Observational method as risk management tool: the Hvalfjörður tunnel project, Iceland

ABSTRACT Rock tunnel construction is associated with considerable geotechnical uncertainty, often due to limited knowledge about the ground conditions. This warrants the use of stringent risk management procedures to reduce the likelihood of cost increases, delays, and structural failure events. The observational method is often promoted as a tool to achieve cost-effective designs in cases of large geotechnical uncertainty, but its practical use is still limited. One reason may be the lack of guidelines and experiences from previous projects where the observational method has been used. In this paper, we therefore present a case study of the design and construction of the tunnel under the Hvalfjörður fjord in Iceland, where the observational method played a key role in the risk management that was performed to deal with the challenging geological conditions at the site. The project was a success and completed four months earlier than originally planned. In light of the case study, we discuss the definition of the observational method in Eurocode 7 and the related contractual aspects to consider in such projects.


Introduction
Design and execution of underground structures involve risks related to the geotechnical conditions. The geotechnical risks may lead to cost overruns, time delays as well as safety issues if they are not managed properly. To avoid such consequences, many authors have recommended employing a risk-based project management approach in geotechnical engineering projects, such as Clayton (2001), van Staveren (2006van Staveren ( , 2013, Stille (2017), and Spross, Olsson, and Stille (2018). General principles of risk-based rock engineering design have been discussed by Spross et al. (2020).
In ISO 31000 (CEN 2018), risk is defined as the "effect of uncertainty on objectives". Today this is probably the most widely used definition of risk in technical contexts. Risk is often characterised by reference to potential events and expressed in terms of the combination of the likelihood of occurrence of an event and the associated consequences. For geotechnical engineering projects, the uncertainty about the geotechnical conditions at the site normally contributes the most to the total project risk. The geotechnical uncertainty arises from epistemic or aleatory uncertainty regarding the geotechnical conditions (Ang and Tang 2007). The epistemic uncertainty is caused by incomplete knowledge about the geotechnical conditions, often owing to very limited investigations of the geotechnical conditions before construction. The aleatory uncertainty is caused by the spatial variability or randomness in the parameters that govern the geotechnical behaviour.
Since the early days of civil engineering, observations have been used by engineers to deal with geotechnical uncertainties and to reduce risks by observing the performance of structures. Historically, modifications of the design based on observations were often made using a trial-and-error process or ad-hoc process. With the development of modern geotechnical engineering, an integrated process of predicting, monitoring, reviewing and modifying the design gradually evolved. This process was eventually named the "observational method" by Peck (1969). Application examples have recently been provided by, for example, Nicholson, Tse, and Penny (1999), Prästings, Müller, and Larsson (2014), Bjureland et al. (2017), Fuentes, Pillai, and Ferreira (2018), Lacasse and DiBiagio (2019), Duncan and Brandon (2019), Powderham and O'Brien (2020), Spross, Bergman, and Larsson (2021), , and Zheng et al. (2021). 2. Purpose and structure of the paper Geotechnical risk management with the observational method has the potential to provide a safe and cost-effective design in projects including geotechnical uncertainties. However, there is a lack of recommendations regarding the implementation of the observational method, in particular regarding its connection to risk management and contractual aspects. This may have restricted the use of the method. In addition, cases of successful implementation of the observational method are rarely published; the aforementioned examples are among the rare exceptions. Therefore, this paper presents a case study of the design and construction of the tunnel under the Hvalfjörður fjord in Iceland. The study presents the experiences from the risk management process during the design and execution phase of the project and highlights key aspects thereof in relation to the implemented observational method and the contractual framework used. A comparison is made with the definition of the observational method in Eurocode 7 (CEN 2004) since this is the current geotechnical design code in many European countries, including Iceland.

Methodology
The management of risks in a geotechnical engineering project is influenced by many factors, such as the perception, communication and allocation of risk, as well as the cooperation between the parties involved. It is therefore normally difficult to isolate and study the effect of each factor separately. For complex situations, or in contexts where it is difficult to study a specific phenomenon, Yin (2018) suggests that a case study is generally an appropriate research methodology. Therefore, due to the complexity of the risk management process in geotechnical engineering projects, a qualitative case study was chosen as the research method.
This case study was conducted in two steps. The first step consisted of an analysis of the tender documents, the contractual documents, the geotechnical site investigation reports, and the geotechnical risk management process in the design and execution phase. The second step consisted of semi-structured interviews performed in 2004 with key personnel involved in the geotechnical risk management process, i.e. the Swedish contractor's project manager, the principal geotechnical engineer, and the engineers responsible for the geotechnical risk management in the design and execution phase. Appendix A provides the interview instrument.
As in all qualitative research, it is important to ensure trustworthiness in case study research. Examples of difficulties to ensure trustworthiness are those related to credibility and generalisability (Guba 1981). The difficulty with credibility was addressed by using different information sources, i.e. written material from the project and interviews with several individuals involved in the risk management process and the design and execution of the project. Concerning the generalisability, it may be questioned whether general conclusions may be drawn from one case study. Stake (2005) argues however that a single case, even if it is unique, can be used to generalise since it could be an example of a broader group of cases. We believe that the conclusions in this paper are not specific for the case study and that the conclusions may be used in a broader context.

The observational method in geotechnical engineering
The conceptual idea behind the observational method is to actively use information obtained from observations during construction to reduce the epistemic uncertainty and, based on this information, modify a preliminary design to the actual geotechnical conditions on site (Peck 1969). The observational method consists of the following steps: establishment of a preliminary design, preparation of contingency measures to put into operation in case of deviations from the preliminary design conditions, monitoring or other observations during the execution of the project, and finally, modification of the preliminary design to the actual conditions based on the observations into a final design. Thus, the final design is not known before project completion. Notably, implementation of contingency measures (i.e. modification of the design) has contractual implications, as this affects the cost and time schedule of the project.
In principle, the most appropriate design method is the one that results in the lowest cost while fulfilling the formal requirements, e.g. structural safety, serviceability, durability, and allowable environmental impact. On the one hand, the observational method is associated with extra costs compared to traditional design methods because of extra design work and extended monitoring while executing the work. Although the final design using the observational method is likely to be better suited to the actual ground conditions, these extra costs can, however, be outweighed by savings from avoiding overdesigning; this can be analysed with statistical decision theory .
According to European design code EN 1997, Eurocode 7 (CEN 2004), limit states can be verified by one or a combination of the following methods: calculations, adoption of prescriptive measures, experimental models and load tests, or an observational method. Eurocode 7 specifies the principles that must be fulfilled when applying the observational method. These principles are further discussed in light of the case study at the end of the paper. The recognition of the observational method as an allowed design method in Eurocode 7 formally allows the designer to exploit observations during the execution of work to decrease the geotechnical uncertainty.

Project description
The road tunnel under the Icelandic Hvalfjörður fjord is located approximately 20 km north of Reykjavík. At the location of the tunnel, the fjord is approximately 3.5 km wide. The tunnel under the fjord connects the northern and southern parts of Iceland ( Figure 1). The tunnel's length is around 5.8 km and it is located approximately 165 m beneath the water's surface at its deepest point. The southern and northern entrances have inclinations of 7.1% and 8.0%, respectively.
The planning of the project started in 1987 when the Icelandic Public Road Administration published a study that presented the benefits of building a tunnel under the Hvalfjörður fjord. After several years of feasibility studies regarding the location of the tunnel, site investigations, and design, the excavation of the tunnel started in June 1996. The tunnel was entirely excavated in October 1997 and opened for traffic in July 1998, which was approximately four months before the originally estimated completion date.
The client was the Icelandic Public Road Administration, and an Icelandic private enterprise obtained the concession to design, build, own, and operate the tunnel for twenty years. After the operating period, the tunnel was transferred to the client. A joint-venture consisting of contractors from Iceland, Denmark and Sweden was contracted with a turn-key contract through a public procurement process, where the joint-venture guaranteed the financing of the project during the construction period. The tunnel works were paid with a lump sum payment, with a minor remeasurement payment made when the tunnel was completed, tested and had been operating for two months. The tunnel project was unique for Iceland for two reasons: it was the first subsea road tunnel built in young geothermally active basaltic lavas, and it was the first design-build-own-transfer project.

Performed site investigations
The geotechnical survey presented in the tender documents was quite limited considering the type of project and the uncertainties involved. It consisted of geological surface mapping along the shores of the fjord, 650 m core drilling of the rock mass, and a geophysical investigation. The core drilling at the southern shore consisted of two vertical holes and one inclined hole about 200 m along the planned tunnel route. The geophysical investigation consisted of 500 km reflection and 40 km refraction seismic surveys along the tunnel route to estimate the level of the rock surface and the potential presence of weak zones. The results from the site investigations were used to identify geological hazards and estimate Q-values of the rock mass, which served as a basis for the design of the support.

Geological conditions
The geology in the area is characterised by the Mid Atlantic Ridge with previous volcanic activities. The bedrock consists of layers of solidified magma of Piacenzian age (approx. 2.6-3.6 million years old) with thin interbeds of sediments of ash and sand, as detailed in Figures 2 and 3. The solidified magma layers often had an impermeable central layer of good-quality basalt (high Q-value), but with vertical fractures due to the solidification process. The outer layers of the solidified magma, called "scoria", usually had less strength and higher permeability than the central layers. The rock was faulted along the fjord with a general spacing of 50-200 m and a filling of scaly or silty material, with thicknesses ranging from a few millimetres to 300 mm. The larger faults were brecciated with a thickness of up to 2 m. Due to earlier magma flows, there were basalt dikes cutting through the sequence of horizontal layers. The contacts were often slightly brecciated or filled with clay or silty material. In some places, there were layers of conglomerate of silt, sand, gravel, stones and boulders between the scoria and basalt. Clays found in dike contacts and faults were tested for swelling properties, the existence of which were detected at four places along the tunnel.
The rock cover was approximately 40 m at the deepest point of the tunnel. The thickness of the postglacial sediment layers on the bottom of the fjord varied between 10 to almost 80 m ( Figure 3). To ensure that there would be appropriate support measures for all possible rock conditions, an engineering geological forecast was prepared and presented in the tender documents.

Design approach for the rock support
The tunnel was excavated by drilling and blasting, and supported by rock bolts and shotcrete. Continuous probe drilling and pre-grouting ahead of the tunnel face were used together with systematic geological mapping of the rock mass at the tunnel face. The detailed design of the rock support was based on five contractual support classes, specified in the engineering geological forecast in the tender documents (Table 1). As the applicability of each support class was defined by a span of Q-values (Barton, Lien, and Lunde, 1974), the installed support could be adjusted to the actual rock mass conditions at the tunnel front, after the Q-value had been determined.  This design and construction method was chosen by the joint-venture to ensure a flexibility that could handle the potentially rapid changes in geotechnical and hydrogeological conditions. The approach was known in Sweden at the time as "active design" (Stille 1986), which is conceptually equivalent to Peck's (1969) observational method. It was believed that a traditional, more conservative, design and construction method, would probably lead to a less flexible and a more costly design, as it would need to apply very cautious design assumptions, but still yield an unpredictable safety margin due to the uncertainties involved.

Hydrogeological conditions
The water depth in the fjord is generally around 10-30 m at the location of the tunnel. The pre-investigations indicated that high water inflow of fresh or salt water could be expected from faults, vertical pipes, and in areas of contact with existing dikes. The contact zone between the dikes and the original basalt layer was found to have high hydraulic conductivity in some places. The water temperature in the tunnel was estimated to rise from 5°C at the tunnel ends to 25°C at the deepest part ( Figure 4).
6. Geotechnical risk management using the observational method

General concepts
Based on the definition of risk management in ISO 31000 (CEN 2018), a geotechnical risk management process can be described as a "systematic application of management policies, procedures and practices to the activities of communication, consulting, establishing the context, and identifying, analysing, evaluating, treating, monitoring, and reviewing geotechnical risks". The main activities of the risk management process are shown in Figure 5. A key activity is the first step, which is to create an understanding of, and to interpret, the geotechnical context in which the project is to be carried out . Using the observational method in the Hvalfjörður project can be viewed as a risk treatment to deal with the identified risk of having considerable geotechnical uncertainties at the site; thus, the observational method becomes an integrated tool in the risk management process, and not a substitute for it. In the following subsections, the risk management process used in the Hvalfjörður project is discussed in relation to the activities in Figure 5.
6.2 Risk management in the design phase

Organisation
In previous tunnel projects in Iceland, there had been problems related to the geological formations significant for the area. Therefore, the client was aware of the technical challenges of the project. In the design phase, risk management was conducted by a system analysis group consisting of engineers that were independent from the client, contractor, and concession owner organisations. The group's main task was to identify and describe the potential hazards and their associated initiating events and warning bells and, based on these, propose additional site investigations, an appropriate design, and suitable construction methods. The risk management process addressed the evolution from hazard to damage ( Figure 6).

Establishing the geotechnical context
The first step of the risk management process consisted of gathering information regarding the geological conditions at the site in order to create an understanding of the geotechnical context. As no one in the system analysis group had any experience in tunnelling in Iceland, an extensive literature review was performed, and several Icelandic and Norwegian experts in this type of project were contacted. Experiences from tunnels in similar geological conditions in Iceland and subsea tunnels in other countries were also studied. This first phase resulted in a preliminary model of the geology, hydrology, and the geothermal conditions at the site.

Risk and hazard identification
This second step included a qualitative fault tree analysis to identify the hazards and chains of events that could lead to technical failure of the project, such as a collapse or inundation of the tunnel. The aim of the initial fault tree analysis was not to quantify the risk, but to identify    as many potential hazards as possible. Two groups of hazards were identified: geotechnical hazards and organisational hazards. This fault tree analysis resulted in a register of the identified geotechnical hazards. Only those damage events that were obviously not present were disregarded in the analysis. Six damage events that were crucial to the technical success of the project were identified: . Water inflow that cannot be controlled. Inflow of salt water or fresh water could stop the tunnelling and make the project too expensive to continue. . Stability problems. Large deformations and/or rockfall in the tunnel, primarily due to weak rock formations. . Heat problems. Inflow of hot water or high temperatures in rock formations that stops the tunnelling work or makes the tunnelling work more complicated and time-consuming. . Harmful gases in the tunnel. Suffocating or poisonous gases that stop the tunnelling work or make the tunnelling work more complicated and time-consuming. . Damage due to seismic activity. Damage to the tunnel from seismic activity (e.g. an earthquake) that stops the tunnelling. . Insufficient tunnel durability. The tunnel starts to deteriorate because of environmental impact, resulting in high maintenance costs.
The second step also included a reassessment of the preliminary geological model and the identified hazards and damage events with assistance from the involved experts. The contractor had considerable influence over the choice of excavation techniques. The experts served as a review group during the risk management process. The second step resulted in a list of the hazards that could endanger the execution of the project.
Organisational hazards were also identified. The process of gathering, documenting, interpreting, and communicating the information from the observations was identified as important since this should be the basis for the decisions made regarding the execution of mitigation actions. Figure 7 shows the chain of events from observation of the indicator "existence of poor rock in front of the tunnel" to the execution of the mitigation action, i.e. increased grouting. Each of the four eventsdesignated A, B, C, and Dwere further analysed with separate fault trees in order to find the subevents that could result in these undesired events. Examples of sub-events were human errors and lack of personal resources, lack of knowledge, and poor cooperation.

Risk analysis
Due to the limited rock cover under the fjord, the unlimited access of water, and the uncertainties regarding the geological formation, the damage event dealing with "Water inflow that cannot be controlled" was considered to be the most critical damage event and a major threat to the completion of the project. This damage event was further analysed in detail while the other damage events were analysed more schematically. The fault tree for the top event "Water inflow that cannot be controlled" is presented in Figure 8. Each of the four events at the bottom of the tree was further analysed in separate fault trees; Figure 9 shows the tree for the subevent "Water from basalt layer" as an example.
The risk analysis step also included the identification of observable damage indicators ("warning bells") as well as methods for observing and measuring these during the execution of work. For example, the identified warning bells for the damage event "water inflow that cannot be controlled" were: . The occurrence of vertical geological formations: may indicate contact with the fjord. . Water temperature: deviations from expected values based on the thermal properties of the rock mass and temperature measurements in the boreholes during the site investigations may indicate connections with deeper layers (high temperature) or shallower layers (low temperature) of the rock mass. . Water salinity: high values may indicate connection with salt water from the fjord. The chloride content in the groundwater was expected to be quite low due to the flow of fresh groundwater from high terrain on both sides of the tunnel. . Water pH value: deviations from expected values depending on the type of rock indicate connections with deeper (high pH) or shallower layers (low pH) of the rock mass. . Coloured drill water/drill cuttings: may indicate the presence of sediments or weak rock mass which can affect the stability of the tunnel. Five possible colours were identified together with their geological interpretation. . Drill penetration rate: indicates changes in the properties of the rock mass, e.g. the type and quality of the rock.

Complementary site investigations as risk treatment
After the risk analysis, some complementary site investigations were planned based on the identified hazards. The aim of the complementary site investigations was  Water inflow that cannot be controlled". Each of the four subevents at the bottom was further developed in its own tree; that of "Water from basalt layer" is presented in Figure 9 as an example.
to increase the amount of data available in the decision-making process, i.e. reducing the uncertainty, where this was possible and economically feasible. The investigation methods were chosen by the contractor after considering recommendations from the expert group. The methods were chosen to give obvious and exclusive indications of the targeted hazards. Thus, the contractor made additional site investigations before the excavation started at the north shore, aiming at understanding better the potential stability problems at the northern tunnel entrance. This was achieved by drilling a long vertical core hole at the north shore to investigate the quality of the rock mass, and six shorter vertical holes to investigate the rock cover.

Organisation
By analysing the contractual framework, the contractors understood that a considerable part of the risks was owned by the joint-venture, meaning that this entity was responsible for these risks and any decision-making regarding their treatment (Spross, Olsson, and Stille 2018). Therefore, risk management had a prominent role in the execution of the tunnel. The risk management process in the construction phase was conducted by the contractor, and the independent expert group was removed from the project since their support was no longer needed.

Planning of risk treatment and monitoring
Risk management served as a basis for the organisation and quality assurance of the project, and improvement of the working procedures. Several risk treatment actions were performed before the start of the execution of the tunnel, such as training the site personnel to perceive hazards, initiating events and warning bells, and gathering materials and equipment for stand-by at the site in the event of, for example, weak rock conditions or the inflow of water. The predetermined rock support classes played an important role in allowing the adjustment of tunnel support in accordance with the observational method.
To determine the support class to be used, continuous probe drilling ahead of the tunnel face was executed, and the Q-value of the rock mass was mapped after every blasted round. The information was used to determine the rock support and to monitor the other hazard indicators, as shown in Table 2 along with the corresponding expected behaviour, threshold values and contingency actions.
The deformation of the tunnel was also observed by convergence measurements, which were conducted in four sections in the tunnel by using angular measurements and extensometer measurements. The monitored locations were a portal with potential stability problems, two sections with wider tunnel spans, and one instance of detected high pore pressure (not preplanned measurement). However, no strict threshold values for allowable deformation were established in advance.

Examples of monitoring and risk treatment during construction
The identified damage indicators in Table 2 were monitored at the tunnel face during the excavation of the tunnel. The amount of support was adjusted along the whole tunnel using the observed Q-value reported by the engineering geologist, in accordance with Table 1.
The measurements of the water temperature and the salinity in the tunnel are presented in Figure 10. The highest water temperature at the face of the tunnel was approximately 58°C, which was more than 30°C higher than the temperature forecast for this location (around station 4000 m in Figure 10). This indicated connections with deeper layers of magma. Therefore, the application of shotcrete was changed, and each newly blasted round was left without shotcrete and the rock surface was watered as a contingency measure, as planned (Table 2). To decrease the air temperature, the ventilation of the tunnel was increased. These measures gave a rather rapid cooling effect, so the plan was successful. The high salinity in some places ( Figure 10) indicated direct contact with seawater  through dikes, fault zones, or vertical pipes. In these areas, the grouting and thickness of the shotcrete were increased as a contingency measure ( Table 2). The excavation also passed through a number of smaller fault zones and dikes. They were treated as planned with the appropriate support classes (Class 3 or 4a; see Table 1) and grouted if the water leakage warranted this action. Close to the fault at station 4087 m, measured pore pressure in two 3-4 m deep holes showed a rapid build-up to around 0.9 MPa (Figure 11), indicating that the area had similar transmissivity to that of the fault (around 10 −3 m 2 /s). Although the rock mass was mapped as Class 2, it was decided to use Class 4a and, in addition, to monitor the deformation (which however turned out to be negligible).
At an area with sandstone cut by several small faults and dikes, the water flow in the probe drill holes measured 28-72 l/m, so the area was pre-grouted. During excavation, stability problems occurred at this location, causing rockfalls, so Class 3 and 4a support was used. Extra bolts, strapping, and thicker shotcrete were used in one part of this area where rock was splitting up in slabs. Afterwards, there was no indication of further displacement.
The deformation measurements at the four monitored sections all showed small deformations (less than 1 mm). The stability of the rock mass was generally better than expected in terms of the Q-value, so the final amount of rock support was around 20% less than the predicted amount for both rock bolts and shotcrete. This saving was enabled by the strict use of the support classes. As it turned out, Class 4b was not used at all, and Class 4a only in 48 m of the tunnel. The porphyritic basalt had generally better quality than the tholeiitic basalt (see Figure 3 for locations). The porphyritic basalt mainly gave block stability problems, which were solved using spot bolting, while the tholeiitic basalt tended to split into thin slabs, which were supported using systematic bolting and shotcrete.

Introduction
The design and execution of the tunnel under the Hvalfjörður fjord as described above has many similarities with the principles of the observational method as defined in Eurocode 7, with five paragraphs (CEN 2004). Because of this, and also considering the lack of well-documented case studies on the practical use of the observational method, we find it highly relevant to discuss the tunnel project in light of the Eurocode definition of the observational method. The first paragraph refers to the suitability of the method: Prediction of geotechnical behaviour is often challenging in cases of substantial epistemic uncertainty (lack of knowledge) regarding the geotechnical conditions. This was also the case with the Hvalfjörður project, as is clear from the project description. In addition, the project team had limited knowledge and experience with subsea tunnelling in Iceland, and there had been failures in other similar tunnel projects. Under these circumstances, the observational method can be expected to be the most cost-effective solution.

Preparations during the design phase
The second paragraph in Eurocode 7 includes five requirements to be fulfilled in the design phase (P stands for principle and implies a mandatory clause). Each of these requirements is considered below: (2)P The following requirements shall be met before construction is started: acceptable limits of behaviour shall be established; This requirement on acceptable limits of behaviour corresponds to the Table 2 threshold values for when the design needs to be modified and/or contingency measures implemented. Additional threshold values were established regarding the deformation of the rock surface in the tunnel. Some of these thresholds were expressed in qualitative terms, e.g. "high and low rate of drill penetration", which made the decision-making process regarding the implementation of contingency measures more difficult, as the interpretation of these limits were rather ambiguous.
the range of possible behaviour shall be assessed and it shall be shown that there is an acceptable probability Figure 11. Pore pressure measured in two drill holes, close to a fault at station 4087 m. that the actual behaviour will be within the acceptable limits; The range of possible behaviour was not established for the behaviour of the rock mass in terms of all observed indicators. Thus, it was not demonstrated that the actual behaviour likely would fall within the range of acceptable behaviour (i.e. not violate the threshold values). An exception was the rock support classes. They were connected to the Q-value mapped at the tunnel face, so the possible Q-values thereby corresponded to a range of possible behaviour. However, the probability that the rock support and grouting would be within the normal behaviour (Class 1-3, Table 2) was not assessed. Thus, it was not known in the design stage how difficult and costly it would be to handle large areas of very poor rock and/or reduce the high inflow of water to acceptable levels, had they occurred. Notably, this is a matter of cost rather than structural safety as long as it is possible to undertake the contingency actions more extensively than expected. We note here, however, that not analysing the probability that the actual behaviour will be within the acceptable limits does imply a considerable economic risk for the risk owner, which for the Hvalfjörður project was the contractor due to the turn-key contract.
a plan of monitoring shall be devised, which will reveal whether the actual behaviour lies within the acceptable limits. The monitoring shall make this clear at a sufficiently early stage, and with sufficiently short intervals to allow contingency actions to be undertaken successfully; the response time of the instruments and the procedures for analysing the results shall be sufficiently rapid in relation to the possible evolution of the system; A monitoring plan was established by the joint-venture before the start of the excavation of the tunnel. The monitoring included deformation measurements and the indicators in Table 2. The monitoring had sufficient response time.
a plan of contingency actions shall be devised, which may be adopted if the monitoring reveals behaviour outside acceptable limits.
A plan of contingency actions was devised before the start of the excavation of the tunnel. In case a threshold value was exceeded (e.g. too much inflow of water into the tunnel), the senior geologist would contact the senior hydrogeologist, the senior rock engineer and the project manager, who would then decide where, how, and to what extent the contingency actions would be undertaken, as detailed in Table 2. The fact that the extent of the action was not predetermined deviates from Eurocode 7, which requires detailed preplanning.

The construction phase
For the construction phase, Eurocode 7 states: (3)P During construction, the monitoring shall be carried out as planned.
(4)P The results of the monitoring shall be assessed at appropriate stages and the planned contingency actions shall be put into operation if the limits of behaviour are exceeded.
(5)P Monitoring equipment shall either be replaced or extended if it fails to supply reliable data of appropriate type or in sufficient quantity.
The monitoring was carried out mainly according to the planning, and contingency actions were undertaken successfully, as described in a previous section. Note that the continuous adjustment of the support based on observed Q-values and support classes can formally be interpreted as a use of contingency actions, as soon as Class 1 is not used, which here was very common. The term "contingency action" can therefore be misleading for this type of application. Still, this possibility to adjust the amount of support was a key component in the use of the observational method in this tunnel project.

Contractual considerations
The uncertainties involved in projects adopting the observational method implies difficulty being able to describe the geotechnical behaviour and, consequently, estimate the cost and time schedule of the project. This is an economical risk to either the client or the contractor depending on the contractual framework. When using the observational method, the contractual framework needs to be both adapted to the specific problem at hand, e.g. if the work could be described as series or parallel works, and flexible enough to be able to handle changes in design and execution.
If the client wants to reduce the risk exposure, a design-and-build, turnkey, or build-own-transfer contract may be used since most of the risks are owned by the contractor in such contracts (Palmström and Stille 2015). However, Ward, Chapman, and Curtis (1991) argue, based on insurance law and practice, that a contractor should not be expected to price risk that is very difficult to quantify accurately. Instead they recommend that price should be based either on actually encountered conditions or possibly on a contractor's risk premium estimated based on the client's supplied in-depth risk analysis. While large allocation of risk to the contractor makes bidding a challenge, it gives, as discussed by Tidlund (2021), the contractor an opportunity to ensure safe and cost-effective design and execution by using the observational method, since the contractor becomes responsible for both of these project phases. However, if the client transfers too much of the risks to the contractor, and the contractor realises itself being unable to manage such large risks, there is a possibility that the contractor will choose not to submit a tender. According to Ward, Chapman, and Curtis (1991), this may in the long run decrease the competition, encouraging low-quality tenders that do not account for the risk accurately. For further general discussions of the effects of different risk allocation strategies in construction contracts, we refer to, e.g. Hanna, Thomas, and Swanson (2013) and Zhang, Zhang, and Gao (2016); however, the amount of research on risk allocation in underground construction projects specifically is very limited and remains an issue for future studies.
In the Hvalfjörður project, the contract was a turnkey contract where the contractor guaranteed the financing of the project during the construction period, and the tunnel works were paid with a lump sum payment, with a minor remeasurement payment made related to the rock support and grouting. Thus, a considerable part of the risks, such as the construction risks related to the excavation of the tunnel, was owned by the contractor. However, the turn-key contract provided the contractor with the possibility of fully adopting the observational method and to implement the planned contingency measures when necessary without permission from the client, as long as the functional requirements in the tender documents were fulfilled. The project was completed within budget and ahead of schedule, partly due to better geotechnical conditions than expected. It can only be speculated about what would have happened if the geotechnical conditions were worse than expected, but the contractor would certainly have tried to get compensation for the associated extra costs and an extended time schedule. Unfair or unreasonable risk allocation is likely only debated when the ground conditions are poorer than expected; for such instances, it is quite likely that the involved parties will interpret the contract differently (Hartman and Snelgrove 1996). Regardless, in a turn-key contract like this, the client benefits very little from the unexpectedly good conditions and still likely has to pay a considerable risk premium to the contractor.

Concluding remarks
The risk management process and its use of the observational method had an important role in the design and execution of the tunnel under the Hvalfjörður fjord, due to the project team's limited knowledge and experience of subsea tunnelling in Iceland and known failures in similar tunnel projects. The observational method's monitoring of geotechnical hazard indicators ensured safe and cost-effective execution of the project. Continuous probe drilling, grouting ahead of the tunnel face, and systematic geological mapping resulted in flexible and efficient execution of the tunnelling work.
Our analysis of the Hvalfjörður project indicates that the contractor was allocated a considerable amount of risk due to the turn-key contract. This put a significant burden on the contractor to be extremely careful in the analysis of the uncertainty of the ground conditions so that a fair risk premium can be added to the tender. In our opinion, it should in most cases be more favourable for the client to share some risks related to the ground conditions, rather than allocating them to the contractor and risking disputes or, in a worst-case scenario, the contractor's bankruptcy before completion of the project.
Regarding technical planning and execution, we believe that the Hvalfjörður project is a good example of how the observational method can be applied to tunnel projects with substantial geotechnical uncertainty. Although it can be argued that the Hvalfjörður project was completed within budget and ahead of schedule only due to better geotechnical conditions than expected, we believe that the organisation of the project was an important factor for its success. We would here specifically like to highlight the well-thought-out risk management process, the good cooperation between the actors involved, and the systematic planning and undertaking of the monitoring and contingency actions during its design and execution.