss of control. Detailed analysis of all injury accidents that occurred between 2002 and 2005 showed that in several dangerous curves a high percentage of uncontrolled vehicles (40% to 87%) invaded their inner roadside area and crashed there with obstacles and other vehicles. 4.2 Pilot road stretches in Germany A subset of 23 road stretches from the road network of the German state Brandenburg was selected for analysis. The TU Dresden geographical information system, described in Dietze and Weller (2011), was used to collect the relevant data. In Figure 4 an overall view is presented of the selected itineraries, highlighted in red (the dark blue dots being network nodes and the light blue lines other road itineraries, not selected for this analysis). Detailed views of each itinerary are presented in Annex 2. Page 21 of 88

22 Figure 4 Overall plan view of the Brandenburg road links selected as pilot itineraries Federal highways B Bundesstraßen (no autobahns) and secondary rural roads ( Landesstraßen ) are included in the analysed subset, with a total length of km. This length includes rural stretches as well as urban areas. Only road stretches in rural areas were considered in the analysis, totalling 125 km of tangents (295 elements) and 36 km of horizontal curves (288 elements). To this end, some itineraries had to be split into two or more road stretches, as shown in Annex 2. A summary of the main horizontal alignment characteristics is presented in Table 4. Table 4 Horizontal alignment geometric characteristics. Analysed subset of Brandenburg road itineraries Type of element Number of Length (m) elements Total Minimum Average Maximum Radius (m) Tangent R> <R <R <R R To apply the safety evaluation methods, the following data on geometric characteristics were collected in the TU Dresden GIS: Start and end kilometre of each tangent, circular curve and transition curve; Curve radius or clothoid parameter; Page 22 of 88

23 Carriageway and shoulder widths. No data were available as regards the average bendiness and the longitudinal profile characteristics, so general assumptions on their value had to be made, in order to apply the Portuguese method. Bendiness was assumed equal on all elements, the selected value being equal to the average CCR value for the selected road stretches (80 degrees per kilometre). Overall, the Brandenburg area considered is quite flat; therefore, a low value (20 m/km, or 2% grade) was assumed for standard hilliness. Shoulders are not paved. AADT on each road element were collected in the road database. Accidents for the 3 years period from 2005 to 2007 were collected. Accidents at intersections, involving turning vehicles (to the road or off the road), and pedestrians were not considered. A total of 126 accidents were considered; of which 95 with injuries and resulting in 7 fatalities, 53 serious injuries and 76 slight injuries. German official statistical definitions for injury severity of accident victims are the same as the ones used in the European CARE road accident database (ETSC, 2007; EC, 2011): fatalities are registered whenever a victim deceases within 30 days of the accident as a result of injuries produced in the accident; and serious injuries are hospitalized for at least 24 hours. 5 Comparison of results obtained with the Portuguese and German pilot road stretches 5.1 Pilot road stretch in Portugal Main results from the analysis of the IP4 stretch using the Portuguese method are summarized in Table 5 and Table 6. The results are presented per type of road element, with disaggregation between curve consistency classes. Accidents and injuries were registered in the period (see 4.1). Table 5 IP4. Consistency class distribution according to the Portuguese method Curve class Element Number Length (km) Traffic Volume (10 6 vehicle-km) Accidents (total) Fatalities Tangent O A B C D * - Number of killed and seriously injured victims KSI* Only a very small proportion of road elements belong to the lower consistency classes: they represent slightly above 5% of the traffic volume (as expressed by the travelled distance by traffic during the analysed period) but account to 12 % of the total number of accidents and 8% of the number of killed and serious injured victims. Overall, the number of curve elements in curve classes C and D is too low to allow conclusions regarding their safety as compared to other types of road elements. Accident rates (total and injury only) of tangents and O and A class elements are of the same order of magnitude (Table 6), the tangent rates being slightly inferior to the curve rates. Page 23 of 88