A large-scale seismic risk assessment framework using enhanced FEMA P-58 and Bayesian network inference: a case study of District 2, Tehran Authors: Reza Hajirostami, Mehdi Banazadeh Published: 2026 ABSTRACT AbstractSeismic risk assessment is a probabilistic approach that evaluates the likelihood of earthquake occurrence, structural response, expected damage levels, economic losses, and potential casualties by incorporating the inherent uncertainties associated with seismic hazards and urban building characteristics. The primary objective of this study is to quantify and spatially characterize the distribution of damage states at the urban scale. Buildings were classified according to their structural system, age, and number of stories. The structures were initially modeled, analyzed, and designed in ETABS, and the beam and column section properties were extracted for each structural type. Finite element models were subsequently developed in OpenSees, and Incremental Dynamic Analysis, IDA, was performed to evaluate the seismic performance of building groups and large-scale seismic risk. The application of this approach to urban-scale seismic risk evaluation distinguishes this research from similar previous investigations. Given the considerable number of models, the extensive dataset, and the necessity for updating results under varying input conditions, a Bayesian Probabilistic Network was employed. In addition, GIS-based mapping was used to present the findings, including the exceedance probabilities of different damage states and the spatial distribution of collapse probability. The outcomes of this study identify areas that may exhibit relatively higher seismic vulnerability, emphasizing the potential need for targeted retrofitting strategies or, enhanced preparedness for post-earthquake emergency response and rescue operations.Keywords: Large scale seismic risk assessment, Incremental dynamic analysis (IDA), Bayesian network, GIS mappingSubject terms: Engineering, Natural hazards, Solid Earth sciences Introduction Urban-scale seismic risk assessment requires the evaluation of the seismic performance of all buildings within a city, together with consideration of lifeline systems and accessibility networks. This process involves substantial uncertainties, which inherently increase its complexity. It is practically impossible to account for all parameters that influence seismic risk at the urban scale, from uncertainties associated with earthquake occurrence to variability in structural characteristics and urban infrastructure conditions. However, the greater the availability of reliable data and the more comprehensively these parameters are incorporated into the modeling framework, while maintaining computational feasibility, the closer the results will approximate reality. Several earlier studies have attempted to address this problem using different approaches. One prominent example is the Risk Assessment Tools for Diagnosis of Urban Areas against Seismic Disasters, RADIUS, project, which provided nine cities with tools to estimate earthquake damage and manage seismic risk. The RADIUS methodology employs the intensity of past earthquakes to estimate potential future damage. This intensity is measured using the Modified Mercalli Intensity, MMI, scale, which is correlated with a ground motion parameter known as Peak Ground Acceleration, PGA. To estimate PGA values, the project recommends the use of one of three established empirical relationships1–3. For building damage assessment, structures are categorized into ten common structural types that significantly influence seismic performance, and a damage estimation model is developed for each category. However, because RADIUS was designed to be relatively simple, it omits several critical parameters and uncertainties, including detailed earthquake source mechanisms, occurrence probabilities, variations in structural design, and local soil conditions. The exclusion of these influential factors may introduce potential inaccuracies into urban seismic risk assessment outcomes. The Federal Emergency Management Agency, FEMA, developed the HAZUS platform as a standardized framework for estimating losses from earthquakes, hurricanes, and floods. HAZUS is a GIS-based system designed for application across the United States. It provides a consistent procedure for assessing earthquake hazards, estimating physical damage to buildings and infrastructure, and evaluating casualties and economic losses. The method employs spectral analysis to determine the maximum structural response during ground shaking. Building vulnerability is subsequently classified according to four predefined damage states. In a related study, Del Gaudio et al.4 evaluated the performance of reinforced concrete buildings during the L’Aquila earthquake. They developed analytical models to generate pushover curves for vulnerability assessment and defined damage levels according to the EMS-98 scale. Their methodology incorporated random variables and employed Monte Carlo simulation to account for the probability of structural failure. Similarly, Riga et al.5 adopted a different approach to assess earthquake risk at the urban scale. They applied the Capacity Spectrum Method to evaluate the seismic risk of Thessaloniki, Greece. Their study emphasized the quantification of uncertainty in large-scale urban risk assessment and provided a structured probabilistic framework for analysis. Boukri et al.6 conducted a city-wide earthquake risk assessment for buildings in Algeria. Their primary objective was to apply the RADIUS methodology to evaluate urban vulnerability and to estimate the associated economic losses and social impacts. Their findings demonstrated the practical applicability of structured, city-scale assessment methods for evaluating the potential consequences of earthquakes in rapidly developing urban areas. In another study, Bayraktarli et al.7 investigated the application of Bayesian probabilistic networks in earthquake risk assessment. They proposed a framework for risk management across three temporal phases, before, during, and after an earthquake. Using OpenSees, they modeled buildings and compared two categories, retrofitted and non-retrofitted structures. This work demonstrated the capability of probabilistic network models to support decision-making in large-scale urban risk management. In a comprehensive urban risk assessment, they integrated building data within a GIS environment and evaluated 7,186 reinforced concrete moment-resisting frame buildings, incorporating parameters such as building height, floor area, and local soil conditions. The dynamic properties of the buildings were estimated based on the procedures proposed by Priestley8 and the Turkish Seismic Code (1998)9. For retrofitted buildings, strength and displacement capacities were calculated using the methodology presented by Dazio10. Structural performance was evaluated using two primary indicators, maximum displacement and residual displacement. The Bayesian network was subsequently employed to determine the necessity of retrofitting based on the assessed risk levels. Bayraktarli et al.11 applied Bayesian networks to assess earthquake risk in Adapazari, Turkey. They first evaluated the seismic hazard by considering sixteen different combinations of earthquake magnitude and source-to-site distance. For each combination, twenty synthetic ground motion records were generated. Buildings were classified according to the number of stories, year of construction, and occupancy type, such as residential or office use. Dynamic analyses were performed using OpenSees to evaluate structural response. From these analyses, the maximum interstory drift ratio, MIDR, was calculated. Three damage states were defined, and fragility curves were developed to represent the probability of reaching each damage state. Using GeNIe software, Bayesian probabilistic networks were employed to estimate the seismic risk for each individual building. Previous studies that did not incorporate Bayesian networks generally relied on methodologies such as RADIUS, HAZUS, or RISK-E, the European platform for assessing urban risk. These approaches do not typically use nonlinear dynamic analysis to evaluate structural performance. Instead, damage assessment is commonly based on predefined empirical fragility curves derived from observational data. The application of Bayesian networks enables the incorporation of multiple sources of uncertainty, including those related to seismic hazard and urban-scale vulnerability. However, many studies have been limited to simplified representations, such as single-degree-of-freedom systems or basic damage models, primarily because nonlinear dynamic analysis is computationally demanding and detailed structural data at the city scale are often unavailable. Reliance on generalized damage models may therefore introduce inaccuracies into risk estimates. The Federal Emergency Management Agency has established a comprehensive performance-based assessment framework in FEMA P-58 guidelines12. This methodology is fundamentally based on Incremental Dynamic Analysis, IDA, a systematic but computationally intensive procedure. The resulting IDA curves are subsequently used to derive building-specific fragility functions. Within the FEMA P-58 framework, building damage and associated economic losses are estimated while accounting for uncertainties in structural response and collapse fragility13,14. Moreover, the guidelines provide simplified procedures for estimating building response, which are particularly suitable for preliminary design stages and situations where full response history analysis is impractical15. The framework evaluates building-level losses by aggregating component-level damage, including both structural and nonstructural elements16. Nouri et al.17 applied the FEMA P-58 framework to assess the effectiveness of dampers in retrofitting deficient steel moment-resisting frames. Their analyses indicated that the inclusion of dampers increased structural stiffness, reduced yield drift, and prevented collapse under major earthquake intensity levels. Retrofitting also reduced repair time by nearly 50% under severe ground shaking, demonstrating that dampers represent a cost-effective seismic strengthening strategy. Bayesian Network, BN, inference provides a robust framework for modeling the complex and uncertain interdependencies inherent in seismic risk, linking variables such as earthquake intensity, soil liquefaction potential, and building damage18,19. By integrating outcomes from detailed FEMA P-58 performance assessments, Bayesian networks can incorporate critical uncertainties and conditional dependencies, thereby enhancing the accuracy and reliability of comprehensive risk evaluations20. Furthermore, Bayesian approaches facilitate iterative resource allocation strategies, enabling the efficient distribution of computational effort in simulations to achieve a desired level of accuracy with fewer analysis runs20. Research significance In this study, the FEMA P-58 framework is adapted and extended through the introduction of a rational building classification scheme and the integration of a Bayesian probabilistic network to establish a systematic and comprehensive approach for urban seismic risk assessment. The proposed methodology is founded on a transparent analytical basis and can be progressively refined as more detailed structural data become available, thereby enhancing the reliability and precision of the results. In previous research, large-scale seismic risk assessments have predominantly relied on predefined charts and simplified methodologies, while the application of the FEMA P-58 framework in conjunction with Incremental Dynamic Analysis, IDA, for such assessments has not yet been investigated. The principal novelty of this study lies in the integration of this rigorous analytical framework with a Bayesian probabilistic network, thereby establishing a systematic, data-driven, and comprehensive methodology for urban seismic risk evaluation. This approach not only enables the assessment of structural seismic resilience following earthquake events but also facilitates the evaluation of seismic risk across diverse urban areas. The findings of this research provide actionable insights for building insurance companies, municipal authorities, and policymakers, supporting risk-informed decision-making at the strategic level aimed at enhancing urban safety, resilience, and sustainable development. This framework is applied to evaluate the seismic risk of District 2 of Tehran Municipality. Data description The input data used in this study were obtained from multiple authoritative sources and integrated to support the seismic risk assessment framework for District 2 of Tehran. Geographic information, including the spatial distribution of active faults in Tehran, the boundaries of the nine zones of District 2, and the precise location of individual buildings, was acquired in GIS format from the Tehran Disaster Mitigation and Management Organization. In addition, an Excel-based dataset provided by the same organization was used, containing the percentage distribution of each structural type according to the classification defined in this study for the nine zones of District 2 of Tehran Municipality. This dataset, together with the GIS information, formed the core input for large scale seismic risk assessment. The seismic hazard function for Tehran was adopted from the study conducted by Bazarchi et al.21, which provides a region-specific probabilistic seismic hazard model. Damage state probabilities for different performance groups were derived from the Performance Assessment Calculation Tool (PACT) library. These fragility models were used to quantify the relationship between seismic intensity measures and expected damage states for various building categories. Nonlinear structural response parameters, including Incremental Dynamic Analysis (IDA) results and collapse fragility curves, were obtained through dynamic nonlinear analyses. These results were used to characterize structural performance under increasing seismic intensity levels. Finally, a Bayesian Network framework was employed to integrate the aforementioned datasets and propagate uncertainties through the model. Using the outputs from IDA analyses and the probabilistic relationships defined within the Bayesian network, the probability of damage states for each of the nine zones in District 2 of Tehran was estimated. To ensure reproducibility of the proposed framework, all major steps of the analysis, including data preprocessing, structural modeling, Incremental Dynamic Analysis (IDA), and Bayesian Network implementation, were conducted using documented and version-controlled scripts. Structural analyses were performed in OpenSees, while the probabilistic modeling and Bayesian inference were implemented using Python-based libraries. Key modeling assumptions, parameter discretizations, and network structures are explicitly described in this study. The complete set of scripts and computational procedures is publicly available in a DOI-minting repository (see Data Availability section), enabling independent replication and further extension of the proposed methodology. Bayesian probabilistic network Bayesian concept and conditional probability The term Bayesian in this context refers to Bayes’ theorem, attributed to the 18th-century mathematician and philosopher Thomas Bayes, and it forms the basis for updating the probability of an event based on new evidence. In Eq. (1), P(AB) represents the joint probability of events A and B, P(A) is the probability of event A, and P(A∣B) denotes the conditional probability of A given that B has occurred. Bayes’ theorem provides a formal framework for updating event probabilities as new information becomes available. In Eq. (2), the term P(B) is calculated according to the law of total probability22. In this equation, denotes the complement of A. A Bayesian probabilistic network consists of a set of parent and child nodes representing random variables and their causal relationships. The conditional dependencies between the nodes are expressed through Conditional Probability Tables (CPTs). Figure 1 illustrates a simple example of a Bayesian network. In Fig. 1, nodes A1, A2, and A3 are defined as parent nodes. Node A4 is the child of all three parent nodes, while node A5 is the child of A3. Both A4 and A5 act as parent nodes for A6. All these conditional dependencies among the nodes are represented using Conditional Probability Tables (CPTs). Probabilistic inference in Bayesian networks (BNs) can be performed in two ways, forward analysis and backward analysis. Forward analysis computes the probability distribution of each node in the BN based on the prior information of the parent nodes and the conditional probability distributions of the nodes. Backward analysis involves determining the probability distribution of parent nodes given observations at one or more child nodes23. Therefore, in a Bayesian network, it is possible not only to infer the state of a child node from the conditions and information of its parent node(s), but also, by acquiring information from a given node, to perform backward analysis to evaluate the state of the parent nodes. A simplified Bayesian network. (This figure was created by the authors). Construction of the Bayesian network The objective of the Bayesian network in this study is to evaluate urban-scale seismic risk by extending the FEMA P-58 methodology. Accordingly, the nodes defined in the network represent the parameters that influence damage assessment in accordance with this approach. Figure 2 illustrates the Bayesian network employed in this study. Given the large volume of data, the Bayesian probabilistic network was constructed and analyzed in Python using the Bayesian Network and TabularCPD class. Table 1 summarizes the node names, their attributes, possible states, and the conditional probability tables of the parent nodes. The following sections provide detailed descriptions of the nodes and the links connecting them. Bayesian probabilistic network for seismic risk assessment. (This figure was created by the authors). Table 1Bayesian network node specifications.NodeNameTypeStateValueNSNumber of storiesBuilding specifications1 ~ 3 (2Stories)4 ~ 7 (5Stories)8 & more (9Stories)0.42130.53350.0452YBYear builtBuilding specificationsBefore 19901990 ~ 2005After 20050.22120.45810.3207SSStructural systemBuilding specificationsMasonryReinforced ConcreteSteel0.14270.45460.4027RRupture distanceSeismic InformationR<10 km10 km < R<25 km25 km < R<40 km0.135450.648100.21645HCHazard curveSeismic Information0.5 < Sa < 0.80.8 < Sa < 1.151.15 < Sa < 1.551.55 < Sa < 2.02.0 < Sa < 2.5Sa > 2.5The hazard curve is selected based on the natural period of the structures.AccPeak floor accelerationSeismic response parametersPFA < 0.1 g0.1 g < PFA<0.2 g0.2 g < PFA<0.3 g0.3 g < PFA<0.45 g0.45 g < PFA<0.6 g0.6 g < PFA<0.8 g0.8 g < PFA<1.0 gPFA > 1.0 gCPD of the Acc node is calculated based on each state of its parent nodesDriMaximum story driftSeismic response parameters∆<0.010.01<∆<0.020.02<∆<0.030.03<∆<0.040.04<∆<0.050.05<∆<0.060.06<∆<0.07∆>0.07CPD of the Dri node is calculated based on each state of its parent nodesPG1 ~ 16Performance group 1 ~ 16Damage Ratio (Acceleration Base PGs)D < 0.10.1 < D < 0.250.25 < D < 0.50.5 < D < 1.0The amount of each PGi nodes is calculated based on the states of Acc nodePG17 ~ 30Performance group 17 ~ 30Damage Ratio (Drift Base PGs)D < 0.10.1 < D < 0.250.25 < D < 0.50.5 < D < 1.0The amount of each PGi nodes is calculated based on the states of Dri node Bayesian network node specifications. 1 ~ 3 (2Stories) 4 ~ 7 (5Stories) 8 & more (9Stories) 0.4213 0.5335 0.0452 Before 1990 1990 ~ 2005 After 2005 0.2212 0.4581 0.3207 Masonry Reinforced Concrete Steel 0.1427 0.4546 0.4027 R<10 km 10 km < R<25 km 25 km < R<40 km 0.13545 0.64810 0.21645 0.5 < Sa < 0.8 0.8 < Sa < 1.15 1.15 < Sa < 1.55 1.55 < Sa < 2.0 2.0 < Sa < 2.5 Sa > 2.5 PFA < 0.1 g 0.1 g < PFA<0.2 g 0.2 g < PFA<0.3 g 0.3 g < PFA<0.45 g 0.45 g < PFA<0.6 g 0.6 g < PFA<0.8 g 0.8 g < PFA<1.0 g PFA > 1.0 g ∆<0.01 0.01<∆<0.02 0.02<∆<0.03 0.03<∆<0.04 0.04<∆<0.05 0.05<∆<0.06 0.06<∆<0.07 ∆>0.07 D < 0.1 0.1 < D < 0.25 0.25 < D < 0.5 0.5 < D < 1.0 D < 0.1 0.1 < D < 0.25 0.25 < D < 0.5 0.5 < D < 1.0 Node NS This node represents the probabilistic distribution of buildings based on the number of stories. Two-story buildings represent structures with 1 to 3 stories, five-story buildings represent structures with 4 to 7 stories, and nine-story buildings represent structures with 8 stories or more. These categories were modeled and analyzed accordingly. Node NS This node represents the probabilistic distribution of buildings based on the number of stories. Two-story buildings represent structures with 1 to 3 stories, five-story buildings represent structures with 4 to 7 stories, and nine-story buildings represent structures with 8 stories or more. These categories were modeled and analyzed accordingly. Node YB This node represents the construction year of the buildings, categorized according to the edition of the Iranian Seismic Code 2800 in effect. Buildings constructed before 1990 were designed only for gravity loads, buildings constructed between 1990 and 2005 were designed in accordance with the second edition of Code 2800, and buildings constructed after 2005 were modeled and analyzed according to the fourth edition of the Iranian Seismic Code 2800. Node YB This node represents the construction year of the buildings, categorized according to the edition of the Iranian Seismic Code 2800 in effect. Buildings constructed before 1990 were designed only for gravity loads, buildings constructed between 1990 and 2005 were designed in accordance with the second edition of Code 2800, and buildings constructed after 2005 were modeled and analyzed according to the fourth edition of the Iranian Seismic Code 2800. Node SS This node represents the structural system of the buildings. Approximately 14% of the buildings in District 2 of Tehran Municipality are masonry, 46% are reinforced concrete, and 40% are steel structures. Node SS This node represents the structural system of the buildings. Approximately 14% of the buildings in District 2 of Tehran Municipality are masonry, 46% are reinforced concrete, and 40% are steel structures. Node R This node represents the source-to-site distance. Within the framework of the seismic risk assessment methodology employed in this study, this parameter primarily influences the selection of accelerograms. Node R This node represents the source-to-site distance. Within the framework of the seismic risk assessment methodology employed in this study, this parameter primarily influences the selection of accelerograms. Node HC This node represents the seismic hazard curve, which is derived based on regional seismicity, soil type, structural period, and other relevant parameters. It expresses the annual probability of occurrence of different spectral accelerations (Sa). Node HC This node represents the seismic hazard curve, which is derived based on regional seismicity, soil type, structural period, and other relevant parameters. It expresses the annual probability of occurrence of different spectral accelerations (Sa). Nodes Acc and Dri These nodes represent the maximum acceleration and inter-story drift, respectively, obtained from IDA. At this stage, the structures were initially modeled, analyzed, and designed in ETABS according to their number of stories, construction year, and structural system. Table 2 presents the characteristics of the buildings, the design codes, and the structural systems. Tables 3 and 4 provide the dimensions and properties of beams and columns, and the calculated fundamental periods of the steel and reinforced concrete (RC) structures, respectively. Subsequently, considering the source-to-site distance (Node R) in the selection of accelerograms, a total of 81 IDA simulations were performed. The results were then used to construct conditional probability tables for different ranges of acceleration and drift, conditioned on the occurrence of specific spectral acceleration (Sa) intervals, represented by Node HC. Nodes Acc and Dri These nodes represent the maximum acceleration and inter-story drift, respectively, obtained from IDA. At this stage, the structures were initially modeled, analyzed, and designed in ETABS according to their number of stories, construction year, and structural system. Table 2 presents the characteristics of the buildings, the design codes, and the structural systems. Tables 3 and 4 provide the dimensions and properties of beams and columns, and the calculated fundamental periods of the steel and reinforced concrete (RC) structures, respectively. Subsequently, considering the source-to-site distance (Node R) in the selection of accelerograms, a total of 81 IDA simulations were performed. The results were then used to construct conditional probability tables for different ranges of acceleration and drift, conditioned on the occurrence of specific spectral acceleration (Sa) intervals, represented by Node HC. Nodes PGi These nodes represent the characteristics of the performance groups extracted from the Performance Assessment Calculation Tool (PACT) library. Nodes PG1 to PG16 correspond to performance groups that are sensitive to acceleration, while nodes PG17 to PG30 correspond to performance groups that are sensitive to inter-story drift. The list of performance groups is provided in Table 5. Nodes PGi These nodes represent the characteristics of the performance groups extracted from the Performance Assessment Calculation Tool (PACT) library. Nodes PG1 to PG16 correspond to performance groups that are sensitive to acceleration, while nodes PG17 to PG30 correspond to performance groups that are sensitive to inter-story drift. The list of performance groups is provided in Table 5. Table 2Structural characterestics and design codes.Building categoryRepresentative modelTotal height (m)PlanSeismic design codeStructural system1–3 Stories – before 19902 stories6.43 × 6 m*Gravity loads only*Steel IMFRC IMRF4–7 Stories – before 19905 stories16.04 × 6 mGravity loads onlySteel IMFRC IMRF8 or more Stories – before 19909 stories28.85 × 6 mGravity loads onlySteel IMFRC IMRF1–3 Stories – 1990–20052 stories6.43 × 6 mStandard 2800–2nd editionSteel IMFRC IMRF4–7 Stories – 1990–20055 stories16.04 × 6 mStandard 2800–2nd editionSteel IMFRC IMRF8 or more Stories – 1990–20059 stories28.85 × 6 mStandard 2800–2nd editionSteel IMFRC IMRF1–3 Stories – after 20052 stories6.43 × 6 mStandard 2800–4th editionSteel IMFRC IMRF4–7 Stories – after 20055 stories16.04 × 6 mStandard 2800–4th editionSteel IMFRC IMRF8 or more Stories – after 20059 stories28.85 × 6 mStandard 2800–4th editionSteel IMFRC IMRF3 × 6 m*: Three 6-meter bays in each plan direction Gravity Loads only*: No seismic code. Structural characterestics and design codes. 3 × 6 m*: Three 6-meter bays in each plan direction Gravity Loads only*: No seismic code. Table 3Section properties of structural elements and fundamental periods of steel structures.Building categoryStoryColumns Sec.Beam Sec.(I)Computational periodS2* – before 19901,2BOX 160 × 160 × 14PL330 × 8 + 2PL150 × 121.095S2–1990–20051,2BOX 200 × 200 × 16PL330 × 8 + 2PL150 × 120.835S2 – after 20051,2BOX 240 × 240 × 16PL330 × 10 + 2PL150 × 150.689S5* – before 19901,2BOX 200 × 200 × 20PL330 × 8 + 2PL150 × 122.1543BOX 180 × 180 × 16PL330 × 8 + 2PL150 × 124,5BOX 160 × 160 × 16PL330 × 8 + 2PL150 × 12S5–1990–20051,2BOX 280 × 280 × 25PL350 × 10 + 2PL170 × 151.4763BOX 240 × 240 × 20PL350 × 10 + 2PL170 × 154,5BOX 220 × 220 × 20PL330 × 10 + 2PL150 × 15S5 – after 20051,2BOX 300 × 300 × 25PL350 × 12 + 2PL170 × 171.3653BOX 260 × 260 × 20PL350 × 12 + 2PL170 × 174,5BOX 240 × 240 × 20PL330 × 10 + 2PL150 × 15S9* – before 19901,2,3BOX 240 × 240 × 25PL330 × 8 + 2PL150 × 123.4504,5,6BOX 200 × 200 × 20PL330 × 8 + 2PL150 × 127,8,9BOX 180 × 180 × 16PL330 × 8 + 2PL150 × 12S9–1990–20051,2,3BOX 340 × 340 × 30PL380 × 12 + 2PL200 × 182.0934,5,6BOX 300 × 300 × 30PL380 × 12 + 2PL180 × 177,8,9BOX 300 × 300 × 20PL350 × 10 + 2PL170 × 15S9 – after 20051,2,3BOX 400 × 400 × 30PL400 × 15 + 2PL200 × 201.8364,5,6BOX 340 × 340 × 30PL400 × 12 + 2PL200 × 187,8,9BOX 300 × 300 × 25PL350 × 12 + 2PL170 × 17S2*: Tow-Story Steel Structure represents 1–3 Stories.S5*: Five-Story Steel Structure represents 4–7 Stories. Section properties of structural elements and fundamental periods of steel structures. S2*: Tow-Story Steel Structure represents 1–3 Stories. S5*: Five-Story Steel Structure represents 4–7 Stories. Table 4Section properties of structural elements and fundamental periods of RC structures.Building categoryStoryColumns Sec.Beam Sec.Stirrup detailsComputational periodC2 – before 19901,2Cl400 × 400-8Φ14B400 × 400-6Φ14,4Φ14Φ10@2000.666C2–1990–20051,2Cl450 × 450-12Φ16B400 × 500-6Φ16,3Φ16*Φ10@1500.536C2 – after 20051,2Cl450 × 450-12Φ18B400 × 500-5Φ18,3Φ18Φ10@1500.515C5 – before 19901,2,3Cl500 × 500-12Φ16B400 × 500-4Φ16,3Φ16Φ10@2001.3414,5Cl400 × 400-12Φ16B400 × 500-4Φ16,3Φ16Φ10@200C5–1990–20051,2Cl600 × 600-20Φ18B400 × 600-6Φ20,4Φ20Φ12@1501.0503Cl500 × 500-16Φ18B400 × 600-6Φ20,4Φ20Φ12@1504,5Cl500 × 500-12Φ18B400 × 500-4Φ20,3Φ20Φ10@150C5 – after 20051,2Cl600 × 600-20Φ20B400 × 600-6Φ20,4Φ20Φ12@1201.0503Cl500 × 500-16Φ20B400 × 600-6Φ20,4Φ20Φ12@1204,5Cl500 × 500-12Φ20B400 × 500-4Φ20,3Φ20Φ12@150C9 – before 19901,2,3Cl600 × 600-20Φ20B400 × 500-4Φ16,3Φ16Φ10@2002.3954,5,6Cl500 × 500-16Φ20B400 × 500-4Φ16,3Φ16Φ10@2007,8,9Cl500 × 500-12Φ20B400 × 500-4Φ16,3Φ16Φ10@200C9–1990–20051,2,3Cl600 × 600-28Φ20B400 × 700-7Φ20,5Φ20Φ12@1501.7864,5,6Cl600 × 600-20Φ20B400 × 600-7Φ20,5Φ20Φ12@1507,8,9Cl500 × 500-12Φ20B400 × 500-5Φ20,3Φ20Φ10@150C9 – after 20051,2,3,4Cl600 × 600-32Φ20B400 × 700-8Φ20,5Φ20Φ12@1201.7185,6,7Cl600 × 600-24Φ20B400 × 600-7Φ20,5Φ20Φ12@1508,9Cl500 × 500-16Φ20B400 × 500-5Φ20,3Φ20Φ10@150B400 × 500-6Φ16, 3Φ16*: Beam 400 mm (width) × 500 mm (height) reinforced with 6Φ14 (top) and 6Φ14 (bottom). Section properties of structural elements and fundamental periods of RC structures. B400 × 500-6Φ16, 3Φ16*: Beam 400 mm (width) × 500 mm (height) reinforced with 6Φ14 (top) and 6Φ14 (bottom). Table 5Performance groups.CodePG nameTypeAcceleration (Acc)/driftShort descriptionPG1E2022.010Building contentAccUnsecured fragile objects on shelves, unknown restraintPG2E2022.020Building contentAccHome entertainment equipment, unknown installationPG3E2022.023Building contentAccDesktop electronics including computers, monitors, etc.PG4B3011.014ArchitecturalAccClay tile roofPG5B3031.002aArchitecturalAccMasonry Chimney - unreinforcedPG6C3027.001ArchitecturalAccRaised Access Floor, non seismically rated.PG7C3032.003cArchitecturalAccSuspended CeilingPG8D1014.021MechanicalAccHydraulic ElevatorPG9D2021.023bMechanicalAccCold or Hot Potable Water PipingPG10D2022.012aMechanicalAccHeating hot Water PipingPG11D3031.012eMechanicalAccChiller - Vibration isolated equipmentPG12D3041.012aMechanicalAccHVAC Galvanized Sheet Metal DuctingPG13D3052.011bMechanicalAccAir Handling Unit -Unanchored equipmentPG14D4011.032aMechanicalAccFire Sprinkler Drop Standard Threaded SteelPG15D5011.011cMechanicalAccTransformer/primary service - Unanchored equipmentPG16D5012.023eMechanicalAccLow Voltage Switchgear - Vibration isolatedPG17B2022.038ArchitecturalDriftstick-built curtain wallPG18B2023.032ArchitecturalDriftFull tempered Monolithic StorefrontPG19C3011.001aArchitecturalDriftWall Partition, Type: Gypsum + Wallpaper, Full HeightPG20B1041.021aStructuralDriftIMF, Beam one sidePG21B1041.021bStructuralDriftIMF, Beam both sidesPG22B1044.022StructuralDriftconcrete walls with double curtainPG23B1044.082StructuralDriftreinforced concrete walls with boundary columnsPG24B1049.011StructuralDriftReinforced concrete flat slabs- columnsPG25C2011.011bStructuralDriftconcrete stair with no seismic jointPG26B1031.011aStructuralDriftSteel Column Base PlatesPG27B1031.021aStructuralDriftWelded column splicesPG28B1035.001StructuralDriftPost-Northridge RBS connection, beam one side of column onlyPG29B1035.011StructuralDriftPost-Northridge RBS connection, beams both sides of columnPG30C2011.001bStructuralDriftsteel stair without seismic joint Performance groups. S9*: Nine-Story Steel Structure represents 8 or more Stories. Seismic risk assessment As previously noted, the seismic risk assessment in this study is conducted using an enhanced implementation of the FEMA P-58 methodology12. This framework is fundamentally based on the following conditional integral formulation. In this formulation, DV denotes the Decision Variable; DM refers to the Damage Measure; EDP indicates the Engineering Demand Parameter; IM corresponds to the Intensity Measure; and f(IM) represents the probability density function of the selected intensity measure. The overall procedure adopted in this research is summarized in the following steps. Step 1: This stage represents the probabilistic characterization of seismic intensity, derived from the seismic hazard curve, and corresponds to the HC node in the Bayesian network. The seismic hazard function for Tehran was adopted from the study conducted by Bazarchi et al.21 The fundamental period of each structural typology was first calculated, and the corresponding hazard curve, associated with this fundamental period, was obtained through interpolation. To improve the relevance and robustness of the analysis, only spectral acceleration values exceeding 0.5 g were considered. Consequently, the Conditional Probability Distribution (CPD) of the HC node represents the conditional probability of occurrence of different Sa intervals, given that the ground motion intensity exceeds 0.5 g, rather than the annual exceedance probability of Sa. Step 2: In this stage, IDA was performed to derive the seismic demand parameters, with the maximum inter-story drift ratio and peak floor acceleration considered as the primary engineering demand parameters governing structural and non-structural damage. The initial stage of the IDA procedure involves selecting ground motion records consistent with the expected source-to-site rupture distance. For near-field conditions, 14 pairs of ground motion records recommended by PEER were employed. For far-field conditions, a combination of PEER ground motions and additional accelerograms recorded at stations located at distances compatible with the target source-to-site rupture distance was utilized. Specifically, 22 pairs of records were used for source-to-site rupture distances of 15–25 km, and 14 pairs for distances of 25–40 km. A complete list of the selected ground motion records is provided in Table 6. Table 6Ground motion specifications.Record NameMwYearMechanismStation nameImperial Valley6.531997Strike-slipEl Centro Array #6, #7, #11, Delta, Plaster CityIrpinia Italy6.91980Normal–obliqueSturno (STN),Superstition Hills6.541987Strike-slipParachute Test Site, El Centro IMP Co Center, POE, Calipatria Fire StationLoma Prieta6.931989Reverse-obliqueSaratoga - Aloha Ave, Capitola, Gilroy Array, Halls ValleyErzican6.71992Strike-slipErzincanCape Mendocino7.011992ReversePetrolia, Rio Dell Overpass, College of the RedwoodsLanders7.281992Strike-slipLucerne, Yermo Fire Station, Coolwater, Whitewater Trout FarmNorthridge6.691994ReverseRinaldi Receiving Sta, Sylmar - Olive View Med FF, Beverly Hills – 14,145, Canyon Country, LA - Baldwin HillsKocaeli7.511999Strike-slipIzmit, Duzce, Arcelik, IznikChi-Chi7.621999Reverse-obliqueTCU065, TCU102, CHY101, TCU045, CHY087Duzce7.141999Strike-slipDuzce, Bolu, MudurnuKobe6.91995Strike-slipNishi-Akashi, Shin-Osaka, TadokaSan Fernando6.611971ReverseLa Hollywood Stor Lot, Pearblossom PumpFriuli6.51976ReverseTolmezzo, CodroipoManjil7.371990Strike-slipManjil, QazvinHector Mine7.131999Strike-slipHEC Ground motion specifications. The active faults influencing seismic hazard in the study area include the North Tehran Fault, the North Rey Fault, and the South Rey Fault. The probability associated with each source-to-site rupture distance scenario (node R) was computed based on the spatial position of Tehran’s District 2 relative to these faults, assuming a uniform likelihood of seismic activation along the fault length. Incremental dynamic analysis (IDA) To account for the full range of structural member behavior prior to global collapse, the nonlinear behavior of structural materials must be explicitly considered. Nonlinear material modeling is generally performed using either the concentrated plasticity or distributed plasticity formulation24. In nonlinear structural analysis adopting the concentrated plasticity approach, as implemented in this study, the inelastic behavior is assumed to be localized at discrete regions along the members. The nonlinear response of structural elements was represented by assigning rotational spring elements at beam ends and column bases, incorporating the Bilin-Materials constitutive model. This model was originally introduced by Ibarra, Medina, and Krawinkler25 and later refined by Lignos and Krawinkler26 to account for parameters such as asymmetric cyclic deterioration and residual strength. To incorporate the inherent uncertainties in structural response derived from IDA, each ground motion record was appropriately scaled to cover a wide range of seismic intensities. Studies by Shome and Cornell27 demonstrated that the first-mode spectral acceleration serves as a suitable intensity measure for moment-resisting frame systems. Accordingly, in this study, the first-mode spectral acceleration was adopted as the intensity measure. Each ground motion record was initially scaled to multiple intensity levels, and dynamic analyses were then performed for each structural typology under the effects of the scaled records at varying intensities. To provide a comprehensive representation of structural behavior, the Hunt & Fill algorithm was employed28, and the criteria for determining the global collapse point in the IDA were applied as described below. Occurrence of a slope equal to 20% of the initial elastic slope in the Incremental Dynamic Analysis (IDA) curve.Exceeding a maximum inter-story drift ratio of 0.1. Occurrence of a slope equal to 20% of the initial elastic slope in the Incremental Dynamic Analysis (IDA) curve.Exceeding a maximum inter-story drift ratio of 0.1. Numerical divergence of the structural analysis algorithm. Numerical divergence of the structural analysis algorithm. Figures 3 and 4 present the IDA 50% curve (i.e., the median IDA curve) for reinforced concrete and steel structures, respectively, and Figs. 5 and 6 illustrate the Collapse curves for reinforced concrete and steel structures, respectively, which are used to compare and assess the performance of different structural typologies and to illustrate the effects of parameters such as year built, number of stories, structural system, and source-to-site distance. The following observations can be drawn from the comparison of these curves. Median IDA curves for reinforced concrete structures: (a) year built, (b) number of stories, and (c) source-to-site distance. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Median IDA curves for steel structures: (a) year built, (b) number of stories, and (c) source-to-site distance. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Collapse curves for reinforced concrete structures: (a) year built, (b) number of stories, and (c) source-to-site distance. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Collapse curves for steel structures: (a) year built, (b) number of stories, and (c) source-to-site distance. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). New structures, benefiting from the more stringent provisions of contemporary seismic codes, typically exhibit enhanced earthquake performance, with reduced inter-story drifts under a given ground motion intensity and a lower probability of collapse. New structures, benefiting from the more stringent provisions of contemporary seismic codes, typically exhibit enhanced earthquake performance, with reduced inter-story drifts under a given ground motion intensity and a lower probability of collapse. In general, the maximum inter-story drift ratio in steel structures is lower than in concrete structures, with this distinction being more pronounced for high-rise buildings. Assuming all other conditions are equal, the maximum drift and probability of collapse in a 9-story concrete building are significantly higher than in a 2-story concrete building, whereas no substantial difference is observed for steel structures. In general, the maximum inter-story drift ratio in steel structures is lower than in concrete structures, with this distinction being more pronounced for high-rise buildings. Assuming all other conditions are equal, the maximum drift and probability of collapse in a 9-story concrete building are significantly higher than in a 2-story concrete building, whereas no substantial difference is observed for steel structures. Overall, near-field earthquakes induce larger inter-story drifts and higher probabilities of collapse compared to far-field events, with this effect being more pronounced in steel structures. Overall, near-field earthquakes induce larger inter-story drifts and higher probabilities of collapse compared to far-field events, with this effect being more pronounced in steel structures. By performing IDA and extracting the maximum inter-story drift ratio and peak floor acceleration at different spectral acceleration (Sa) levels, the conditional probability tables for the drift and acceleration nodes are obtained. These tables represent the conditional probability of occurrence of various drift and acceleration ranges given different Sa intervals, for different structural typologies and source-to-site distances. The graphs presented in Figs. 7 and 8 serve as examples of these conditional probability tables, illustrating these two parameters for a 5-story concrete and steel building constructed after 2005, with a rupture distance of 15 km < R < 25 km. Overall, it can be observed that, under identical conditions, the maximum drift is higher in the concrete structure, whereas the maximum acceleration is higher in the steel structure. Probability distribution of maximum drift values for 5-story buildings constructed after 2005 with 15 < R < 25: (a) reinforced concrete, (b) steel structure. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Probability distribution of peak floor acceleration values for 5-story buildings constructed after 2005 with 15 < R < 25: (a) reinforced concrete, (b) steel structure. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Step 3: The third stage of the proposed methodology focuses on estimating the percentage of damage based on the calculated seismic demand parameters. To this end, the PACT library provides a comprehensive database derived from past earthquake records and experimental studies of structural elements, non-structural components, and building contents. The elements are classified into acceleration-sensitive and drift-sensitive categories. By integrating this library with Monte Carlo simulations, the conditional probabilities of various damage levels, given different drift and acceleration intervals, were evaluated for 16 acceleration-sensitive and 14 drift-sensitive performance groups. In total, 64,000 Monte Carlo simulations were conducted for each performance group, resulting in the conditional probability distribution tables for the PGi nodes. The graphs in Fig. 9 illustrate the conditional probability tables for the two performance groups B1031.011a and C3011.001a, representing a structural and a non-structural element, respectively, both of which are drift-sensitive. From these graphs, it is evident that the non-structural element experiences significantly higher damage than the structural element under the same drift level. Comparison of the probabilistic distribution of damage levels for drift-sensitive structural and non-structural components. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Bayesian network analysis and results The initial dataset for this study was obtained from the Tehran Emergency Management and Disaster Prevention Organization. District 2 of Tehran Municipality, covering an area of 47.1 km², is subdivided into nine urban zones. Bayesian analyses were performed independently for each zone, utilizing the available data on the number of buildings corresponding to each structural typology. The conditional probability tables (CPTs) of the PGi nodes, representing the output of the Bayesian network, were generated through forward analysis. Subsequent Bayesian evaluation provided the conditional probabilities of varying damage levels for each performance group within the different structural typologies. By separately averaging the results for drift-sensitive and acceleration-sensitive performance groups, the probability distributions of damage levels for each building type were obtained. As presented in Table 1, four damage levels were considered, ranging from slight to complete damage, corresponding to Slight Damage, Moderate Damage, Severe Damage, and Complete Damage, respectively. Damage exceeding 50% was classified as complete damage, as such a level of destruction typically necessitates reconstruction. Figure 10 illustrates the resulting damage probability distributions for various structural typologies. These distributions represent the likelihood of different damage states occurring given a seismic event with Sa > 0.5 g, with damage quantified in terms of the percentage of reconstruction cost relative to the original building. Probabilistic distribution of damage levels for different structures. (This figure was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). By combining the proportion of each structural typology within each urban zone of District 2 of Tehran with the conditional probabilities of damage for both drift-sensitive and acceleration-sensitive components, zone-specific probability distributions of total damage were derived. The resulting damage maps, overlaid with the zoning of District 2, are presented in Fig. 11. Nine zones of District 2 in Tehran with probabilistic damage distribution for acceleration-sensitive (dark grey shade) and drift-sensitive (light grey shade) components of each zone. (The geographic zoning is based on GIS data from the Tehran Municipality Crisis Management Organization. The probabilistic damage distribution bar charts was generated by the authors using the scripts provided in the public repository (DOI: 10.5281/zenodo.19560389)). Limitations Despite the robustness of the proposed seismic risk assessment framework, several limitations related to data availability, modeling assumptions, and computational constraints should be acknowledged. A primary limitation arises from the datasets provided by the Tehran Disaster Mitigation and Management Organization. The GIS dataset obtained from this organization did not include key building attributes such as construction year, number of stories, structural system, or occupancy information. As a result, an additional Excel-based dataset was used, which only provided the percentage distribution of structural types according to the classification defined in this study across the nine subzones of District 2. This constraint limited the level of detail available for building-specific vulnerability characterization and introduced a degree of aggregation uncertainty in the exposure model. Furthermore, the study is restricted to District 2 of Tehran due to data availability, which limits the generalizability of the results to other urban regions. Although the seismic hazard model adopted from Bazarchi et al.21 provides a region-specific probabilistic representation for Tehran, epistemic uncertainties in hazard estimation and ground motion characterization may still influence the overall risk outputs. These uncertainties are partially propagated through the Bayesian Network framework but cannot be entirely eliminated. Due to the large spatial scale of the study and the associated computational demands, simplifying assumptions were introduced in both the Incremental Dynamic Analysis (IDA) and the Bayesian Network modeling. In particular, nonlinear structural analyses in OpenSees were performed using two-dimensional frame models instead of full three-dimensional representations. While this simplification introduces approximation error, it significantly reduces computational cost and enables large-scale probabilistic analysis. Moreover, the Bayesian Network framework does not explicitly include some potentially influential variables, such as liquefaction potential, and building occupancy characteristics. Finally, the accuracy of the proposed framework depends on the structure of the Bayesian Network and the assumed conditional dependencies between variables. Although the model captures key interactions between hazard, fragility, and damage, necessary simplifications for computational tractability may limit its ability to fully represent complex real-world system behavior. Conclusions In this study, a framework for large-scale seismic risk assessment was developed that does not rely on predefined fragility curves or overly simplified modeling approaches. Instead, by enhancing the FEMA P-58 methodology and incorporating Incremental Dynamic Analysis (IDA) alongside a Bayesian probabilistic network, the proposed approach enables a systematic assessment of seismic risk a large scale. Through an appropriate classification of buildings based on construction year, number of stories, and structural system, this approach enables not only the prediction of the seismic performance of individual structures, but also the identification of urban areas with higher vulnerability to earthquake hazards. The results of this assessment can be used to support strategic retrofit planning and to inform disaster risk management policies. The results indicated that the percentage of structural damage tends to decrease with the reduction of building age. Moreover, damage to drift-sensitive performance groups appears to be more significant in reinforced concrete structures, while acceleration-sensitive performance groups show relatively higher damage in steel structures. It is also observed that, particularly in reinforced concrete structures, the extent of damage may increase with building height. Furthermore, for earthquake scenarios with Sa > 0.5 g, Zones 5 and 6 in District 2 of Tehran are expected to experience the highest level of complete damage. The proportion of complete damage in these two zones was estimated to be approximately 44% for drift-sensitive elements and 59% for acceleration-sensitive elements. Access to more comprehensive and detailed primary data, such as accurate structural characteristics of all buildings or more granular building classification information, would further enhance the reliability and precision of the results. Moreover, employing the Bayesian probabilistic network allows the model to be updated as new information becomes available. This includes forward or backward inference adjustments when new data are obtained from any node in the network, such as building-specific properties or observed earthquake damage. Consequently, the probabilities associated with other nodes may be revised, potentially improving the adaptability and accuracy of the seismic risk assessment framework. Author contributions Reza Hajirostami: Conceptualization, Data curation; Formal analysis, Investigation, Methodology; Software, Visualization, Writing – original draft.Mehdi Banazadeh: Conceptualization, Investigation, Methodology, Project administration, Resources, Supervision, Validation; Writing – review & editing. Funding This research received no external funding. Data availability The datasets generated and/or analyzed during the current study are available from the corresponding author Mehdi Banazade (mbanazadeh@aut.ac.ir), upon reasonable request.In addition, all codes, scripts, and processed datasets supporting the findings of this study have been publicly archived in a DOI-minting repository. The repository is accessible at: https://doi.org/10.5281/zenodo.19560389The repository includes: Data preprocessing scripts, Bayesian Network modeling scripts, Seismic risk assessment workflow, Sample input and output datasets.All data sources used in this study are described in detail within the manuscript, along with their respective access links where applicable. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets generated and/or analyzed during the current study are available from the corresponding author Mehdi Banazade (mbanazadeh@aut.ac.ir), upon reasonable request.In addition, all codes, scripts, and processed datasets supporting the findings of this study have been publicly archived in a DOI-minting repository. The repository is accessible at: https://doi.org/10.5281/zenodo.19560389The repository includes: Data preprocessing scripts, Bayesian Network modeling scripts, Seismic risk assessment workflow, Sample input and output datasets.All data sources used in this study are described in detail within the manuscript, along with their respective access links where applicable.