Application of a system-based and a scenario-based risk analysis to the Driskos tunnel Reflections about accuracy of collected data and uncertainties of risk analysis methods Raphaël Defert, BG Bonnard & Gardel Ingénieurs Conseils, France Ioannis Rentzeperis, Egnatia Odos, Greece Konstantinos Koutsoukos, Egnatia Odos, Greece Philippe Pons, BG Bonnard & Gardel Ingénieurs Conseils, France Didier Lacroix, CETU, France ABSTRACT Application of European Directive 2004/54/EC to road tunnels (over 500 m length on the trans- European road network) makes it necessary to perform risk analysis. Such risk analysis is also part of the safety documentation which is submitted to Administrative Authority for the operational approval of each tunnel. The Directive specifies a number of cases where risk analysis is required, and requires a Specific Hazard Investigation for all tunnels. Scenario-based and system-based risk analyses are the two main methodologies of risk analysis used for road tunnels. For the Driskos tunnel, situated near Ioannina in Greece in Egnatia Odos motorway, both methodologies have been applied: A scenario-based method has been used for the Specific Hazard Investigation (which aims at studying all kinds of risks); A system-based risk analysis has been performed to support the decision about restrictions or not of the transport of Dangerous Goods (DG) in the Driskos tunnel. The compared advantages and limitations of each type of methodology have been widely discussed. This paper focuses on, using the Driskos tunnel risk analyses as reference, relevance and limits of each type of methodology, taking into account the unavoidable uncertainties of the collected data and the risk analyses themselves. KEYWORDS: uncertainties, risk analysis, European Directive 2004/54/EC, quantitative risk assessment, Dangerous Goods INTRODUCTION As a result of the Article 13 of European Directive 2004/54/EC requirements, the French Tunnel Study Centre (CETU) and the consultant BG have been mandated by Egnatia Odos SA to jointly carry out risk analyses for the Driskos tunnel. The risk analyses have been carried out based on the French methodology (Guide to Road Tunnel Safety Documentation from CETU): 1. Risk analysis according to article 13 of the European Directive: a Specific Hazard Investigation has been carried out according to the French methodology described in the booklet 4 (Guide to Road Tunnel Safety Documentation - SHI). 2. Risk analysis due to the transport of DG, according to Annex I article 3.7 of the European Directive in 2 steps (booklet 3 (Guide to Road Tunnel Safety Documentation Risk analyses relating to dangerous goods transport): Calculation of Intrinsic Risk of the Driskos tunnel, as a first step to check if it's necessary or not to perform a complete comparative risk analysis in a second step. Comparative risk analysis between Driskos route (EO) and an alternative route through Ioannina. 113
From a methodological point of view, these risk studies are based on two different methodologies: Specific Hazard Investigation (scenario-based risk analysis): the method consists in applying a transversal approach (incorporating the tunnel s safety devices, operation and intervention resources, traffic volumes, etc.). The aim of this method is to analyse the development of some fire scenarios and their consequences (with the help of simulation software), in order to evaluate whether the tunnel measures ensure safe conditions or not. Comparative risk analysis (system-based risk analysis): it is supported by the use of a software (QRA model for this case) that fully quantifies the risks on the base of input data. Indeed, quantification of the frequency and severity components of the risks make it possible to determine the frequency/severity curves, which allows comparison of different possible choices. Egnatia Odos motorway is the Greek part of the European transportation route E90. It is a major motorway, that extends from the western port of Igoumenitsa to the eastern Greek-Turkish border at Kipoi. It is part of the trans-european network (TEN). The Driskos tunnel, operated by Egnatia Odos SA, is located near the city of Ioannina, in Epirus. It is a twin bores tunnel with a length of 4.560 m. Figure 1 Egnatia Odos motorway and Driskos tunnel This article covers the following topics successively: Presentation of the Specific Hazard Investigation of the Driskos tunnel; Presentation of the Comparative Risk Analysis of the Driskos tunnel; Discussion about uncertainties of these risk analysis methods; Conclusions. SPECIFIC HAZARD INVESTIGATION OF DRISKOS TUNNEL The Specific Hazard Investigation (SHI) is based on a reference condition. For Driskos tunnel, it corresponds to the current state of the tunnel, excluding additional ventilation measures (shaft) in the middle of the tunnel, which was not considered in the risk analysis. In a first step, the SHI describes the characteristics of the tunnel, its operation and environment, and takes into account the traffic forcasts for the 2020 target year. The safety provisions of the tunnel are compared with the minimum requirements of Directive 2004/54/EC. A functional description of the tunnel is made to analyse and check how the main structural facilities, equipment, operational and emergency measures contribute to the safety functions. Within the SHI, the objectives are to identify the hazards and to choose representative scenarios to be investigated in details. For that purpose, the following steps have been carried out: Taking into account, in an exhaustive way, the dangerous events that may occur; Prioritizing these dangerous events as a function of their frequency and potential severity; Selecting a few number of scenarios for an in-depth analysis. 114
The frequency/severity matrix below is a representation of the considered dangerous events with their 2 components (frequency and severity), as a support for the selection of scenarios: Frequency Breakdown of car Frequency-Severity Matrix of Driskos Tunnel A B Breakdown of HGV Breakdown of coach Material accident of HGV Material accident of car Material accident of coach Car accident with injury HGV accident with injury Fire of car under control or not Coach accident with injury C Fire of HGV under control Fire of HGV not controlled D Fire of coach under control Fire of coach not controlled E Fire of DG-HGV under control Fire of DG-HGV not controlled F Figure 2 I II III IV V Severity Frequency/Severity matrix of the Driskos tunnel The frequencies are classified according to different categories, from "Very frequent A" (return period < 1 year) to "Extremely rare F" (return period > 10 000 years). The severities are classified according to different categories, from "Minor I" (material damages) to "Major catastrophe - V" (> 50 fatalities). For the in-depth analysis, 3 fires of different heat release rates have been chosen: 30 MW (fire of an unladen HGV/HGV loaded with materials of low combustibility), 100 MW (full burning of a HGV loaded with combustible materials), 200 MW (fire of a tank of flammable liquids, classified as DG). Associated scenarios, with their contextual parameters, are presented below: Scenario Heat release (MW) Bore Position DeltaP (Pa) 1 30 Left bore PM 2283-50 2 100 Right bore PM 3370-50 3 200 Right bore PM 3370-50 Left bore Right bore Figure 3 List of fire scenarios Zone 4 These three fire scenarios have been analysed in the framework of very severe scenarios including a fully developed fire, adverse weather conditions and the worst possible location in the tunnel. Moreover, sensitivity studies have been performed to assess the influence of several parameters on the possible consequences of fires. 115
The theoretical frequency for an HGV fire in Driskos tunnel is about 1 fire every 75 years (all types of fires, fully developed or under control). Frequency of scenarios 1, 2 and 3 above (fig. 3) is much lower, because: Scenario are fully developed fires; - 50 Pa is an adverse difference of pressure, associated to a very limited frequency; Worst location (according to the ventilation design), in terms of possible effects, has been selected for scenario 2 & 3; Scenario 3 concerns a very limited number of HGV (HGV conveying flammable DG). For every scenario, space-time diagrams are computed. The following parameters are studied: visibility, temperature and CO concentration. These graphs are used to describe the evacuation conditions of the users. As an example, the following graph shows temperature conditions for scenario 2: Figure 4 Space-time diagram for temperature (scenario 2) The space-time diagram above shows the temperatures varying with space (abscissas) and time (ordinates). In this example, some transitory smokes upstream fire can be observed before jet fans are fully operated. However, this effect is only transitory in the first minutes, when fire is not fully developed. On the basis of all the results and analysis, it appears that in all cases, the longitudinal ventilation is able to push all the smoke towards the exit of the tunnel, so that the vehicles blocked behind the fire are finally kept in a safe atmosphere. However, in a few cases, when the traffic is low and the difference of pressure between portals is important and adverse, there is a risk of transitory smokes upstream during a few minutes. During this first phase, road users that are in the traffic jam could then be into smokes as long as jet fans are not fully operational. The consequences (especially temperature) would then depend on: Initial conditions (traffic and difference of pressure); Time to activate appropriated actions (especially jet fans activation); Heat release rate; Number and behaviour of the users; Location of fire. It must be highlighted that this kind of situation occurs in all tunnels with longitudinal ventilation and is considered acceptable in non-urban tunnels, given the very low probability of a fully-developed fire under such very unfavourable conditions. 116
Fourth International Symposium on Tunnel Safety and Security, Frankfurt am Main, Germany, March 17-19, 2010 COMPARATIVE RISK ANALYSIS OF DRISKOS TUNNEL Directive 2004/54/EC requires that a risk analysis is performed prior to the choice of a regulation regarding the transportation of DG through a tunnel. Such regulation must additionally comply with the provisions of the European Agreement concerning the International Carriage of DG by Road (ADR). According to ADR, any restriction to the transportation of DG through a tunnel must be made by assigning the tunnel to a category from A to E and by installing the corresponding signs. The principle of these categories is as follows: Figure 5 ADR categories for the transport of DG in tunnel In a first step, the Intrinsic Risk (IR) of the Driskos tunnel has been calculated with the help of the QRA model (PIARC). IR is an indicator of the risks due to the whole DG traffic in the tunnel. As the value of IR for the Driskos tunnel is higher than the threshold 10-3 fatalities/year, a complete Comparative Risk Analysis is compulsory. For the Driskos tunnel, the Comparative Risk Analysis (CRA) aimed at comparing the various possible regulations (A to E), considering that if all or part of the DG are banned from the tunnel, they will use one or several alternative routes which may be more or less dangerous than the tunnel route. Therefore, two routes have been considered: "Driskos route": it is the route (Egnatia Odos motorway) which pass through Driskos tunnel, from Arachthos-Zagori interchange (point A on the map below) to Ioannina interchange (point B). This route includes the Driskos tunnel as well as two short cut-and-cover tunnels. "Alternative route": the alternative route is the old national road that was used for the transport of Heavy Good Vehicles (HGV) before the construction of the Egnatia Odos motorway. This route is longer and partly urban, and includes a 430 m long cut-and-cover. The points A and B are the interchanges where DG-HGV would leave the motorway if Driskos tunnel was forbidden to DG. Cut and cover A Cut and cover Driskos tunnel Cut and cover Alternative route Driskos route Figure 6 B Routes considered in the Comparative Risk Analysis 117
This study has been carried out using a QRA Model developed under auspices of PIARC and OECD. It is dedicated to road transport of DG. For the two routes, the input data are: Characteristics of routes (number of lanes, speed limit, etc.); Characteristics of tunnels (safety devices, operation); Global traffic; Traffic of DG and repartition of different goods (flammable liquids, LPG, etc.); Accident rates; Population: on open air sections, a 2D model has been used for the description of the population; Meteorological data. A prospective investigation has been performed up to year 2020. In order to take into consideration more realistic input data, a DG traffic counting has been carried out for five days on the alternative route because DG were forbidden in Driskos tunnel at the time the study was carried out. The calculated risks are characterized by two components: The severity (in terms of number of victims); The frequency. The severity is assessed by means of probit equations establishing the correlations between the physical consequences of the scenarios (heating effects, toxic effects, shock wave, etc.) and the physiological consequences of these effects on the exposed people. The calculated severities are the consequences of the scenarios, during the few minutes which follow the accident. In addition, the results obtained should not be considered in an absolute way, but only by comparison (one route with the other), due to uncertainties of that sort of modelling. The risk indicators considered are: F/N curves, that plot the frequency F(N) of accidents with N or more fatalities; "Expected Value" (EV) which represents the average number of fatalities per year. For each regulatory hypothesis (A to E), EV are calculated to characterize the total risk on the network consisting of the two routes. For the Driskos tunnel, due to the nature of the DG found during the counting and the risk analysis methodology, the risk indicators are the same for categories B and C on the one hand, and for categories D and E on the other hand. As a consequence, the risk analysis can only differentiate three solutions for the regulations: A; B/C (i.e. B or C); D/E (i.e. D or E). Global Risk (tunnel route + alternative route) associated to the 5 categories in tunnel Cumulated Frequency (per year) 1.E+00 1.E-01 1.E-02 1.E-03 Itinéraire tunnel : Driskos_route_11 Itinéraire alternatif : alternatif Catégorie A - EV = 2.282E-2 Catégorie B/C - EV = 4.995E-3 Catégorie D/E - EV = 4.630E-3 1.E-04 1.E-05 1.E-06 1.E-07 1.E-08 1 10 100 1000 10000 Figure 7 Number of fatalities Comparison of the different regulations (F/N curves and EV) 118
The results show that the category A entails a slightly higher overall EV (23.10-3 ) than the other categories (5.10-3 ). However, this difference is not sufficient to base a decision on this only criterion, because of the uncertainties and the results of the sensitivity studies. Due to these results, other criteria have been considered. The other criteria proposed in Booklet 3 are: Decider's "risk aversion" (gives a higher weight to accidents which result in many fatalities, even if they are very rare); DG-HGV accidents not involving DG (considers the fatalities due to accidents involving DG- HGV in which the DG are not released and play no role); Route vulnerability to a DG-HGV accident from economic and environmental points of view; Economic considerations. The table below presents a synthesis of the 4 criteria by the 3 groups of ADR regulation: Table 1 Comparative analysis on the other criteria Decider's "risk aversion" DG-HGV accidents not involving DG Route vulnerability to a DG-HGV accident Economic considerations Category A Category B/C Category D/E From this table, it appears that categories A, B and C are the most appropriated restrictions regimes for the transport of DG. The overall comparison gives some advantage to categories A and especially B/C, so that the risk analysis suggests choosing category B (allowing only certain DG through the Driskos tunnel). However it is important to point out that this decision should be taken by considering not only the above four criteria and also the greek policy for the passage of DG through major highways. In addition, traffic data (whole traffic and DG traffic) for an extended period should be collected in order to fully analyse the issues of categorization. DISCUSSION ABOUT ACCURACY OF COLLECTED DATA AND UNCERTAINTIES OF RISK ANALYSES METHODS The two risk analyses that have been carried out for Driskos tunnel are based on two different methodologies that can be compared from the point of view of uncertainties: Specific Hazard Investigation (SHI) is a scenario-based risk analysis, which can be considered as a semi-quantitative method, because the work of quantification is not completely carried out. For example, the number of fatalities for scenarios studied is of few importance for analysis and not explicitly given, and the frequencies of each scenario (with contextual parameters) are not calculated (deterministic approach). Comparative Risk Analysis (CRA) is a system-based risk analysis, for which the components "Frequency" et "Severity" are completely quantified, but in a rougher way regarding the physical consequences of a scenario. This method allows calculating risk indicators (e.g. EV and F/N curves) given a tunnel, its environment, traffic, accident and other data. The calculation are carried out by means of a QRA methodology which can be considered as a "black box", in which data are entered, and from which risk indicators are produced. 119
The main difference between the two approaches is that for SHI, the aim is not to calculate precise numbers of fatalities and their frequencies of occurrence for some scenarios (which anyway cannot be precise due to the large uncertainties), as it is done in the CRA. The aim of the SHI is to analyse and understand, with the help of simulation models (software), the development of scenarios and their consequences on road users. This allows appreciating whether the tunnel measures ensure safe conditions or not, in the few but representative situations described in the scenarios. The figure below summarises these principles: Specific Hazard Investigation Input data + assumptions about circonstancies of specific scenarios Analyse of the consequences, with the help of a simulation model Semi-quantified risk results, that are interpreted by the risk manager Comparative Risk Analysis Input data + assumptions about generic circonstancies Risk assessment model Quantified risks results To + 2 minutes : fermeture du tunnel Cumulated frequency 1/(100m*year) 1.E-03 1.E-04 1.E-05 1.E-06 Catégorie A - EV = 1.324E-5 Catégorie B - EV = 1.188E-5 Catégorie C - EV = 8.349E-6 Catégorie D/E - EV = 0 1.E-07 1.E-08 1.E-09 1.E-10 Figure 8 Comparison of CRA and SHI principles 1.E-11 1 10 100 1000 10000 EV = Expected Value = Fatalities / year Number of fatalities The SHI process involves important analysis work by the risk manager, such as: Decision choices at every step of the study (what scenarios, what contextual parameters, etc.); Step by step calculations and evaluation of the results; Conclusions, interpretations of results and finally recommendations. On the other hand, the CRA process generates automatically the above two tasks by a risk assessment model. The above methods present 3 different sources of uncertainties: Using a model, which necessitates a more or less major simplification of the real conditions, induces some uncertainties. Simplifications are a necessity regarding the complexity of a real case. Those methodological simplifications induce intrinsic approximations that contribute to uncertainties at a basic level; Data used for assessment of the risks are more or less uncertain (time to launch ventilation, traffic data, etc.); Tools/software that are used to calculate the severity of dangerous events, e.g. fire simulation models (used to calculate ambient conditions in case of fire in a tunnel), are based on intrinsic simplifications. It is very difficult to know how large the uncertainties are, because they are generally not quantified themselves. As a matter of fact, such an evaluation would even not be possible in most cases, due to the number of data and models used in a risk analysis, especially a system-based risk analysis. For further examination of the uncertainties in general and beyond any consideration of the Driskos case study, a comparison of the two processes SHI and CRA was performed. The common situation considered was a scenario of a fuel tanker fire: 120
In SHI, scenario 3 is a 200 MW fire (standardised Heat Release Rate), which is considered to be representative of a fuel tanker fire; In CRA, a scenario of pool fire, associated to flammable liquid (fuel tanker), is considered. The two different ways to assess the risks associated to this scenario, and the related uncertainties, will then be compared in the following. Uncertainties in the simplification of the reality The table below compares the simplifications of the reality for each study: Table 2 Simplification of the reality for each study Choice of contextual factors Source term Behaviour of road users Assessment of the consequences on road users Assessment of the frequencies SHI process CRA process For the 200 MW fire, several contextual For fuel tanker, a scenario of pool fire is factors have been selected: included in the QRA model. Calculations are carried out for: 3 traffic conditions (hourly traffic) 3 different configurations of traffic 3 positions of the fire in the tunnel 5 positions of fire in each bore Several ΔP between portals 1 ΔP between portals (sensitivity study) Several kinetics of operating Possibly 3 kinetics of operating conditions (closure of the tunnel, conditions (according to periods of time ventilation, etc.) in the day) Standardized source term (Heat Release Rate, CO production, etc.) for a 200 MW fire, which correspond to a simplification of the reality taking into account all the possible pool fire surface according to the geometric characteristics of the tunnel Some parameters are standardized (velocity, time to react, etc.) / Several behaviours have been considered in the SHI: Users stay or not in their vehicles Users take or not cross-passages Users walk or not in the right direction Hypotheses are taken into account concerning the lethality thresholds for people. These thresholds are compared with the physical effects calculated by the simulation model, but no number of fatalities is calculated. Frequencies are assessed for families of events (fires, accidents, etc.), and are only used to select scenarios. Heat Release Rate and other parameters roughly calculated according to the geometry of the pavement (slope, cant) and the capacity of the drainage system (surface of the pool fire). Standardized hypotheses are made in the QRA model concerning the behaviour of road users (velocity, time to react, etc.), according to the different alert safety devices in the tunnel (radio, signalling, etc.). Hypotheses are taken into account concerning the lethality thresholds for 3 percentages of lethality. These thresholds are compared with the physical effects calculated by the QRA model, and the number of fatalities for each configuration is deduced. This is a simplification in that the percentage of lethality is a continuous value. Moreover, it is supposed a repartition of the categories of road users inside the tunnel: Pedestrians Road users in their vehicles with closed windows, etc. Probability distributions are assigned in QRA model for each parameter (e.g. probability for each position of fire in the tunnel). These distributions (range of values that the parameters can assume with associated probability) induce uncertainties 121
From this table, it appears a priori that uncertainties linked to the simplified representation of the reality are not fundamentally different for consequences (except for the use of different softwares, if they have not the same level of accuracy, see thereafter). Actually, they are probably higher for CRA because the calculation of number of fatalities requires a more detailed discretization of the reality. In CRA, uncertainties related to the assessment of frequencies have to be considered in addition. Indeed, it must be considered that simplifications induce 2 sources of uncertainties for CRA: Uncertainties on severities (discrete values of parameters having an influence on severity); Uncertainties on frequencies (probability distribution of these discrete values). On the contrary, only 1 source of uncertainties exist in a scenario-based approach. Indeed, SHI is more a deterministic approach and the uncertainties concern mainly severity component. In addition, it should be noticed that for SHI, the risk manager has the possibility to choose all the contextual elements that seem necessary, until he estimates that the set of scenarios is sufficiently representative of all cases that could happen in the tunnel. In any case, these simplifications introduce an important level of uncertainties that remain very difficult to quantify. Uncertainties in the input data The table below compares the level of detail of input data for each method: Table 3 Level of details of input data for each study SHI Geometric Length characteristics Cross-section area of the tunnel Slope Cant Number of lanes Safety devices Closure devices Detection/operation devices (CCTV, radio, signalling, etc.) Ventilation system (number, localisation, power of jet fans, etc.) and ventilation scenarios Water mist system and its technical characteristics (if any) Delays Delay to close the tunnel Delay to launch ventilation and/or water mist system Delay to inform users Delay to inform emergency services Traffic data Hourly traffic (+ other traffic parameters of lower influence) CRA Length Cross-section area Slope Cant Number of lanes Closure devices Detection/operation devices (CCTV, radio, signalling, etc.) Ventilation system (number, localisation, power of jet fans, etc.) and ventilation scenarios Specifications of drainage system Delay to close the tunnel Delay to launch ventilation Delay to inform users Hourly traffics for 3 periods of time AADT of fuel tanker traffic Average length of vehicles Average number of people in vehicles (cars, coaches, etc.) Speed of vehicles Not taken into account Emergency services Possibility to take account, in a qualitative way, of intervention of emergency services Fire rates Fire rates are considered to have an idea of Fire rates are considered to calculate all the the frequency of occurrence frequency of occurrence Environment Difference of pressure between portals Difference of pressure between portals 122
From this table, it appears that the number of input data is of the same order of magnitude for both methodologies. The main difference concerns traffic data necessary to calculate frequencies of occurrence (proportional to the fuel tanker traffic) and number of fatalities (length of the traffic jam, number of road users, etc.): they are more detailed in the CRA. For both methods, it should be noticed that the risk manager has the possibility to carry out sensitivity studies for all these parameters. For example, the variations of the results according to different values of a parameter (detection times, traffic levels, etc.) allow to determine the impact of each of these parameters and to define a severity range associated to each scenario. The uncertainties of these families of data are very various between them. The table below gives a qualitative appreciation of their level of accuracy, and of their level on influence on the results: Input data Order of magnitude of uncertainties Influence on results (from *: low influence to ***: high influence) Geometric characteristics of the tunnel * Safety devices ** Delays *** Traffic data *** Fire rates (1) ** Environment ** Figure 9 Order of magnitude of uncertainties for each input data (1) In contrast to traffic accidents on open roads, there is relatively little data on incidents in road tunnels and published statistics for few countries only (information on tunnel fires is still quite sparse). Uncertainties in the modelling tools The table below compares the uncertainties of each method for modelling tools: Table 4 Modelling tools for each study Simulation model SHI For SHI, 1D or 3D simulation models can be used according to the ventilation system. CRA In CRA, some rough models are used to assess distances of effect inside the tunnel. For fires, a 1D model is used, which cannot take account of stratification phenomena. Then uncertainties are much higher when a 3D model should be used. From this point of view, it clearly appears that SHI makes it possible to choose adapted simulation models, even if they include some unavoidable uncertainties. On the contrary, in system-based methods, modelling tools that are used have to be adapted to all the different configuration of tunnels (geometry, ventilation system, etc.). For this reason, these models are generally more simple and rougher than specific software that can be used by a risk manager, which has the possibility to adapt the sophistication of the model used to the complexity of the tunnel studied. In scenarios-based methods, it is therefore possible to reduce uncertainties linked with modelling tools. 123
CONCLUSION From the analysis above, the following conclusions can be drawn: As system-based methods aims at quantifying every component of the risk (frequency and severity components), related uncertainties are higher than for scenario-based methods. For scenario-based methods, uncertainties only affect the consequences of the examined scenarios. For system-based risk analysis, uncertainties affect the whole F/N curve. Moreover, system-based methods are generally based on dedicated software for the calculation of consequences. This model has to be adapted to all the different systems, and is most of the time simpler and rougher than specific software that can be used by a risk manager. Indeed, a risk manager has the possibility to adapt the sophistication of the model to the complexity of the tunnel. In scenario-based methods, it is therefore possible to reduce uncertainties linked with modelling tools. To reduce uncertainties, whenever possible, it is more appropriated to use specific data than generic data or default values. If the use of specific data is not possible, it is at least necessary to check the origin of the data used (are the conditions relating to infrastructure, traffic, etc. similar to the case studied?). In any case, there are many sources of uncertainties. Consequently, the results (space-time diagrams, risk indicators, etc.) may be more pertinent in a comparative approach (e.g. of various routes, or of an existing state to a reference state of a tunnel). Indeed, the best way to reduce the uncertainty is to compare situations in which you only modify one parameter at a time. In a comparative approach, the uncertainty is certainly lower than in an absolute approach, because there are fewer differing data and models. It could appear that some uncertainties are larger than the differences between the values you wish to compare: in these cases, it is therefore difficult to conclude whether the alternative measures ensure an equivalent protection or not. Results coming from risk studies must be interpreted as an order of magnitude and not as absolute values. For these reasons, risk analysis should only be performed by specialists with sufficient experience and understanding of the methods they use and their intrinsic limitations. ACKNOWLEDGMENT: The author would like to thank Mr. Athanasios Tsantsanoglou, supervisor of EOAE for the Risk Analysis study of Driskos tunnel, for his valuable support and assistance throughout the duration of the study. 124