Systemic risks, vulnerability analysis and integrated disaster risk management for disaster resilient societies

Nadejda Komendantova, Tatiana Ermolieva, Taher Zobeydi (IIASA, Laxenburg,  Austria); Seda Kundak, Ali Yılmaz, Cihan Mert Sabah (Istanbul Technical University, Istanbul, Turkey); Iuliana Armas, Dragos Toma-Danila (University of Bucharest, Bucharest, Romania); Marcel Hürlimann, Nieves Lantada, Amparo Nunez (Barcelona UPC, Barcelona, Spain) Early Warning Centre (EWC) at KNMI

Introduction

The alarming tendency of increasing systemic losses due to the combination of natural and human-induced risks calls for risk-based approaches to economic developments and catastrophe management, for the design of robust interdependent ex-ante mitigation and ex-post emergency response options, to deal with the risks of all kinds. The measures aim to decrease societal vulnerabilities, increase resilience and Building Back Better (BBB) capacity.

The systemic risks and vulnerability analysis methodology for PARATUS (Promoting disaster preparedness and resilience by co‐developing stakeholder support tools for managing the systemic risk of compounding disasters, https://iiasa.ac.at/projects/paratus) project, WP2 (Responsible: IIASA, Cooperation and Transformative Governance group of Advanced Systems Analysis Program, CAT-ASA) includes innovative approaches to exposure and vulnerability analysis as well as the development of methods and tools enabling projecting individual and systemic vulnerabilities into the future taking into account multiple vulnerability criteria and drivers, which can be modified by dynamically changing risks and vulnerabilities reduction measures (Komendantova et al., 2025a,b; Ermolieva et al., 2025).

In this article we describe the ongoing work on the development of methodologies, models and modeling tools for the current and the future vulnerability analysis by dealing with projections of key vulnerability drivers under alternative development scenarios (without and with mitigation measures) such as, e.g., Business-as-Usual (BAU, without accounting for risks), Risk Averse (RA, accounting for risks), and other (section 2). The methodologies and modeling tools are briefly illustrated in section 3 using data and relevant studies in earthquake-prone cities Istanbul (Turkey) and Romania (Bucharest).

2. Future vulnerability scenarios

Vulnerability scenarios are methods and ways to characterize plausible states (of economic systems, regions, society, individuals) of being, in some sense, vulnerable in the future. Future development trends and vulnerability scenarios facilitate a discussion of current conditions and potential future spatio-temporal socio-economic, technological, demographic development trends, which can affect future vulnerabilities. The process of developing future vulnerability scenarios also allows for a discussion of desirable or non-desirable scenarios of “futures”, the development patterns and the respective vulnerabilities.

The concept of vulnerability in the context of natural disasasters and/or compond (systemic) natural and human-induced disasters expresses the multi-dimensionality of disasters’ impacts&damages on exposed systems and actors, e.g., socio-economic, physical (infrustructure, buildings), environmental, institutional, financial, etc. The analysis of the multifaceted vulnerability profiles goes beyond simple risk analysis as it considers the complex interplay of various factors like social, economic, and environmental conditions that make individuals, communities, or systems susceptible to harm.

2.1. Multifaceted vulnerability profiles: current and future. Investigating and measuring this multi-faceted vulnerability profile is a complex and valuable task requiring to identfy the respective vulnerability drivers such as, e.g., spatio-temporal economic developments and population growth patterns, incomes (profiles by population age groups), male/female/children/elderly dependencies ratios, literacy/education level (socio-economic vulnerability), infrastructure safety and reliability (physical or structural vulnerability), natural resource availability (environmental), financial capacity and institutional developments (institutional, financial), etc. Accomplishing the task enables to plan and allocate mitigation (structural and financial) measures and resources where they are critically needed to decrease societal vulnerabilities and increase resilience, to better face the risks and achieve future sustainable development goals.

In the presence of risks of all kinds, regional past, current, and future development trends (and therefore, vulnerabilities) can evolve along the four rather broad story lines, which influence the (sustainable) development planning and the future vulnerabilities, individual and systemic (compound):

1: Area (urban or rural) development plan focuses solely on risk-adjusted and hazard-sensitive (risk-averse) development strategies without paying attention to current economic and urban/rural growth tendencies/trends;

2: Area development plan pays adequate attention to the economic and urban/rural growth tendencies and also envisages investments into and implementation of risk-adjusted (risk-averse) measures;

3: Area development plan does not invest into hazard-mitigation measures and also does not prioritize economic growth trends;

4: Area development plan highly prioritizes economic and growth trends as they are and does not account for possible hazards, therefore, does not implement risk-adjusted planning and investments into hazard mitigation and resilience-building measures.

There could be other story lines accounting for specific characteristics of case study areas. Vulnerability analysis approaches for Bucharest (Romania) and Istanbul (Turkey) have been proposed in Armas et al., 2017; Erdogan and Terzi, 2022; Kundak et al., 2025) relying on current sitiation and data. From the discussions with the case study teams, the future “vulnerability” scenarios can be characterised by different scenario combinations of factors (criteria and sub-criteria) affecting the individual and the systemic vulnerabilities:

• Socio-economic vulnerability scenario includes future scenarios of population growth and (re)alocation, increasing incomes, social security and health care benefits, etc.
• Structural (physical) vulnerability scenario includes possible buildings’ reinforcements, construction of new building according to seismic norms, etc.
• Infrastructure vulnerability scenario includes future plans regarding expansion/shrinking of transportation and energy provision networks, etc.
• Environmental vulnerability scenario can include green areas expansion, conversion of specific land use types, etc.,
• Systemic vulnerability scenarios integrate all combinations of individual vulnerability scenarios with respective projections of drivers.

2.2. Multifaceted vulnerability profiles: urban areas vulnerability. In urban areas the damages (and therefore the vulnerabilities) triggered by natural disasters can occur due to a general layout of a city, urban texture (the arrangement of buildings, streets, road network, bridges, green spaces, and other urban elements), usage areas, existing houses’ composition (wooden, steel, stone, reenforced concrete, etc.), transportation systems and infrastructure, planning and management weaknesses in a city. Urban vulnerability indicators (criteria) can be divided into social, economic, environmental, physical, systemic vulnerability criteria. The systemic vulnerability criteria is a compund indicator combining all the individual vulnerabilities.

2.3. Current vulnerability analysis. The current vulnerability analysis (Armas et al., 2017; Erdogan and Terzi, 2022; etc.; Kundak et al., 2025) utilize historical data on the vulnerability criteria, sub-criteria and indicators (vulnerability drivers) at the level of tracts (Bucharest) and neighborhoods (Istanbul), which are fine-scale administrative units. The criteria and sub/criteria are similar to those in Figure 1.

In studies by Armas et al. (2017), the socio-economic vulnerability included “four social vulnerability dimensions – social, education, housing, and social dependence …”. The four criteria were constructed based on the following sub-criteria (drivers) data: social – dwelling population density, widows female population in total population, elderly (e.g., over 65 years), female population in total population, room occupancy per household; educational – minimum level of education, unemployed population (inactive population), women with more than 3 children (in total women who gave birth); housing – housing density, average room area per person on census tract, average household room area on census tract, average no. of private/owned houses, number of rooms, etc.

In Istanbul urban area (Erdogan and Terzi, 2022), the vulnerability assessment by neighborhoods of Istanbul (Turkey), four main criteria (critical urban services, infrastructure facilities, structural, socioeconomic) were identified and further subdivided into sub-criteria. For example, the “socioeconomic” criterion comprised population density, daytime density, average household size, education status, average household income, population over 65, child population ratio, women population ratio. The “critical urban services” consist of road network, distance to fuel stations, accessibility to fire stations (in m), accessibility to police stations (in m), accessibility to open spaces (in m); criterion “infrastructure” included damaged electricity line length (in km), damage distribution of water pipelines, damage ratio, damage distribution of natural gas pipelines; and structural criteria included such sub-criteria as building age, building construction type, building density (building/ha), Peak Ground Acceleration (PGA-gal).

Very often, the current vulnerability analysis is based on the Principle Component Analysis (PCA), Multi-criteria decision analysis (MCDA), the Analytical Hierarchy Process (AHP) method, statistical and machine learning (SML) approaches.

The collection of data was made possible through household surveys and questionaries about current status of the vulnerability “drivers” (sub-criteria), both are time and resource consuming processes, which cannot be repeated frequently.

In studies regarding future vulnerabilities, the data at required resolutions (e.g., tracts, neighborhoods, households) can be missing. Therefore, the dynamic revision of vulnerability indices in the context of newly introduced risk mitigation measures and socio-economic, demographic, technological development trends (drivers) can become difficult.

2.4. Future vulnerability analysis: Statistical and machine learning (SML) approaches. For the development of future vulnerability scenarios when the necessary data can depend on multiple uncertain global and local drivers and trends, the analysis can rely on statistical and machine learning (SML) models.

SML approaches to the analysis of vulnerabilities can effectively supplement traditional vulnerability analysis and modeling methods. The SML models can be used to analyze future socio-economic, structural and other vulnerabilities based on projections of respective vulnerability drivers (called “criteria” in Armas et al., 2017 and in Erdogan and Terzi, 2022). For example, the socio-economic vulnerability is affected by such factors (drivers, criteria) as: social, educational, housing, and social dependence. The sub-drivers (sub-criteria) of these factors are: social – dwelling population density, widows female population in total population, elderly people, female population in total population, room occupancy per household; educational – minimum level of education, unemployed population (inactive population), women with more than 3 children (in total women who gave birth); housing – housing density, average room area per person on census tract, average household room area on census tract, average no. of private/owned houses, number of rooms, etc.

For modeling and predicting structural vulnerability, the number of building stores, building material, distance to epicenter, and other variable characterizing structural and physical properties of buildings and earthquakes can be used in statistical methods to explain the damage volume. In some regions, also foundation type, land type, roof type, ground floor type, and superstructure type based on construction materials are considered as predictor variables.

Taking into account that the data on drivers (criteria) and sub-drivers (sub-criteria) are massive, the SML models can be trained based on pre-calculated vulnerability indicators and relevant covariates. Then, the SML vulnerability models can be used to study future vulnerabilities assuming different scenarios of drivers (in SML also called predictors, covariates).

3. Future vulnerability scenarios: spatio-temporal development limitations, vulnerability drivers (criteria and sub-criteria).

Trained SML models can serve as a Scenario Analysis Tool of plausible future socio-economic and structural vulnerabilities, and they enable the following:

1. Accounting for projections/scenarios of relevant covariates (population and income growth, construction of new buildings, new hospitals, access to green areas, new hazards and climate change patterns, etc.);
2. Accounting for development limitations, e.g., “Red lines”, Technological and ATAHO indexes (in Istanbul, Turkey), see Figure 2 and 3;
3. Feasible mitigation precautionary and reconstruction “Building-Back-Better (BBB)” measures;
4. Targeted “point” mitigation (buildings reinforcement, location of shelters) measures green areas, etc., which are relevant for case studies in Bucharest (Romania) and Istanbul (Turkey) (on optimal shelters allocation see e.g. Qiu et al., 2024; Xu et al., 2016);
5. Other scenarios

In Istanbul (Turkey), the limiting city “expansion” factors are: “Red) lines” (Figure 2), ATAHO risk index (Figure 3), Technological hazard (NaTech) index (Figure 4). Locations (neighborhoods) with high ATAHO and NaTech risk indexes should be considered to allocate less population and properties in the future. Figure 4 visualizes the overlapping areas with high population density (in blue), locations with heavy earthquake damages (ATAHO) (in orange), and areas of possible technological risks triggered by earthquake (i.e., Natech risk index) (in red). Figure 5 presents the results of the studies on multi-hazard risk scores (https://doi.org/10.1038/s44304-025-00065-8) in Istanbul. Both Figures 4 and 5 emphasize the need for targeted vulnerability reduction measures/interventions in high-risk areas, which are marked by high negative criteria scores, indicating their low resilience.

The SML vulnerability models can be effectively linked into the framework of Integrated and Spatially-detailed Catastrophe Risk Modeling and Management (ISCRiMM) Decision Support Systems for designing optimal and robust interdependent ex-ate and ex-post measures decreasing individual and systemic vulnerabilities and increasing resilience and BBB capacities (Ermolieva et al., 2023; Ermolieva et al., 2025 forthcoming; Komendantova et al., 2025).

3.1. Vulnerability classes and targeted vulnerability analysis and mitigation.

From the Quantile-based (QB) SML modeling, it is possible to “classify” locations according to their current vulnerability levels (quantiles), both individual and systemic vulnerabilities, i.e., identify combinations of vulnerability levels/classes. Locations can be subdivided into vulnerability classes characterized by various combinations of high, moderate, low socio-economic, infrastructural, structural, and systemic vulnerabilities. It is possible to distinguish, e.g., the following representative combinations of vulnerability classes, which correspond to different population groups :

• locations with low socio-economic vulnerability (or socio-economic status, SES in Istanbul (Turkey)), high population density, high buildings’ and infrastructures’ vulnerability, etc.;
• locations with high socio-economic vulnerability (often lower education and income levels, near-marginalized socioeconomic groups), high buildings’ and infrastructures’ vulnerability, low critical urban services and infrastructure availability;
• locations with high socio-economic vulnerability (often lower education and income levels), moderate buildings’ and infrastructures’ vulnerability, moderate critical urban services and infrastructure availability; and so on.

3.2. Quantile-based analysis of structural vulnerability: damages vs building types. Locations with high, moderate, low structural vulnerability depending on the buildings’ compositions can be identified through quantile-based analysis. Assume that the dependent variable is “damage” and the covariates are buildings’ types (wooden, reinforced concrete, stone, prefabricated, steel, masonry, others). By varying the level of quantile (from 0 to 1) it is possible to derive the probability distribution of “damage” variable, i.e., the value/level and the respective probability. The coeficients in front of the covariates or, in other words, the “weights” of the covariates, indicate the contribution of the covariate to each “damage” quantile (vulnerability level).

Figure 6. Quantile regression processes: damage vs seismicity and buildings’ types. The x-axis defines the quantile (on the scale from 0 to 1) and the y-axis defines the coefficient (weight) of the covariate to the respective “damage” quantile.

3.2. Future vulnerability analysis: Vulnerability Analysis and Assessment (VAA) tool: The future vulnerability scenarios can integrate and investigate targeted measures to decrease specific vulnerabilities and vulnerability classes in respective locations. The VAA tool is being developed at IIASA (Cooperation and Transformative Governance group of Advanced Systems Analysis Program (CAT-ASA, IIASA) for PARATUS project to visualize the current vulnerabilities and the future vulnerabilities scenarios based on the data and the scenarios of respective vulnerability drivers. The SML models for analyzing and projecting the vulnerabilities by classes are linked with the VAA tool.

Figure 7 displays the front page of the VAA tool providing an introduction into the PARATUS project and the earthquake-related case studies, including the methodology for current vulnerability analysis and future vulnerability scenarios development. Figure 8 visualizes, e.g., the 0.5th quantile of the “Human vulnerability index” for Bucharest (Romania) analyzed jointly with Bucharest PARATUS team using data and studies in Armas et al.2017 and Toma-Danila et al. 2015, 2017. For the case study in Istanbul, the VAA tool, e.g., allows to pick up the variable (vulnerability driver), allocation scenario (risk averse, RA), and compare it to the BAU (as it is discussed in section 2). Alternative approaches to the development of RA scenarios for different vulnerability drivers can be defined through “Robust downscaling and allocation” approaches.

For example, Figure 10 visualises the ratio of two spatially detailed population projection scenarios (BAU and RA) (www.citypopulation.de/en/turkey/istanbul/admin/). In Scenario 1 (BAU), the population (aggregate projections come from “Open Data Portal of the Istanbul Metropolitan Municipality”, scenario 5) in locations increases proportionally to historical growth rates. The scenario is defined as BAU because it does not account for the risks (Figures 2-5) in locations. In Scenario 2 (RA), the increase of population is allocated accounting for thresholds and “riskiness of locations/neighbourhoods”. In Figure 10, more reddish colours indicate neighbourhoods where population values are higher in Scenario 2 (RA), and with lighter colours (yellowish and greenish) – where population values are higher in Scenario 1 (BAU). Thus, the relocation scenario 2 (RA) enables to relax the Spatial Vulnerability Burden (SVB) in many central Istanbul neighborhoods (colored with light green and yellow in Figure 10) characterized by higher risk indices of all kinds (Figures 2-5). The SVB refers to possible “disproportionate” impact of natural hazards or environmental risks on specific geographic locations or populations. The SVB concept highlights how certain areas or groups are more susceptible to negative consequences due to their location, socioeconomic status, or other factors.