Conditional spectrum‐based ground motion selection. Part I: Hazard consistency for risk‐based assessments

The conditional spectrum (CS, with mean and variability) is a target response spectrum that links nonlinear dynamic analysis back to probabilistic seismic hazard analysis for ground motion selection. The CS is computed on the basis of a specified conditioning period, whereas structures under consideration may be sensitive to response spectral amplitudes at multiple periods of excitation. Questions remain regarding the appropriate choice of conditioning period when utilizing the CS as the target spectrum. This paper focuses on risk‐based assessments, which estimate the annual rate of exceeding a specified structural response amplitude. Seismic hazard analysis, ground motion selection, and nonlinear dynamic analysis are performed, using the conditional spectra with varying conditioning periods, to assess the performance of a 20‐story reinforced concrete frame structure. It is shown here that risk‐based assessments are relatively insensitive to the choice of conditioning period when the ground motions are carefully selected to ensure hazard consistency. This observed insensitivity to the conditioning period comes from the fact that, when CS‐based ground motion selection is used, the distributions of response spectra of the selected ground motions are consistent with the site ground motion hazard curves at all relevant periods; this consistency with the site hazard curves is independent of the conditioning period. The importance of an exact CS (which incorporates multiple causal earthquakes and ground motion prediction models) to achieve the appropriate spectral variability at periods away from the conditioning period is also highlighted. The findings of this paper are expected theoretically but have not been empirically demonstrated previously. Copyright © 2013 John Wiley & Sons, Ltd.     

Effect of ground motion selection methods on seismic collapse fragility of RC frame buildings

Variation in the seismic collapse fragility of reinforced concrete frame buildings predicted using different ground motion (GM) selection methods is investigated in this paper. To simulate the structural collapse, a fiber‐element modelling approach with path‐dependent cyclic nonlinear material models that account for concrete confinement and crushing, reinforcement buckling as well as low cycle fatigue is used. The adopted fiber analysis approach has been found to reliably predict the loss in vertical load carrying capacity of structural components in addition to the sidesway mode of collapse due to destabilizing P–Δ moments at large inelastic deflections. Multiple stripe analysis is performed by conducting response history analyses at various hazard levels to generate the collapse fragility curves. To select GMs at various hazard levels, two alternatives of uniform hazard spectrum (UHS), conditional mean spectrum (CMS) and generalized conditional intensity measure (GCIM) are used. Collapse analyses are repeated based on structural periods corresponding to initial un‐cracked stiffness and cracked stiffness of the frame members. A return period‐based intensity measure is then introduced and applied in estimating collapse fragility of frame buildings. In line with the results of previous research, it is shown that the choice of structural period significantly affects the collapse fragility predictions. Among the GM selection methods used in this study, GCIM and CMS methods predict similar collapse fragilities for the case study building investigated herein, and UHS provides the most conservative prediction of the collapse capacity, with approximately 40% smaller median collapse capacity compared to the CMS method. The results confirm that collapse probability prediction of buildings using UHS offers a higher level of conservatism in comparison to the other selection methods. Copyright © 2017 John Wiley & Sons, Ltd.     

Introduction of conditional mean spectrum and conditional spectrum in the practice of seismic safety evaluation in China

This paper aims at implementing and introducing the use of conditional mean spectrum (CMS) and conditional spectrum (CS) as the main input parameters in the practice of seismic safety evaluation (SSE) in China, instead of the currently used uniform hazard spectrum (UHS). For this purpose, a procedure for M-R-epsilon seismic hazard deaggregation in China was first developed. For illustration purposes, two different typical sites in China, with one to two dominant seismic zones, were considered as examples to carry out seismic hazard deaggregation and illustrate the construction of CMS/CS. Two types of correlation coefficients were used to generate CMS and the results were compared over a vibration period range of interest. Ground motion records were selected from the NSMONS (2007–2015) and PEER NGA-West2 databases to correspond to the target CMS and CS. Hazard consistency of the spectral accelerations of the selected ground motion records was evaluated and validated by computing the annual exceedance probability rate of the response spectra and comparing the results to the hazard curve corresponding to each site of concern at different periods. The tools developed in this work and their illustrative application to specific case studies in China are a first step towards the adoption of CMS and CS into the practice of seismic safety evaluation in this country.     

A comparison of intensity‐based demand distributions and the seismic demand hazard for seismic performance assessment

This paper compares the seismic demands obtained from an intensity‐based assessment, as conventionally considered in seismic design guidelines, with the seismic demand hazard. Intensity‐based assessments utilize the distribution of seismic demand from ground motions that have a specific value of some conditioning intensity measure, and the mean of this distribution is conventionally used in design verification. The seismic demand hazard provides the rate of exceedance of various seismic demand values and is obtained by integrating the distribution of seismic demand at multiple intensity levels with the seismic hazard curve. The seismic demand hazard is a more robust metric for quantifying seismic performance, because seismic demands from an intensity‐based assessment: (i) are not unique, with different values obtained using different conditioning intensity measures; and (ii) do not consider the possibility that demand values could be exceeded from different intensity ground motions. Empirical results, for a bridge‐foundation‐soil system, illustrate that the mean seismic demand from an intensity‐based assessment almost always underestimates the demand hazard value for the exceedance rate considered, on average by 17% and with a large variability. Furthermore, modification factors based on approximate theory are found to be unreliable. Adopting the maximum of the mean values from multiple intensity‐based assessments, with different conditional intensity measures, provides a less biased prediction of the seismic demand hazard value, but with still a large variability, and a proportional increase the required number of analyses. For an equivalent number of analyses, direct computation of the seismic demand hazard is a more logical choice and provides additional performance insight. Copyright © 2013 John Wiley & Sons, Ltd.     

Seismic performance of wood‐frame houses in south‐western British Columbia

The seismic performance of conventional wood‐frame structures in south‐western British Columbia is analytically investigated through incremental dynamic analysis by utilizing available UBC‐SAWS models, which were calibrated based on experimental test results. To define an adequate target response spectrum that is consistent with information from national seismic hazard maps, record selection/scaling based on the conditional mean spectrum (CMS) is implemented. Furthermore, to reflect complex seismic hazard contributions from different earthquake sources (i.e. crustal events, interface events, and inslab events), we construct CMS for three earthquake types, and use them to select and scale an adequate set of ground motion records for the seismic performance evaluation. We focus on the impacts of adopting different record selection criteria and of using different shear‐wall types (Houses 1–4; in terms of seismic resistance, House 1>House 2>House 3>House 4) on the nonlinear structural response. The results indicate that the record selection procedures have significant influence on the probabilistic relationship between spectral acceleration at the fundamental vibration period and maximum inter‐story drift ratio, highlighting the importance of taking into account response spectral shapes in selecting and scaling ground motion records. Subjected to ground motions corresponding to the return period of 2500 years, House 1 is expected to experience very limited extent of damage; Houses 2 and 3 may be disturbed by minor damage; whereas House 4 may suffer from major damage occasionally. Finally, we develop statistical models of the maximum inter‐story drift ratio conditioned on a seismic intensity level for wood‐frame houses, which is useful for seismic vulnerability assessment. Copyright © 2010 John Wiley & Sons, Ltd.     

Evaluation of the exact conditional spectrum and generalized conditional intensity measure methods for ground motion selection

Two existing, contemporary ground motion selection and modification procedures – (i) exact conditional spectrum (CS‐exact) and (ii) generalized conditional intensity measure (GCIM) – are evaluated in their ability to accurately estimate seismic demand hazard curves (SDHCs) of a given structure at a specified site. The amount of effort involved in implementing these procedures to compute a single SDHC is studied, and a case study is chosen where rigorous benchmark SDHCs can be determined for evaluation purposes. By comparing estimates from ground motion selection and modification procedures with the benchmark, we conclude that estimates from CS‐exact are unbiased in many of the cases considered. The estimates from GCIM are even more accurate, as they are unbiased for most – but not all – of the cases where estimates from CS‐exact are biased. We find that it is possible to obtain biased SDHCs from GCIM, even after employing a very diverse collection of intensity measures to select ground motions and implementing its bias‐checking feature, because it is usually difficult to identify intensity measures that are truly ‘sufficient’ for the response of a complex, multi‐degree‐of‐freedom system. Copyright © 2015 John Wiley & Sons, Ltd.     

Conditional spectrum‐based ground motion record selection using average spectral acceleration

The use of a seismic intensity measure (IM) is paramount in decoupling seismic hazard and structural response estimation when assessing the performance of structures. For this to be valid, the IM needs to be sufficient;that is, the engineering demand parameter (EDP) response should be independent of other ground motion characteristics when conditioned on the IM. Whenever non‐trivial dependence is found, such as in the case of the IM being the first‐mode spectral acceleration, ground motion selection must be employed to generate sets of ground motion records that are consistent vis‐à‐vis the hazard conditioned on the IM. Conditional spectrum record selection is such a method for choosing records that are consistent with the site‐dependent spectral shape conditioned on the first‐mode spectral acceleration. Based on a single structural period, however the result may be suboptimal, or insufficient, for EDPs influenced by different period values, for example, peak interstory drifts or peak floor accelerations at different floors, potentially requiring different record suites for each. Recently, the log‐average spectral acceleration over a period range, AvgSA, has emerged as an improved scalar IM for building response estimation whose hazard can be evaluated using existing ground motion prediction equations. Herein, we present a recasting of conditional spectrum record selection that is based on AvgSA over a period range as the conditioning IM. This procedure ensures increased efficiency and sufficiency in simultaneously estimating multiple EDPs by means of a single IM. Copyright © 2017 John Wiley & Sons, Ltd.     

Occurrence of negative epsilon in seismic hazard analysis deaggregation,and its impact on target spectra computation

This paper investigates circumstances behind the occurrence of negative ε (the normalized difference between the spectral acceleration of a recorded ground motion and the median response predicted by a ground motion prediction equation) in probabilistic seismic hazard deaggregation. Negative ε values are of engineering interest because of their impact on the conditional mean spectrum (CMS), which is a proposed alternative to the uniform hazard spectrum (UHS) as a target spectrum for ground motion selection. In the case where target ε values from deaggregation are positive, the CMS calculation produces relatively lower response spectra than the UHS. Positive target ε values occur almost universally in active seismic regions at long return periods of engineering interest, but the possibility of negative target ε values is important because in the case of negative target ε, some relationships between the CMS and UHS would reverse. This paper describes the calculation of target ε, performs parametric studies to determine when negative ε values occur in deaggregation, and investigates the potential impact on target spectrum calculation and ground motion selection. The case studies indicate that special seismicity models and certain ground motion prediction equations have the most significant effect on ε values and a combination of these characteristics in Eastern North America creates the most likely situation for negative target ε to occur. CMS results are nonintuitive when the target ε is negative, but it is not clear that this is a common practical concern because negative target ε occurs only in well‐constrained areas. Copyright © 2011 John Wiley & Sons, Ltd.     

Probabilistic performance‐based optimum design of seismic isolation for a California high‐speed rail prototype bridge

Previous comparison studies on seismic isolation have demonstrated its beneficial and detrimental effects on the structural performance of high‐speed rail bridges during earthquakes. Striking a balance between these 2 competing effects requires proper tuning of the controlling design parameters in the design of the seismic isolation system. This results in a challenging problem for practical design in performance‐based engineering, particularly when the uncertainty in seismic loading needs to be explicitly accounted for. This problem can be tackled using a novel probabilistic performance‐based optimum seismic design (PPBOSD) framework, which has been previously proposed as an extension of the performance‐based earthquake engineering methodology. For this purpose, a parametric probabilistic demand hazard analysis is performed over a grid in the seismic isolator parameter space, using high‐throughput cloud‐computing resources, for a California high‐speed rail (CHSR) prototype bridge. The derived probabilistic structural demand hazard results conditional on a seismic hazard level and unconditional, i.e., accounting for all seismic hazard levels, are used to define 2 families of risk features, respectively. Various risk features are explored as functions of the key isolator parameters and are used to construct probabilistic objective and constraint functions in defining well‐posed optimization problems. These optimization problems are solved using a grid‐based, brute‐force approach as an application of the PPBOSD framework, seeking optimum seismic isolator parameters for the CHSR prototype bridge. This research shows the promising use of seismic isolation for CHSR bridges, as well as the potential of the versatile PPBOSD framework in solving probabilistic performance‐based real‐world design problems.     

Practice‐oriented estimation of the seismic demand hazard using ground motions at few intensity levels

This paper examines the calculation of the seismic demand hazard in a practice‐oriented manner via the use of seismic response analyses at few intensity levels. The seismic demand hazard is a more robust measure for quantifying seismic performance, when seismic hazard is represented in a probabilistic format, than intensity‐based assessments, which remain prevalent in seismic design codes. It is illustrated that, for a relatively complex bridge–foundation–soil system case study, the seismic demand hazard can be estimated with sufficient accuracy using as little as three intensity measure levels that have exceedance probabilities of 50%, 10% and 2% in 50 years which are already of interest in multi‐objective performance‐based design. Compared with the conventional use of the mean demand from an intensity‐based assessment(s), it is illustrated that, for the same number of seismic response analyses, a practice‐oriented ‘approximate’ seismic demand hazard is a more accurate and precise estimate of the ‘exact’ seismic demand hazard. Direct estimation of the seismic demand hazard also provides information of seismic performance at multiple exceedance rates. Thus, it is advocated that if seismic hazard is considered in a probabilistic format, then seismic performance assessment, and acceptance criteria, should be in terms of the seismic demand hazard and not intensity‐based assessments. Copyright © 2013 John Wiley & Sons, Ltd.     

Alternative non‐linear demand estimation methods for probability‐based seismic assessments

Alternative non‐linear dynamic analysis procedures, using real ground motion records, can be used to make probability‐based seismic assessments. These procedures can be used both to obtain parameter estimates for specific probabilistic assessment criteria such as demand and capacity factored design and also to make direct probabilistic performance assessments using numerical methods. Multiple‐stripe analysis is a non‐linear dynamic analysis method that can be used for performance‐based assessments for a wide range of ground motion intensities and multiple performance objectives from onset of damage through global collapse. Alternatively, the amount of analysis effort needed in the performance assessments can be reduced by performing the structural analyses and estimating the main parameters in the region of ground motion intensity levels of interest. In particular, single‐stripe and double‐stripe analysis can provide local probabilistic demand assessments using minimal number of structural analyses (around 20 to 40). As a case study, the displacement‐based seismic performance of an older reinforced concrete frame structure, which is known to have suffered shear failure in its columns during the 1994 Northridge Earthquake, is evaluated. Copyright © 2008 John Wiley & Sons, Ltd.     

A probabilistic performance‐based approach for mitigating the seismic pounding risk between adjacent buildings

Existing design procedures for determining the separation distance between adjacent buildings subjected to seismic pounding risk are based on approximations of the buildings' peak relative displacement. These procedures are characterized by unknown safety levels and thus are not suitable for use within a performance‐based earthquake engineering framework. This paper introduces an innovative reliability‐based methodology for the design of the separation distance between adjacent buildings. The proposed methodology, which is naturally integrated into modern performance‐based design procedures, provides the value of the separation distance corresponding to a target probability of pounding during the design life of the buildings. It recasts the inverse reliability problem of the determination of the design separation distance as a zero‐finding problem and involves the use of analytical techniques in order to evaluate the statistics of the dynamic response of the buildings. Both uncertainty in the seismic intensity and record‐to‐record variability are taken into account. The proposed methodology is applied to several different buildings modeled as linear elastic single‐degree‐of‐freedom (SDOF) and multi‐degree‐of‐freedom (MDOF) systems, as well as SDOF nonlinear hysteretic systems. The design separation distances obtained are compared with the corresponding estimates that are based on several response combination rules suggested in the seismic design codes and in the literature. In contrast to current seismic code design procedures, the newly proposed methodology provides consistent safety levels for different building properties and different seismic hazard conditions. Copyright © 2012 John Wiley & Sons, Ltd.     

Sliding fragility of block‐type non‐structural components. Part 1: Unrestrained components

Motivated by the development of performance‐based design guidelines with emphasis on both structural and non‐structural systems, this paper focuses on seismic vulnerability assessment of block‐type unrestrained non‐structural components under sliding response on the basis of seismic inputs specified by current seismic codes. Two sliding‐related failure modes are considered: excessive relative displacement and excessive absolute acceleration. It is shown that an upper bound for the absolute acceleration response can be assessed deterministically, for which a simple yet completely general equation is proposed. In contrast, fragility curves are proposed as an appropriate tool to evaluate the excessive relative displacement failure mode. Sample fragility curves developed through Monte‐Carlo simulations show that fragility estimates obtained without taking into account vertical base accelerations can be significantly unconservative, especially for relatively large values of the coefficient of friction. It is also found that reasonable estimates of relative displacement response at stories other than the ground in multistorey buildings cannot in general be obtained by simply scaling the ground acceleration to the peak acceleration at the corresponding storey. Failure modes considered in this study are found to be essentially independent of each other, a property that greatly simplifies assessment of conditional limit states. Copyright © 2002 John Wiley & Sons, Ltd.     

Ground motion selection for simulation‐based seismic hazard and structural reliability assessment

This paper examines four methods by which ground motions can be selected for dynamic seismic response analyses of engineered systems when the underlying seismic hazard is quantified via ground motion simulation rather than empirical ground motion prediction equations. Even with simulation‐based seismic hazard, a ground motion selection process is still required in order to extract a small number of time series from the much larger set developed as part of the hazard calculation. Four specific methods are presented for ground motion selection from simulation‐based seismic hazard analyses, and pros and cons of each are discussed via a simple and reproducible illustrative example. One of the four methods (method 1 ‘direct analysis’) provides a ‘benchmark’ result (i.e., using all simulated ground motions), enabling the consistency of the other three more efficient selection methods to be addressed. Method 2 (‘stratified sampling’) is a relatively simple way to achieve a significant reduction in the number of ground motions required through selecting subsets of ground motions binned based on an intensity measure, IM. Method 3 (‘simple multiple stripes’) has the benefit of being consistent with conventional seismic assessment practice using as‐recorded ground motions, but both methods 2 and 3 are strongly dependent on the efficiency of the conditioning IM to predict the seismic responses of interest. Method 4 (‘generalized conditional intensity measure‐based selection’) is consistent with ‘advanced’ selection methods used for as‐recorded ground motions and selects subsets of ground motions based on multiple IMs, thus overcoming this limitation in methods 2 and 3. Copyright © 2015 John Wiley & Sons, Ltd.     

Hazard-consistent response spectra in the Region of Murcia (Southeast Spain): comparison to earthquake-resistant provisions

Hazard-consistent ground-motion characterisations of three representative sites located in the Region of Murcia (southeast Spain) are presented. This is the area where the last three damaging events in Spain occurred and there is a significant amount of data for comparing them with seismic hazard estimates and earthquake-resistant provisions. Results of a probabilistic seismic hazard analysis are used to derive uniform hazard spectra (UHS) for the 475-year return period, on rock and soil conditions. Hazard deaggregation shows that the largest hazard contributions are due to small, local events for short-period target motions and to moderate, more distant events for long-period target motions. For each target motion and site considered, the associated specific response spectra (SRS) are obtained. It is shown that the combination of two SRS, for short- and long-period ground motions respectively, provides a good approximation to the UHS at each site. The UHS are compared to design response spectra contained in current Spanish and European seismic codes for the 475-year return period. For the three sites analysed, only the Eurocode 8 (EC8) type 2 spectrum captures the basic shape of the UHS (and not the EC8 type 1, as could be expected a priori). An alternative response spectrum, anchored at short- and long-period accelerations, is tested, providing a close match to the UHS spectra at the three sites. Results underline the important contribution of the frequent, low-to-moderate earthquakes that characterize the seismicity of this area to seismic hazard (at the 475-year return period).     

A simulation‐based framework for risk assessment and probabilistic sensitivity analysis of base‐isolated structures

A versatile, simulation‐based framework for risk assessment and probabilistic sensitivity analysis of base‐isolated structures is discussed in this work. A probabilistic foundation is used to address the various sources of uncertainties, either excitation or structural, and to characterize seismic risk. This risk is given, in this stochastic setting, by some statistics of the system response over the adopted probability models and stochastic simulation is implemented for its evaluation. An efficient, sampling‐based approach is also introduced for establishing a probabilistic sensitivity analysis to identify the importance of each of the uncertain model parameters in affecting the overall risk. This framework facilitates use of complex models for the structural system and the excitation. The adopted structural model explicitly addresses nonlinear characteristics of the isolators and of any supplemental dampers, and the effect of seismic pounding of the base to the surrounding retaining walls. An efficient stochastic ground motion model is also discussed for characterizing future near‐fault ground motions and relating them to the seismic hazard for the structural site. An illustrative example is presented that emphasizes the results from the novel probabilistic sensitivity analysis and their dependence on seismic pounding occurrences and on addition of supplemental dampers. Copyright © 2011 John Wiley & Sons, Ltd.     

The effect of causal parameter bounds in PSHA‐based ground motion selection

In this paper the effect of causal parameter bounds (e.g. magnitude, source‐to‐site distance, and site condition) on ground motion selection, based on probabilistic seismic hazard analysis (PSHA) results, is investigated. Despite the prevalent application of causal parameter bounds in ground motion selection, present literature on the topic is cast in the context of a scenario earthquake of interest, and thus specific bounds for use in ground motion selection based on PSHA, and the implications of such bounds, is yet to be examined. Thirty‐six PSHA cases, which cover a wide range of causal rupture deaggregation distributions and site conditions, are considered to empirically investigate the effects of various causal parameter bounds on the characteristics of selected ground motions based on the generalized conditional intensity measure (GCIM) approach. It is demonstrated that the application of relatively ‘wide’ bounds on causal parameters effectively removes ground motions with drastically different characteristics with respect to the target seismic hazard and results in an improved representation of the target causal parameters. In contrast, the use of excessively ‘narrow’ bounds can lead to ground motion ensembles with a poor representation of the target intensity measure distributions, typically as a result of an insufficient number of prospective ground motions. Quantitative criteria for specifying bounds for general PSHA cases are provided, which are expected to be sufficient in the majority of problems encountered in ground motion selection for seismic demand analyses. Copyright © 2016 John Wiley & Sons, Ltd.     

Closed‐form aftershock reliability of damage‐cumulating elastic‐perfectly‐plastic systems

Major earthquakes (i.e., mainshocks) typically trigger a sequence of lower magnitude events clustered both in time and space. Recent advances of seismic hazard analysis stochastically model aftershock occurrence (given the main event) as a nonhomogeneous Poisson process with rate that decays in time as a negative power law. Risk management in the post‐event emergency phase has to deal with this short‐term seismicity. In fact, because the structural systems of interest might have suffered some damage in the mainshock, possibly worsened by damaging aftershocks, the failure risk may be large until the intensity of the sequence reduces or the structure is repaired. At the state‐of‐the‐art, the quantitative assessment of aftershock risk is aimed at building tagging, that is, to regulate occupancy. The study, on the basis of age‐dependent stochastic processes, derived closed‐form approximations for the aftershock reliability of simple nonevolutionary elastic‐perfectly‐plastic damage‐cumulating systems, conditional on different information about the structure. Results show that, in the case hypotheses apply, the developed models may represent a basis for handy tools enabling risk‐informed tagging by stakeholders and decision makers. Copyright © 2013 John Wiley & Sons, Ltd.     

Improved risk‐targeted performance‐based seismic design of reinforced concrete frame structures

This paper presents a procedure for seismic design of reinforced concrete structures, in which performance objectives are formulated in terms of maximum accepted mean annual frequency (MAF) of exceedance, for multiple limit states. The procedure is explicitly probabilistic and uses Cornell's like closed‐form equations for the MAFs. A gradient‐based constrained optimization technique is used for obtaining values of structural design variables (members' section size and reinforcement) satisfying multiple objectives in terms of risk levels. The method is practically feasible even for real‐sized structures thanks to the adoption of adaptive equivalent linear models where element‐by‐element stiffness reduction is performed (2 linear analyses per intensity level). General geometric and capacity design constraints are duly accounted for. The procedure is applied to a 15‐storey plane frame building, and validation is conducted against results in terms of drift profiles and MAF of exceedance, obtained by multiple‐stripe analysis with records selected to match conditional spectra. Results show that the method is suitable for performance‐based seismic design of RC structures with explicit targets in terms of desired risk levels.     

Site dependence and record selection schemes for building fragility and regional loss assessment

When performing loss assessment of a geographically dispersed building portfolio, the response or loss (fragility or vulnerability) function of any given archetype building is typically considered to be a consistent property of the building itself. On the other hand, recent advances in record selection have shown that the seismic response of a structure is, in general, dependent on the nature of the hazard at the site of interest. This apparent contradiction begs the question: Are building fragility and vulnerability functions independent of site, and if not, what can be done to avoid having to reassess them for each site of interest? In the following, we show that there is a non‐negligible influence of the site, the degree of which depends on the intensity measure adopted for assessment. Employing a single‐period (e.g., first‐mode), spectral acceleration would require careful record selection at each site and result to significant site‐to‐site variability of the fragility or vulnerability function. On the other hand, an intensity measure comprising the geometric mean of multiple spectral accelerations considerably reduces such variability. In tandem with a conditional spectrum record selection that accounts for multiple sites, it can offer a viable approach for incorporating the effect of site dependence into fragility and vulnerability estimates. Copyright © 2017 John Wiley & Sons, Ltd.