Uncertainty, Reliability and Geotechnical Decision-Making

Introduction and context

Geotechnical engineering is performed under incomplete knowledge. Soil and rock are spatially variable, ground investigations sample only limited locations, laboratory tests idealize field behaviour, and construction can modify drainage, stress history and groundwater conditions. Consequently, a design based on a single “representative” soil parameter may conceal substantial uncertainty.

Uncertainty does not imply that design is impossible. It requires engineers to distinguish what is known from what is inferred, quantify important sources of variation, evaluate the likelihood and consequences of adverse outcomes, and select actions that provide an acceptable balance between safety, cost, constructability and resilience. This is the central purpose of reliability-based geotechnical decision-making.

The subject is strategic because uncertainty affects not only calculations but also the selection of investigation methods, procurement strategy, contingency allowances, monitoring requirements, construction sequencing, insurance, maintenance and eventual decommissioning. The engineer’s task is therefore broader than calculating a factor of safety: it is to govern ground-related risk throughout the asset life cycle.

Learning objectives

By the end of this chapter, the reader should be able to:

1. distinguish aleatory variability from epistemic uncertainty and identify their implications for design;

2. statistically characterize geotechnical data, including dependence, trends and outliers;

3. explain spatial variability using correlation length and random-field concepts;

4. formulate probabilistic limit-state and reliability models;

5. apply first-order reliability concepts and interpret probability of failure;

6. perform sensitivity analysis and identify the value of additional information;

7. construct and use a geotechnical risk register;

8. support design selection, value engineering and the observational method; and

9. establish lifecycle governance for uncertainty, monitoring and decision escalation.

1. The nature and classification of uncertainty

1.1 Variability and lack of knowledge

Aleatory variability is inherent variation in the ground or loading system. Examples include natural variation in undrained strength from one location to another and fluctuations in groundwater level. More data may characterize this variability but cannot eliminate it.

Epistemic uncertainty arises from incomplete knowledge. Examples include uncertain geological boundaries, inadequate sampling, imperfect correlations between index and strength parameters, and uncertainty in the model used for settlement prediction. Investigation, testing, improved modelling and monitoring can reduce epistemic uncertainty.

The distinction is useful but not absolute. A parameter such as “undrained shear strength” may vary naturally and also be uncertain because the testing method, sample quality and anisotropy are imperfectly understood. In practice, both components should be represented transparently rather than treated as an unexplained safety factor.

1.2 Sources of geotechnical uncertainty

Important sources include:

• geological uncertainty: uncertain stratigraphy, buried channels, lenses, faults and weathering profiles;
• measurement uncertainty: sampling disturbance, apparatus error and spatially limited observations;
• parameter uncertainty: uncertainty in mean, variance, distribution and correlation of strength or compressibility;
• model uncertainty: simplifications in bearing-capacity, seepage, deformation or numerical models;
• load uncertainty: uncertain traffic, surcharge, seismic demand, construction loads and future development;
• construction uncertainty: deviations from drawings, workmanship, installation tolerances and unexpected groundwater;
• human and organizational uncertainty: communication failures, inappropriate competence, optimistic assumptions and weak change control.

A mature assessment records these sources separately. Combining every uncertainty into one arbitrary reduction factor makes it difficult to identify useful risk-reduction measures.

2. Statistical characterization of ground parameters

2.1 Descriptive statistics

For observations \(x_1,x_2,\ldots,x_n\), the arithmetic sample mean is

\[

\bar{x}=\frac{1}{n}\sum_{i=1}^{n}x_i.

\]

The unbiased sample variance is

\[

s^2=\frac{1}{n-1}\sum_{i=1}^{n}(x_i-\bar{x})^2,

\]

and the sample standard deviation is \(s=\sqrt{s^2}\). The coefficient of variation is

\[

\mathrm{COV}=\frac{s}{\bar{x}},

\]

provided the mean is meaningful and not close to zero. COV is dimensionless and useful for comparing variables with different units, but it can be misleading for parameters that may be negative, near zero or strongly skewed.

The standard error of the mean, under independent observations with finite variance, is \(s/\sqrt{n}\). It describes uncertainty in estimating the population mean, not the variability of individual soil values. Confusing these two quantities can lead to dangerously narrow design ranges.

Median, quartiles and robust measures such as the median absolute deviation are valuable where data are skewed or contain influential observations. For example, permeability commonly spans several orders of magnitude and is often better analysed on a logarithmic scale.

2.2 Distribution selection

A probability distribution should be selected from physical reasoning, data quality and goodness-of-fit—not convenience alone. A normal distribution may be appropriate for a parameter that is approximately symmetric and unbounded in practice over its relevant range. A lognormal distribution is often used for positive quantities such as hydraulic conductivity, although it remains an approximation. Beta-type or truncated distributions may be suitable for bounded parameters.

A fitted distribution should be checked using graphical methods, quantile comparisons and statistical tests, while recognizing that small samples have low power to distinguish competing distributions. Engineering judgment is required: an apparently good statistical fit in the observed range may give unrealistic tails outside it.

Correlation between variables is equally important. Unit weight, strength and stiffness may not be independent; treating them as independent can overestimate or underestimate reliability. Correlation coefficients should be based on defensible evidence, with scenario analysis used when data are sparse.

2.3 Trend, transformation and representative values

A data set should be examined for trends with depth, elevation, depositional environment or geological unit before calculating a single overall mean. If strength increases with effective overburden stress, a depth-normalized or piecewise model may be more appropriate than one distribution for the entire deposit.

For layered ground, the relevant design value is often not the overall mean. A shallow foundation may be controlled by a weak seam; a slope may be controlled by the weakest continuous mechanism; and a pile group may mobilize resistance through several strata. The “representative value” must therefore be defined in relation to the failure mechanism and the volume of ground involved.

Characteristic values used in codes are not automatically equal to means or lower fractiles. Their definition depends on the code, parameter, failure mode and consequence class. They should not be transferred between design standards without checking the underlying reliability basis.

3. Spatial variability and random fields

3.1 Why location matters

Two soil samples may be statistically similar but have very different design implications depending on their separation and geological continuity. Spatial variability is commonly described by a random field, in which a parameter varies as a function of position.

The autocorrelation function expresses the similarity of values separated by distance. A common conceptual model is

\[

\rho(\Delta x)=\exp\left(-\frac{|\Delta x|}{\theta}\right),

\]

where \(\rho\) is the correlation coefficient at separation \(\Delta x\), and \(\theta\) is a scale parameter called a correlation length under this exponential convention. Other covariance models, such as squared-exponential or Matérn forms, are also possible. The selected model must state its convention because definitions of correlation length differ.

A short correlation length means rapid changes over short distances. A long correlation length indicates more continuous zones. Vertical and horizontal correlation lengths are commonly different because deposition and geological processes produce anisotropy.

3.2 Averaging and failure mechanisms

Spatial averaging can reduce the variance of an averaged parameter, but the reduction depends on the averaging volume and correlation structure. For a line average over length \(L\), the variance is

\[

\mathrm{Var}(\bar{X}_L)=\frac{2}{L^2}\int_0^L (L-r)C(r)\,dr,

\]

where \(C(r)\) is the covariance function and the field is assumed stationary along the line. This relation demonstrates that averaging is not equivalent to simply dividing variance by the number of samples unless observations are independent.

A large loaded area may mobilize an average strength, whereas a narrow slip surface can seek a weak continuous path. Reliability models must therefore match the spatial scale of the failure mechanism. Using an averaged laboratory strength for a failure mechanism controlled by a thin weak layer is unconservative.

3.3 Investigation planning

Investigation locations should be selected to identify geological structure as well as estimate parameter statistics. Closely spaced tests may provide little additional information if they are strongly correlated, while strategically located tests across suspected boundaries may greatly reduce model uncertainty. An investigation plan should state the decisions it is intended to support, not merely the number of boreholes.

4. Limit states and reliability measures

4.1 Limit-state formulation

A limit state separates acceptable and unacceptable performance. Define a performance or limit-state function \(g(\mathbf{X})\), where \(\mathbf{X}\) is a vector of uncertain variables. The convention used here is:

\[

g(\mathbf{X})>0 \quad \text{safe or satisfactory},

\]

\[

g(\mathbf{X})=0 \quad \text{limit state},

\]

\[

g(\mathbf{X})<0 \quad \text{failure or unacceptable performance}.

\]

For a simplified bearing-capacity problem, if \(Q\) is resistance and \(S\) is load effect,

\[

g=Q-S.

\]

For a slope, \(g\) might be resisting moment minus driving moment. For serviceability, it may be an allowable settlement minus predicted settlement. Multiple limit states should be considered: ultimate failure, excessive deformation, hydraulic instability, seismic liquefaction, environmental release and disruption of service.

4.2 Probability of failure and reliability index

The probability of failure is

\[

P_f=P[g(\mathbf{X})\leq 0].

\]

If the limit-state function is normally distributed with mean \(\mu_g\) and standard deviation \(\sigma_g\), then

\[

\beta=\frac{\mu_g}{\sigma_g},

\qquad

P_f=\Phi(-\beta),

\]

where \(\beta\) is the reliability index and \(\Phi\) is the standard normal cumulative distribution function. This direct relationship is valid only when \(g\) is normally distributed, or when the calculation has been transformed into an equivalent standard-normal space with appropriate interpretation.

Reliability index is a comparative measure, not a universal physical constant. A target reliability depends on consequence class, reference period, redundancy, detectability, uncertainty quality and applicable regulation. A numerical target should never replace engineering judgment or statutory requirements.

4.3 First-order reliability method

The First-Order Reliability Method (FORM) transforms basic variables into standardized normal variables and searches for the most probable point on the failure surface. The reliability index is the shortest distance from the origin to that surface in transformed space. For nonlinear limit states, FORM linearizes the surface near this design point.

FORM is efficient for moderate-dimensional problems and provides sensitivity information, but it may be inaccurate for highly nonlinear, discontinuous or multimodal failure surfaces. Second-order methods, response surfaces, importance sampling, subset simulation or direct Monte Carlo simulation may be required.

For a linear limit state with independent normal variables,

\[

g=a_0+\sum_{i=1}^{n}a_iX_i,

\]

\[

\mu_g=a_0+\sum a_i\mu_i,

\qquad

\sigma_g^2=\sum a_i^2\sigma_i^2.

\]

If variables are correlated, covariance terms must be included:

\[

\sigma_g^2=\sum_i\sum_j a_i a_j\,\mathrm{Cov}(X_i,X_j).

\]

This expression shows why ignoring correlation can distort reliability.

4.4 System reliability

A geotechnical asset may fail through several modes. In a series system, failure of any component causes system failure; in a parallel or redundant system, several components must fail before loss of function. Individual mode probabilities cannot simply be added unless dependence and overlap are handled appropriately. A union bound, \(P(\cup F_i)\leq\sum P(F_i)\), is conservative but may be very loose.

Common-cause failures—such as an unexpected high groundwater level affecting both slope stability and basal heave—can dominate system risk. System analysis should therefore identify dependence, not merely list independent hazards.

5. Worked example: reliability of a simplified resistance–load model

Consider a preliminary foundation check with resistance \(Q\) and load effect \(S\), assumed independent and normally distributed:

\[

\mu_Q=1200\ \text{kN},\quad \sigma_Q=120\ \text{kN},

\]

\[

\mu_S=900\ \text{kN},\quad \sigma_S=90\ \text{kN}.

\]

Let \(g=Q-S\). Then

\[

\mu_g=1200-900=300\ \text{kN}.

\]

Because the variables are independent,

\[

\sigma_g=\sqrt{120^2+90^2}

=\sqrt{22500}=150\ \text{kN}.

\]

Thus,

\[

\beta=\frac{300}{150}=2.0.

\]

Using the standard normal distribution,

\[

P_f=\Phi(-2.0)\approx0.0228,

\]

or approximately 2.3 percent for the stated model and reference conditions.

This is not a complete foundation design. The normal assumptions may be inappropriate for a positive resistance, model uncertainty has not been explicitly included, and the values may represent a particular reference period. Nevertheless, the example illustrates an important result: a nominal margin of 300 kN does not by itself indicate reliability; dispersion in both resistance and load controls the probability of an adverse outcome.

If resistance and load were positively correlated, the variance of \(Q-S\) would include the term \(-2\mathrm{Cov}(Q,S)\). Positive correlation would reduce the variance of the difference in this particular formulation, whereas negative correlation would increase it. The physical origin of the correlation must be examined before using such a result.

6. Sensitivity analysis and value of information

6.1 Purpose of sensitivity analysis

Sensitivity analysis identifies which assumptions most influence performance or decision quality. It may be local, examining derivatives near a design point, or global, varying inputs across their distributions. Useful techniques include one-at-a-time analysis, standardized regression coefficients, variance-based indices, tornado diagrams and scenario analysis.

For the linear model above, the contribution of variable \(X_i\) to output variance is related to \(a_i^2\mathrm{Var}(X_i)\), with covariance terms included where relevant. A parameter with a large mean uncertainty but little influence on the limit state may not justify additional investigation; a moderately uncertain parameter with high influence may deserve priority.

6.2 Value of information

Additional investigation has value when it changes a decision or reduces expected loss. The expected value of perfect information is conceptually the difference between the expected outcome under decisions made with perfect knowledge and the expected outcome under current knowledge. The expected value of sample information is the value of a realistic investigation, including its cost and imperfect effectiveness.

A useful decision sequence is:

1. define the decision alternatives and consequences;

2. identify uncertain variables that distinguish the alternatives;

3. estimate how investigation results would update beliefs;

4. determine whether the preferred action would change;

5. compare avoided expected loss with investigation and delay costs.

This prevents the common practice of collecting data without identifying the decision it will improve. Bayesian updating can formalize the process: prior beliefs about a parameter are combined with likelihood from new observations to produce a posterior distribution. The method requires an explicit prior and a defensible observation model; it does not manufacture certainty from poor data.

7. Risk registers for geotechnical projects

A geotechnical risk register is a live governance tool rather than a one-time table. Typical fields include:

• unique risk identifier;
• hazard and initiating cause;
• affected asset, location and construction stage;
• uncertain quantity or event description;
• consequence categories, such as safety, cost, programme, environment and reputation;
• likelihood or probability range;
• severity and detectability;
• existing controls and their owner;
• risk rating before and after controls;
• trigger or leading indicator;
• response action, budget and due date;
• residual risk and escalation route.

Risk matrices are useful for communication but can create false precision. A rating such as “medium” should be supported by a narrative describing assumptions, uncertainty range and consequences. Where credible, quantitative analysis should supplement the matrix.

Risk responses include avoidance, reduction, transfer and acceptance. Transfer through contract or insurance does not remove the physical hazard and may introduce disputes if ground conditions are poorly defined. Acceptance requires an informed authority, a documented rationale and sufficient contingency.

8. Decision-making under uncertainty

8.1 Design selection and value engineering

Value engineering should optimize whole-life function, not merely reduce initial construction cost. A cheaper foundation scheme may have greater sensitivity to groundwater, tighter tolerances, higher monitoring demands or greater disruption if conditions differ from assumptions. A decision model should compare alternatives using:

• expected construction and maintenance cost;
• residual safety and serviceability risk;
• programme and constructability;
• investigation and monitoring requirements;
• carbon, environmental and social impacts;
• reversibility and adaptability;
• exposure to market and regulatory change.

Expected monetary value can be informative, but rare catastrophic outcomes may require risk aversion, minimum safety constraints or robust decision criteria rather than simple averaging.

8.2 The observational method

The observational method is a planned approach in which design assumptions are stated, field behaviour is monitored, and predefined actions are taken if observations approach specified limits. It is not an excuse to construct an incomplete design.

A credible observational-method plan contains:

1. a sufficiently safe initial design;

2. predicted ranges of behaviour and governing mechanisms;

3. measurable parameters and reliable instrumentation;

4. alert and action thresholds;

5. response measures designed and resourced in advance;

6. responsibilities, data review intervals and escalation procedures; and

7. a process for updating the model as evidence accumulates.

For an excavation, observations may include wall movement, strut load, piezometric levels, basal heave indicators and adjacent building settlement. Thresholds should distinguish normal variation from trends requiring intervention. Trigger-action response plans are most effective when actions can be implemented quickly, such as reducing excavation rate, installing additional support, lowering groundwater or imposing a surcharge restriction.

8.3 Robustness and adaptability

A robust design performs acceptably across a range of plausible conditions, even if it is not optimal for the most likely condition. Adaptability allows future actions—such as adding anchors, increasing drainage capacity or modifying operating loads—to be implemented if evidence changes. These properties are particularly valuable where uncertainty is irreducible and consequences are severe.

9. Illustrative case study: deep excavation beside an operating hospital

Background

The following is a fictional illustrative case. A six-storey basement is proposed beside an operating hospital in a dense urban area. The site contains fill over soft alluvial clay, with a variable sand seam and a groundwater table approximately 2 m below ground level. Adjacent structures include a sensitive diagnostic facility and buried utilities.

Problem

The initial value-engineering proposal replaced a stiff diaphragm wall and internal bracing system with a thinner wall and fewer levels of support. The proposal reduced estimated construction cost but increased predicted wall movement and relied on uncertain low-permeability behaviour of the clay. The project team needed to decide whether the saving justified the increased risk.

Analysis

The ground model was divided into geological units rather than treated as homogeneous. Cone penetration tests, high-quality sampling, piezometers and utility surveys were used to characterize strength, compressibility and groundwater connectivity. The sand seam was identified as a potential preferential drainage path. Three limit states were evaluated: basal instability, excessive wall movement and groundwater drawdown affecting neighbouring foundations.

A probabilistic screening model varied undrained strength, permeability, sand-seam continuity, wall stiffness and construction-stage loads. Sensitivity analysis showed that uncertainty in sand-seam continuity and groundwater response influenced risk more than uncertainty in the mean clay unit weight. This finding redirected investigation expenditure toward additional piezometers and targeted drilling rather than more routine density tests.

Approach and solution

The selected scheme retained the stiffer wall but used a staged excavation sequence and optimized support levels. A contingency design for additional pumping and a supplemental internal brace was prepared before construction. The monitoring plan included inclinometers, piezometers, precise levelling points and automated alerts at the hospital.

The risk register assigned ownership to the temporary-works designer, contractor, instrumentation specialist and client project manager. Green, amber and red thresholds were defined. An amber trend required engineering review within 24 hours; a red trigger required work stoppage and implementation of the predefined response.

Outcome

During excavation, piezometers detected higher-than-expected hydraulic response in one zone, while wall movement remained within the predicted envelope. The excavation sequence was locally slowed and pumping was modified to avoid excessive drawdown. No unacceptable hospital movement occurred. The preferred design cost more initially than the rejected thin-wall option, but the decision reduced expected disruption and preserved the hospital’s operational requirements.

Learning points

The case demonstrates that value engineering must consider consequence and controllability, not only first cost. It also illustrates that targeted information can be more valuable than larger quantities of generic data, and that monitoring is effective only when linked to predesigned actions and accountable decision-makers.

10. Governance, ethics and professional responsibility

Uncertainty must be communicated honestly. Reporting a single parameter value without its basis, variability and applicability can mislead clients and contractors. Engineers should distinguish measured facts, interpreted ground models, assumptions and professional judgment.

Ethical practice requires that safety-critical uncertainty is not concealed to achieve a commercial target. Risk acceptance should be assigned to an appropriately authorized party, and residual risks should remain visible through design changes, handover documents and maintenance plans. Contract documents should allocate investigation responsibilities carefully and avoid language that implies impossible certainty about ground conditions.

Regulatory compliance remains mandatory. Probabilistic analysis supplements, but does not override, applicable building regulations, geotechnical standards, occupational safety requirements, environmental controls or permit conditions. Where standards prescribe partial factors or characteristic values, alternative reliability calculations should be reconciled with the code basis rather than substituted casually.

11. Digital methods, AI and contemporary developments

Digital ground models, geographic information systems, cloud-based monitoring and numerical simulation can improve integration of geological and observational data. Machine-learning methods may identify patterns in cone penetration data, classify geological units or detect anomalies in monitoring streams. However, predictions are conditional on training data and may fail outside the geological domain represented in those data.

AI should therefore support, not replace, engineering accountability. A defensible workflow records data provenance, preprocessing, uncertainty bounds, validation performance, domain of applicability and human review. Automated alarms require safeguards against sensor drift, missing data and false positives. Digital twins are useful when they are continuously calibrated and linked to decisions; a visually impressive model without an action framework is not risk governance.

Emerging practice increasingly combines probabilistic numerical modelling, Bayesian updating, remote sensing, distributed fibre-optic monitoring and lifecycle carbon assessment. The research challenge is to integrate these technologies without creating a false impression that more computation eliminates geological uncertainty.

12. Best-practice workflow

A practical lifecycle workflow is:

1. define performance objectives, consequence categories and decision time horizons;

2. develop a geological and hydrogeological conceptual model;

3. identify plausible failure and serviceability mechanisms;

4. plan investigations to discriminate between decisions, not merely to accumulate data;

5. characterize distributions, trends, correlations and spatial scales;

6. formulate limit states and include model uncertainty explicitly where practicable;

7. perform sensitivity and scenario analysis before detailed optimization;

8. compare alternatives using safety, cost, programme, environmental and lifecycle criteria;

9. document assumptions, residual risks, triggers and ownership in a live risk register;

10. implement monitoring and the observational method where it adds controllability;

11. update the model as construction evidence becomes available; and

12. capture lessons for operation, maintenance and future projects.

Challenges and limitations

Reliability calculations can be technically sophisticated but still misleading if the conceptual ground model is wrong. Statistical estimates from small, biased or nonrepresentative samples may imply unjustified precision. Distribution tails are rarely well observed, yet rare events may dominate consequences. Model uncertainty is difficult to quantify and is often double-counted or omitted. Correlation between failure modes can invalidate simplistic system calculations.

There is also a communication challenge. Probability may be interpreted as a guarantee, while a low probability with catastrophic consequences may be unacceptable. Conversely, excessive conservatism can impose unnecessary cost and environmental impact. The appropriate response is transparent assumptions, independent review for high-consequence work, staged investigation and decisions proportionate to risk.

Chapter Summary

Geotechnical design is inherently uncertain because the ground is variable, observations are limited and analytical models are imperfect. Effective practice distinguishes aleatory variability from epistemic uncertainty, characterizes data statistically and spatially, and matches representative parameters to the scale and mechanism of failure.

Limit-state functions provide a foundation for probabilistic analysis. Probability of failure and reliability index can be estimated directly for suitable distributions or through methods such as FORM, simulation and response surfaces. Correlation, spatial continuity, model uncertainty and system dependence are essential considerations.

Sensitivity analysis identifies influential assumptions, while value-of-information thinking directs investigation toward decisions that can actually change. Risk registers convert analysis into accountable governance. Design selection and value engineering should consider whole-life consequences, robustness and adaptability rather than initial cost alone. The observational method is a planned feedback system requiring safe initial provisions, reliable monitoring and predefined actions.

Finally, uncertainty governance continues through construction, operation and maintenance. Digital tools and AI can improve analysis and detection, but they do not replace geological judgment, ethical responsibility, regulatory compliance or clear ownership of decisions.

Key Terms

Aleatory variability: Inherent natural or operational variation that cannot be eliminated by collecting more data.

Epistemic uncertainty: Uncertainty caused by incomplete knowledge and potentially reducible through investigation or improved modelling.

Limit state: Boundary separating satisfactory from unsatisfactory performance.

Limit-state function: Mathematical function defining the performance boundary, commonly written as \(g(\mathbf{X})\).

Probability of failure: Probability that the limit-state function is zero or negative.

Reliability index: A standardized measure of distance from the mean state to the failure surface.

FORM: First-Order Reliability Method, based on linearization near the most probable failure point.

Random field: Spatially varying random quantity representing ground-property variability.

Correlation length: Parameter describing the distance over which values remain statistically related.

Covariance: Measure of joint variation between two random variables.

Characteristic value: Code-defined representative parameter used in design, not necessarily a mean or fixed fractile.

Model uncertainty: Uncertainty arising from simplifications or limitations in the analytical or numerical model.

Sensitivity analysis: Investigation of how outputs respond to changes in uncertain inputs.

Value of information: Expected benefit of obtaining information before making a decision.

Risk register: Live record of hazards, consequences, controls, owners, triggers and residual risk.

Observational method: Planned design and construction approach using monitoring and predefined responses.

Trigger-action response plan: Document linking measured thresholds to specific management actions.

Robustness: Ability to perform acceptably across a range of plausible conditions.

Bayesian updating: Formal combination of prior knowledge and new evidence to revise probability distributions.

System reliability: Reliability of an asset considering multiple interacting or dependent failure modes.

Review Questions

Short Answer Questions

1. Distinguish aleatory variability from epistemic uncertainty and give one geotechnical example of each.

2. Why is the standard error of the mean not a measure of the variability of individual soil values?

3. Define a limit-state function using the convention adopted in this chapter.

4. What information does a correlation length provide in a random-field model?

5. State two essential requirements of a valid observational-method plan.

Long Answer Questions

1. Explain how geological trends, spatial variability and failure mechanisms influence the selection of representative geotechnical parameters.

2. Describe FORM, including its assumptions, interpretation and principal limitations.

3. Discuss how sensitivity analysis and value-of-information reasoning can improve ground-investigation planning.

4. Explain how a geotechnical risk register should be developed, maintained and used during construction.

5. Critically evaluate the role of value engineering in selecting between alternative foundation or temporary-works systems.

Analytical / Application Questions

1. A resistance has mean 1500 kN and standard deviation 150 kN, while the load has mean 1100 kN and standard deviation 100 kN. Assuming independent normal variables and \(g=Q-S\), calculate \(\beta\) and estimate \(P_f\). State the limitations of the calculation.

2. A deep excavation shows increasing wall movement and unexpectedly high piezometric levels near an adjacent building. Develop a trigger-action response plan with at least three monitoring indicators and corresponding actions.

3. A contractor proposes reducing investigation expenditure by omitting targeted drilling across a suspected continuous sand seam. Explain how you would assess the proposal using sensitivity analysis, value of information, consequence assessment and lifecycle risk.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes epistemic uncertainty from aleatory variability?

A. Epistemic uncertainty is inherent random variation, whereas aleatory variability is caused by poor testing.

B. Epistemic uncertainty reflects incomplete knowledge and may often be reduced by investigation.

C. Epistemic uncertainty applies only to loads, while aleatory variability applies only to soil.

D. There is no meaningful distinction in geotechnical engineering.

Correct Answer: B. Epistemic uncertainty reflects incomplete knowledge and may often be reduced by investigation.

Question 2

For independent resistance Q and load S, with g = Q − S, which expression is correct for the variance of g?

A. Var(g) = Var(Q) − Var(S)

B. Var(g) = Var(Q) + Var(S)

C. Var(g) = [Var(Q) + Var(S)]²

D. Var(g) = Var(Q) / Var(S)

Correct Answer: B. Var(g) = Var(Q) + Var(S)

Question 3

If a limit-state function is normally distributed with mean 300 kN and standard deviation 150 kN, what is the reliability index?

A. 0.5

B. 1.0

C. 2.0

D. 4.5

Correct Answer: C. 2.0

Question 4

In the convention used in this chapter, failure occurs when:

A. g > 0

B. g = 1

C. g < 0 or equal to zero

D. the coefficient of variation is zero

Correct Answer: C. g < 0 or equal to zero

Question 5

What is the principal purpose of a correlation length in a spatial-variability model?

A. To define the unit weight of soil

B. To describe the distance over which property values remain statistically related

C. To calculate a factor of safety directly

D. To replace all site investigation

Correct Answer: B. To describe the distance over which property values remain statistically related

Question 6

Which action is most consistent with the observational method?

A. Constructing without a safe initial design because monitoring is available

B. Monitoring behaviour against predefined thresholds linked to planned responses

C. Collecting data only after unacceptable movement occurs

D. Using monitoring to avoid assigning decision responsibility

Correct Answer: B. Monitoring behaviour against predefined thresholds linked to planned responses

Question 7

A risk register is most effective when it includes:

A. Only a numerical risk score

B. Hazards, consequences, controls, owners, triggers and residual risk

C. Only construction costs and programme dates

D. A fixed rating that cannot change during construction

Correct Answer: B. Hazards, consequences, controls, owners, triggers and residual risk

Question 8

Which factor can make an overall mean soil strength inappropriate for design?

A. The failure mechanism is controlled by a localized continuous weak seam

B. The data have been plotted on a graph

C. The soil has a measured unit weight

D. The investigation includes more than one test type

Correct Answer: A. The failure mechanism is controlled by a localized continuous weak seam

Question 9

What is the main purpose of sensitivity analysis?

A. To remove all uncertainty from a design

B. To identify which uncertain inputs most influence the output or decision

C. To guarantee that the selected distribution is normal

D. To replace regulatory requirements

Correct Answer: B. To identify which uncertain inputs most influence the output or decision

Question 10

Which statement about AI in geotechnical risk management is most appropriate?

A. AI predictions are valid regardless of the geological domain of the training data.

B. AI eliminates the need for professional engineering judgment.

C. AI can support anomaly detection and classification but requires validation, uncertainty assessment and human oversight.

D. AI should be used only when no monitoring data exist.

Correct Answer: C. AI can support anomaly detection and classification but requires validation, uncertainty assessment and human oversight.