Quantitative Risk Assessment for Landslides

Transportation Research Record 1786 ■ Paper No. 02-3900 69 Quantitative Risk Assessment for Landslides William J. Roberds, Ken K. S. Ho, and Eric Leroi Landslides are a serious problem in many parts of the world; they cause public casualties, property damage, and loss of service, and they cost significant resources to prevent and mitigate. Traditional approaches to landslide issues generally work well, but they only implicitly consider the significant uncertainties and consequences involved. An integrated risk assessment and risk management approach corrects this limitation. Quantitative risk assessment (QRA) is an established methodology that provides input for practical risk management of landslide issues and supplements traditional approaches. QRA provides a framework for anticipating and definitively evaluating potential slope-related problems so they can be accepted or cost-effectively mitigated and the residual risks can be communicated. In QRA, the potential slope-related problems (i.e., various types and sizes of landslides) are evaluated with respect to their probability (over a particular period) or frequency of occurrence and their consequences (e.g., public casualties) if they do occur. QRA can be done correctly in various ways, with the best way depending on the specific application, ranging from global issues for developing program policy to site-specific issues for detailed slope design. However, users skilled in geotechnical engineering and risk assessment are needed to perform QRA correctly and cost-effectively and to understand the limitations of the results. Perceived limitations of QRA include accuracy issues, acceptability issues (e.g., by regulators), and aversion issues (e.g., liability). However, many of these issues are easily resolved and are not real limitations, so that QRA, although not perfect, is advantageous in many cases. Landslides are a serious problem in many parts of the world. Landslides, if they occur, can cause adverse consequences, such as public casualties, property damage, and loss of service. Significant resources (e.g., costs or land-use restrictions) can be spent attempting to prevent or mitigate landslides, sometimes without success. The traditional approach to resolving landslide issues is to attempt to prevent slope failures by ensuring that the factor of safety (FS), which is the ratio of the capacity to demand, is greater than 1.0. This is typically done using conservative, “deterministic” (i.e., single-value parameters) stability analyses and requiring a minimum acceptable FS significantly greater than 1.0. This in turn may be supplemented with ad hoc sensitivity and uncertainty analyses, with the uncertainty analysis often done qualitatively. The traditional approach has evolved with time and generally works well for most problems, especially those that are relatively simple or with which the engineer and regulator are familiar. However, there are some limitations of the traditional approach: W. J. Roberds, Golder Associates, Inc., 18300 NE Union Hill Road, Suite 200, Redmond WA 98052. K. K. S. Ho, Geotechnical Engineering Office, Civil Engineering Department of the Government of the Hong Kong Special Administrative Region, 12/F, Civil Engineering Building, 101 Princess Margaret Road, Homantin, Hong Kong. E. Leroi, Geoter International, Espace 890, Route National 96, 13360 Roquevaire, France. • It is deterministic; it generally ignores uncertainty in processes, conditions, and events (e.g., the occurrence of “triggers” that lead to failure, such as critical rainfall or earthquake). Such uncertainties may be large and may have a major effect on the results. What input values (e.g., earthquake magnitude) should be used for design? • It often ignores the consequences of slope failure, using the same minimum acceptable FS for all cases. Should an unimportant slope that has little consequence if it fails be designed to the same standard as a critical slope that has huge consequences if it fails? • It does not produce results that lend themselves to cost–benefit evaluation of alternative actions. For example, how much is increasing the FS from 1.2 to 1.4 worth? Hence, the result may be too conservative (with unnecessary costs or restrictions) or not conservative enough (unsafe). • The basis for acceptability may not be clear. What is the appropriate minimum acceptable FS? Why not something lower or higher? • There is still a chance of failure, even if the calculated FS meets the criteria and the expectation is one of no failure. What is the residual chance of failure? Is this acceptable or should it be reduced? What is the liability? For example, the failure rates in Hong Kong (based on 1997–1998 data) were as follows: – Of 17,000 engineered (post-1977) slopes, all of which presumably were shown to meet FS criteria, 0.12% experienced landslides and 0.032% experienced major failures each year. –Of 37,000 nonengineered (pre-1977) slopes (5,500 of which have since been upgraded), only some of which presumably were shown to meet the FS criteria, 0.61% experienced landslides and 0.064% experienced major failures each year. Geotechnical engineering in general, and for landslides in particular, has the following general considerations: • It is generally performance based (i.e., it ensures adequate performance) rather than procedurally based. This includes consideration of the following: –Multiple performance objectives (cheaper and safer), –“What-if” scenarios and other uncertainties, and –Mitigation (proactive) and contingency plans (reactive). • It increasingly must be defensible, which means it must be transparent. • It is often multidisciplinary with respect to –Predicting various types of consequences (e.g., social issues), – Establishing decision criteria (e.g., regarding acceptability and preferences among consequences), and –Communicating results to multiple stakeholders. • The methods are often empirically based with simplified analyses, and the data are generally limited by cost and logistical issues. 70 Transportation Research Record 1786 Paper No. 02- 3900 INTEGRATED RISK ASSESSMENT AND RISK MANAGEMENT Performance of a particular system (e.g., a set of slopes) can be divided into normal performance (i.e., no problems occur) and abnormal performance (i.e., problems do occur). Obviously, there is significant uncertainty in what abnormal consequences will occur because of the uncertainty in the occurrence and characteristics (type and magnitude) of problems that will actually occur for that system. Risk is the likely set of consequences of all the potential problems, which are not considered as part of the normal performance of that system and may be a function of time. Such risks need to be considered with other factors (normal performance) when determining system configurations based on the predicted performance of the system. The integrated risk assessment and risk management approach is well established in other applications but is less so in geotechnical engineering. It consists of a structured framework that incorporates expert judgement in a transparent way. To be comprehensive, a risk assessment consists of the following steps: 1. Identify all potential problems for the system of interest. 2. Assess their likelihood (over a particular period) or frequency of occurrence. 3. Assess their likely consequences if they do occur. 4. Combine 1, 2, and 3 to determine the likely consequences considering the possibility of all potential problems associated with that system over the particular period. Any significant relationships among the various problems (either in occurrences or in consequences) must be considered for the results to be reasonably accurate. Such risk assessment provides important input to risk management, along with risk tolerability, mitigation options, and evaluations of the normal performance of such options. In risk management, the option that provides the best acceptable combination of performance (considering various types of consequences, both normal and abnormal) logically should be selected. Another useful result of risk assessment for risk management is the sensitivity of risks to various factors. Mitigation options can be identified by focusing on those factors that dominate risk. General benefits of the integrated risk assessment and risk management approach include the following, which alleviate some of the limitations of the traditional approach: • It considers uncertainty (e.g., in triggers and parameters). • It considers consequences. • It is transparent and can be communicated to the public, with care. • It determines residual risk to ensure reasonable expectations on performance (including liability issues). Such risk assessments can be comparative or absolute and can be qualitative or quantitative, as discussed in the following sections. QUALITATIVE RISK ASSESSMENT The qualitative risk assessment approach consists of relatively imprecise assessments of the likelihood or frequency and consequences of potential problems. In some cases, such assessments are semiquantitative (e.g., in terms of categories of values). For example, a qualitative (semiquantitative) risk assessment methodology was used by a government agency in Hong Kong as one of the means to assess natural hillside hazards as a function of the design event that should be used in deterministic analysis of the necessary mitigation works. As shown in Table 1 (with definitions provided in Tables 2 and 3), the appropriate design event is a function of the slope susceptibility (a measure of the likelihood or frequency of landslide) and consequence class (a measure of the consequence of landslides for how close facilities are to the slope and the type of facility, which in turn determines how many public casualties will result). Although potentially useful, most qualitative risk assessments have the following limitations: • They are not transparent and are potentially haphazard. • They cannot determine risk acceptability. • They cannot determine the cost–benefit of alternatives. However, they do provide significant insight and can be used to prioritize potential problems or actions. QUANTITATIVE RISK ASSESSMENT Quantitative risk assessment (QRA) approaches have the following general attributes: • They provide a numerical measure of risk by performance (e.g., additional costs and casualties). • They are rational, structured, and transparent. • They are an established engineering technique. • They can be used to communicate the realities of risk to the public (albeit carefully to avoid misinterpretation). • They can assist in determining the acceptability of the residual risk level (if standards or acceptance of comparable activities can be established). • They satisfy the mandate for QRA (which exists in some cases). • They can be used to determine the cost–benefit of alternative actions (among an identified set of alternatives). • They allow for constrained optimization of risk management (in some cases) to create the best possible alternative. Risk Model The general equation for QRA is as follows: p[C(t )] = ∑ { p[C F] p[ F(t )]} (1) all F TABLE 1 Design Event Based on Qualitative Risk Assessment Susceptibility (see Table 2) A B C D I WCE Consequence Class (see Table 3) II III IV V CE N Note: WCE = adopt worst credible event (WCE) for design, which is a very conservative estimate based on all data in catchment and vicinity, as well as in similar terrain (e.g., geomorphological considerations) (~ 1,000 yrs); CE = adopt conservative event (CE), which is a moderately conservative estimate based on historical data in catchment and vicinity (e.g., air photo interpretation, API) (~ 100 yrs); N = further study not required. Roberds et al. Paper No. 02- 3900 TABLE 2 71 Susceptibility for Use in Determining Design Event for Natural Hillsides * Susceptibility A (extreme) B (high) C (moderate) D (low) Example Description a Signs of instability, continued movement and records of repeated failures (API) Records of occasional failures (API) Few records of failures (API), but with indications of relic failures (geomorphological evidence/other evidence from similar terrain) No records of recent and relic failures, little geomorphological evidence and other evidence from similar terrain Approximate Frequency >1/10 yr 1/10 − 1/100 yr 1/100 − 1/1000 yr <1/1000 yr * See a Table 1. See Ho et al. (1) for more examples. where where F = the occurrence of one of the potential sets of problems (i.e., events of a particular type and magnitude), F(t) = the occurrence of that potential set of problems F during particular period t, p[F(t)] = the “hazard” and is defined as the probability of that potential set of problems F occurring during particular period t, where ∑ p[ F(t )] = 1.0, all F C = the occurrence of consequence of a particular type and magnitude, p[C兩F] = vulnerability and is defined as the relative likelihood (probability distribution) of consequence C occurring if potential problem set F occurs, where ∫ p[C F ] all C dC = 1.0, C(t) = the occurrence of consequence C during particular period t, and p[C(t)] = risk and is defined as the relative likelihood (probability distribution) of consequence C occurring during particular period t, considering all possible failures. Where multiple problems can occur during period t, F is a unique combination of individual problems (recognizing that the individual problems may not be independent), and C must consider combinations of the Cs for all those individual problems (recognizing that not all consequences are additive or independent). Often, however, multiple problems are so unlikely that they can be ignored, and Equation 1 simplifies to consideration of individual problems, not combinations. Various parts of Equation 1 can be discretized without losing much accuracy. For example, if most of the uncertainty is in the occurrence of problems, consequences can be discretized into several (or even one) categories with an average value. For example, the risk model is often simplified to the following, assuming failures are not multiple and are independent: E[C(t )] = ∑ {E[C F]P[ F(t )]} all F (2) E[C(t)] = the mean (probability-weighted average) value of C(t), E[C|F] = the mean value of C if individual failure F occurs, and P[F(t)] = the probability of individual failure F occurring during particular period t. The appropriate way to implement the risk model depends on the application and is determined by the judgment of the risk assessor. To illustrate, consider a particular slope that could (but not necessarily would) fail within a specified period t, resulting in some magnitude of ground disturbance F (in 1000 m2) and associated costs C (in U.S. $1 million). However, the value of F and thus of C over t is uncertain beforehand, as expressed by p[F(t)] and p[C(t)], respectively. As shown in Figure 1, p[C(t)] can be determined by assessing (based on available information) p[F(t)] and p[C|F], which is the uncertainty in what C would be for any particular value of F (if it occurred). In this simple example, C and F have each been discretized into four possible values, although many more could be used. Hazard Assessment In the risk model, p[F(t)] is typically called the hazard. The hazard assessment needs to be comprehensive and suitably refined (for accuracy) but not too detailed (for practicality). It needs to recognize that different scales of a particular type of problem have different frequencies as well as consequences. Such probability (over a particular period) or frequency assessment can be based on the following: • Historical data: These data can be gathered either directly or as relationships (e.g., landslides as a function of rainfall). However, this must be done with judicious classification of the hazards involved and the relevant slope data because of the typically small sample size and changing conditions. • Fault tree analysis: This analysis identifies and evaluates the various possible combinations of causative events. • Other probabilistic performance modeling (e.g., first-order second moment, first-order reliability method, and Monte Carlo TABLE 3 Consequence Class (Combining Facility Type and Location) for Use in Determining Design Event for Natural Hillsides* Proximity (for runout) very close (e.g., α >30o) moderately close (e.g., 30o>α >25o) far (e.g., α<25o) Facility Group No. (see Table 6) (for vulnerability) 5 1&2 3 4 IV I II III II III IV V III IV V V *See Table 1. NOTE: α is the angle below horizontal from crest of slope to nearest point of facility. Transportation Research Record 1786 Paper No. 02- 3900 1 1 1 0.8 0.8 0.8 p[C(t)] p[F(t)] 72 0.6 0.6 p[C|F] 0.4 0.4 0.2 0.2 0.6 0.4 0.2 3 2 0 0 0 FIGURE 1 1 2 3 2 F(t) (1000m ) 0 0 1 2 C (US$1million) F 1 (1000m2) 0 0 1 2 3 C(t) (US$1million) 3 Simple example of general equation for QRA. simulation): For accuracy, however, it is necessary to capture all the significant processes, input uncertainties, and correlations, which might be shown graphically in an influence diagram. Typically, the uncertainty in the minimum FS over a particular period t is assessed, p[FS(t)], from which the probability of failure over that period can be mathematically derived: This is the primary reason why the design event approach was developed in Hong Kong (see Table 1). However, QRA is still an option for these cases, and methods have been developed to use QRA for them, although they are sometimes difficult to apply, and the results may be ambiguous. Still, in some cases, QRA may be the best option for risk management. 1.0 P[ F(t )] = P[ FS(t ) < 1.0] = ∫ p[FS(t )]dFS (3) Consequence Assessment 0 However, p[FS(t)] may be biased and multimodal (because of combining different scenarios), with large uncertainty. • Direct subjective assessment: This needs to be consistent with all available information, including experience. For example, for global QRA of old nonengineered slopes in Hong Kong, the annual probability of failure for different scales and mechanisms was determined from historical data (see Table 4 for fill slopes, which have an overall annual frequency of failure of about 1/525 year). Similarly, the probability distribution for run-out angles was determined using historical data (see Table 5). Hazard assessment for natural hillsides is particularly difficult. It requires an understanding of hillside processes, including the following: • Complex failure mechanisms; • Complex debris travel mechanisms, including changes in movement and volume (entrainment or knock on) along the debris trail; and • Widely varying scale. In the risk model, p[C|F] is typically called the vulnerability. Similar to the hazard assessment, the consequence assessment needs to be comprehensive and suitably refined (for accuracy) but not too detailed (for practicality). It also needs to recognize that different scales of a particular type of problem have different consequences as well as frequencies of occurrence (as previously discussed). However, typically the uncertainty in whether a failure occurs is more important than the uncertainty in the consequences if it does occur, so that mean values of failure consequences are often used (see Equation 2). Also similar to hazard assessment, consequence assessment can be based on the following: • Direct subjective assessment: This assessment needs to be consistent with all available information, including experience. • Event tree analysis: This analysis identifies and evaluates the various possible sequences (or combinations) of follow-on events. • Other probabilistic models: These models need to adequately capture the important processes and uncertainties. Generally, consequence assessments are not historically based because conditions vary too much or change too fast to extrapolate TABLE 4 Annual Probability of Failure of Old Nonengineered Fill Slopes Based on Historical Data (Fraction of Slopes That Failed Each Year) Scale of Failure very minor (<20m3) (70% of total failures) minor (20-50m3) (15% of total failures) major (50-1000m3) (15% of total failures, incl. massive) massive (>1000m3) Mechanism of Failure sliding <10m sliding sliding (45%) liquefaction (10%) washout (45%) sliding liquefaction washout Height of Fill Slope 10-20m 1.10E-03 > 20m 1.90E-04 1.90E-05 - 1.70E-04 3.80E-05 1.70E-04 3.80E-05 3.80E-06 1.70E-06 1.50E-04 7.50E-05 7.50E-05 Roberds et al. Paper No. 02- 3900 TABLE 5 Reasonable Lower Bound (Unspecified Percentile) on Run-Out Angles Based on Historical Data Landslide Volume <300 m3 300 − 3000 m3 >3000 m3 Sliding 30o interpolate 20o Washout or Liquefaction 20o interpolate 10o past data from many cases into the future for a particular case. For example, for global QRA of old nonengineered slopes in Hong Kong, the potential loss of life (PLL, which is the mean of the probability distribution of public fatalities) for various landslide cases was determined based on probabilistic modeling: PLL = ( PLL for ref LS) × (scale factor ) × ( proximity factor ) ( 4) where PLL for ref LS = the PLL for a reference landslide (defined as a 10-m-wide failure of 50 m3 in volume, based on past landslide data in Hong Kong), which has been determined for various land uses (see Table 6), scale factor = the ratio of the width of the debris in the landslide of interest to that of the reference landslide (10 m), and proximity factor = a function of debris mobility (run out) versus facility location. Integrated Risk Assessment and Risk Management In the risk model, hazard and consequence are sometimes defined differently; either • Hazard is landslide detachment/initiation only, whereas vulnerability includes run out and consequence (e.g., in the determination of design events for natural hillsides in Hong Kong, see Table 1); or • Hazard includes detachment and run out, whereas vulnerability is consequence only (e.g., in global QRA of old nonengineered slopes in Hong Kong). As long as one of these definitions is used in a consistent way, such differences do not matter, although they may cause some confusion among different users. TABLE 6 2 3 4 5 QRA is intended to be input for risk management and to provide part of the evaluations needed to make good decisions. In making such decisions, specific criteria are needed, that is, the acceptable level of risk and the value of risk reduction (trade-offs) to determine the cost–benefit of alternatives. However, such decision criteria are not well established. In fact, there is no clear precedence for such criteria for landslide issues. However, precedence for other applications and implied by the traditional approach for resolving landslide issues can be invoked. In Hong Kong, these other precedents have been used to establish interim risk guidelines for natural terrain landslides (for a unit area), which are supplemental and not mandatory. In summary, one of these guidelines specifies the following: • Individual risk (in terms of the chance of fatality for the most exposed or vulnerable person): – <10–5 per year for new developments, and –<104 per year for existing developments; • Societal risk (in terms of fatalities among the population): – Frequency F of N or more fatalities (F–N curve) < 10(–2 – logN), and –<0.01 PLL per year (derived from F–N curve limits); • About U.S. $2.5 million/PLL for cost–benefit to determine “as low as reasonably practical.” Comparative QRAs focus on the difference in risks between alternatives (e.g., the PLLs for Alternatives A and B are unknown but it they are 20% higher for A than B). The results of such comparative QRAs can be used only to prioritize alternatives and cannot determine acceptability or costs–benefits of alternatives (unless they have been adequately calibrated with the results of absolute QRAs). For example, in Hong Kong, the New Priority Classification System is a risk-scoring system (semiquantitative risk assessment) for various types of slopes. It has been calibrated to some extent and is used for prioritizing old slopes. QRA applications range from global assessment for screening, prioritizing, and formulating risk management policies to site-specific assessment for determining cost–benefits of alternatives and acceptability of design. Different levels of accuracy are required for these different applications. In global QRA, the scale of the problem and the relative contribution of different components are evaluated to help policy makers formulate risk management policies, priorities, and resource allocation. Global risks are simply the sum of site-specific risks. However, because of the large number of sites involved, it is impractical to do QRAs for each site. Instead, it is adequate simply to average all the sites for the purpose of appreciating the profile of risk posed by the different categories of slopes. PLL for Reference Landslide for Various Land Uses Facility Group No. 1 Example Description* Buildings (Densely Used), Roads (Very High Traffic Density) Buildings (Lightly Used), Roads (High Traffic Density) Open Space (Densely Used), Roads (Moderate Traffic Density) Open Space (Lightly Used), Roads (Low Traffic Density) Country Parks Roads, (Very Low Traffic Density) LS = reference landslide. * See Ho et al. (1) for more examples. 73 PLL for ref LS 3-6 Proportion of Old Slopes (for global QRA, see Table 7) 10% 1-2 20% 0.25 17% 0.03 33% 0.001 20% 74 Transportation Research Record 1786 Paper No. 02- 3900 For example, a global QRA of old nonengineered slopes in Hong Kong (about 37,000) was conducted. It used the assessed frequency of failure for the various types of slopes (e.g., see Table 4), the probability distribution for run out if such failures occurred (e.g., see Table 5), and the vulnerability to such failures and run out (e.g., Table 6) to determine the average annual risk per slope and the collective annual risk associated with the various types of slopes, as summarized in Table 7. Several site-specific QRA case studies from Hong Kong are presented in Ho et al. (1). Hazard and Risk Maps One useful way to present the results of a QRA is on hazard or risk maps that show hazards or risks as a function of location (e.g., contours). Hazard maps should include debris movement, as well as landslide initiation, and risk maps should include land use (and associated vulnerability). However, scale and liability issues are associated with such maps. At large scales, they are potentially useful for land-use planning, but they are generally not adequate for site-specific decisions (which are much smaller in scale and need more detail). QRA LIMITATIONS inappropriate classification of those data). The possibility and magnitude of such errors can be reduced by appropriate quality assurance and QRA methods, such as helping geotechnical experts to assess the probability of extreme events. For example, an experienced facilitator can be used to elicit consensus assessments from a panel of experts for critical parameters and explicitly identify the basis and limitations of an assessment (i.e., any conditionality) to avoid misunderstanding and to facilitate checking and communication. Although different QRA approaches sometimes produce significantly different results, this generally reflects errors or different assumptions. If done correctly, only the degree of approximation should differ among approaches. However, if a QRA is redone using new information, it would not be unexpected for the results to change (e.g., be refined), but it would be unexpected for the updated results to fall outside the previously predicted range. If the updated results fall outside the previously predicted range, the uncertainties were previously underpredicted (e.g., because important processes were ignored or inappropriately modeled). Many of the perceived accuracy issues (e.g., related to inadequate information, understanding, or models) are equally true for the traditional approach. In fact, in QRA, these are at least acknowledged and can be incorporated in the analysis as additional uncertainties. Acceptability Issues The value of QRA, especially with regard to risk management, has not been fully appreciated in the geotechnical field. Often, this is because of perceived (whether real or not) limitations of QRA. These perceived limitations can be categorized as accuracy issues, acceptability issues, and aversion issues, as discussed in the following sections. Accuracy Issues The results of QRA are not precise, even though they may appear to be. Significant expertise is needed to perform QRA correctly: • Geotechnical expertise is needed to develop an appropriate landslide model that considers both processes and parameters. If the QRA is incomplete (i.e., it does not include all the significant hazards), the risks will be underestimated. Similarly, if the model is oversimplified, accuracy will suffer. QRA will necessarily involve judgment, such as interpretation of historical data, for which QRA provides a framework; potential differences of opinion lead to imprecise results. However, the assessment of extreme (low-probability) events is difficult because of a lack of experience and the uncertainties involved. • Some expertise in probabilistic analysis (which unlike geotechnical engineering is not readily recognized and may not be apparent) is needed to adequately quantify the uncertainties. Otherwise, errors can easily be made in the formulation, in the calculations, or in the inputs (e.g., inadequate interpretation of historical data involving TABLE 7 The results of QRA might not be accepted by the regulators or the public. In addition to perceived accuracy issues, acceptability may be difficult to achieve because the results are difficult to verify (except by peer review, for which there are currently few peers) and there is typical skepticism toward new and unfamiliar techniques. Moreover, there are generally no consensus standards for decision making on risk acceptability and trade-offs, although there is some precedence (both explicit and implicit). For qualitative risk assessment, this is even more true because the qualitative approach is less well defined, with problems often concerning consensus assessments (poor definitions) and the determination of acceptability and cost–benefit. Similarly, for relative (rather than absolute) QRA, whether cardinal (% better) or ordinal (ranking), there is insufficient information to determine acceptability and cost–benefit. Aversion Issues The various participants (practitioners, regulators, or owners) may have an aversion to doing a QRA. In addition to accuracy and acceptance issues, this aversion may be because of the following reasons: • Although potentially useful for a practitioner (e.g., in reducing liability), many practitioners are not trained in probability and do not understand it. Results of Global QRA of Old Nonengineered Slopes in Hong Kong Type of slope Number of slopes Global annual failure frequency Average annual risk (PLL) per slope Collective annual risk (PLL) Distribution of total risk Cut Slopes 19,100 1:100 1.2E-04 2.3 75% Fill Slopes 9500 1:525 4E-05 0.38 13% Retaining Walls 8100 1:360 4E-05 0.32 12% Paper No. 02- 3900 Roberds et al. • Similarly, regulators may not be trained in probability and may not understand it, and hence they cannot check it. Also, their mandate is generally to ensure safety (e.g., through conservative analyses), and unless there is regulatory guidance for risk acceptance, they are generally reluctant to make such a determination and acknowledge risk. • Although it is potentially useful for an owner to know and communicate the risks associated with a particular decision (i.e., develop reasonable expectations of performance), as well as to satisfy regulations (if QRA is mandated), owners are generally reluctant to acknowledge risks (e.g., for liability). Moreover, owners often perceive that a QRA will be expensive and take a long time (especially if acceptance is a problem and the project is schedule driven), although this should not be the case if done appropriately (scaled to the problem). Sometimes owners are disappointed in the results because they have unreasonable expectations on predictability and do not realize that the best that can be done beforehand is to quantify the uncertainties in future performance. Summary of Issues As noted earlier, QRA generally has fewer accuracy issues than the traditional approach, correcting some and sharing others, so that some of the acceptability and aversion issues (especially those related to perceived accuracy issues) should be easily resolved. Although not perfect and not appropriate for all cases, QRA may be the best available tool for some cases. Landslide investigations will be useful in improving landslide QRA by developing a better understanding of the processes involved and historical databases. Education also will be useful in improving landslide QRA accuracy, as well as correcting misperceptions about other QRA limitations. CONCLUSIONS Landslides are a serious problem in many parts of the world; they cause public casualties, property damage, and loss of service, and they cost significant resources to prevent and mitigate. Traditional approaches to landslide issues generally work well, but they only implicitly consider the significant uncertainties and consequences involved. Typically, traditional approaches only implicitly consider the significant uncertainties and consequences involved, so that decisions are sometimes not optimal or difficult to justify and expectations on performance may be unreasonable. An integrated risk assessment and risk management approach, in which QRA provides input to risk management, can correct some of these deficiencies and has been developed to the point that it is a practical way for resolving landslide issues in many (but not necessarily all) cases and supplements traditional approaches. 75 QRA provides a framework for anticipating and definitively evaluating potential slope-related problems so that they can be accepted or cost-effectively mitigated and the residual risks can be evaluated and communicated. Applications range from global issues for developing slope safety policy, which is not easily addressed by traditional approaches, to site-specific issues for detailed slope assessment and design. In QRA, the potential slope-related problems (i.e., various types and sizes of landslides) are evaluated in terms of their probability (over a particular period) or frequency of occurrence and their consequences (e.g., public casualties, property damage, and loss of service) if they do occur. Such QRA can be performed correctly in various ways, with the best way depending on the specific application. However, users skilled in both geotechnical engineering and risk assessment are needed to do QRA correctly, as well as costeffectively, and to understand the limitations of the results. QRA provides input to risk management and is neither an end in itself nor the only input (e.g., other input may include nontechnical, social science aspects). Perceived limitations of QRA include accuracy issues, acceptability issues (caused in part by perceived accuracy issues) and aversion issues (caused in part by perceived accuracy and acceptability issues). However, QRA generally has fewer accuracy issues than the traditional approach, correcting some and sharing others, so that some of the acceptability and aversion issues (especially those related to accuracy issues) should be easily resolved and are not real limitations. Hence, although not perfect, QRA may be valuable in supplementing the traditional approach in many cases. Landslide investigations will be useful in improving landslide QRA by developing a better understanding of the processes and mechanisms involved and historical databases. ACKNOWLEDGMENT This paper is published with the permission of the Director of Civil Engineering, Government of the Hong Kong Special Administrative Region. REFERENCE 1. Ho, K., E. Leroi, and B. Roberds. Quantitative Risk Assessment— Application, Myths and Future Direction. In Proc., GeoEng2000, vol. 1, Melbourne, Australia, Nov. 2000, Technomic Publishing, Inc., Lancaster, Pa., 2000, pp. 269–312. This paper is a condensed version of a paper by Ho et al., “Quantitative Risk Assessment—Application, Myths and Future Direction” (1). The reader is referred to that paper for many of the details (including other references and case studies) that could not be presented here. The longer paper focused on experience in Hong Kong, which is continued here. Publication of this paper sponsored by Committee on Engineering Geology.