A probabilistic GIS-based framework for urban flood risk assessment in Chengdu metro network Authors: Qiuping Li, Ting Ni, Lei Li, Hao Wen, Jiuping Xu Published: 2025 ABSTRACT Urban metro network is highly vulnerable to flooding, yet most existing risk assessments rely on deterministic weighting methods that inadequately address uncertainty. To address this limitation, this study develops a probabilistic GIS-based flood risk assessment framework that integrates the Analytic Hierarchy Process (AHP) with its Monte Carlo extension (MC-AHP). Unlike conventional studies, the framework (i) explicitly quantifies uncertainty in expert judgments through β-PERT simulations, (ii) incorporates a comprehensive indicator system covering hazard, exposure, and vulnerability—including passenger flow and under-construction lines often overlooked in prior work, and (iii) provides spatially explicit risk mapping for both operational and planned infrastructure. Application to the Chengdu metro network reveals a concentric flood risk distribution, with the central urban core exhibiting the highest vulnerability despite moderate hazard levels. Compared to AHP, MC-AHP identifies a greater proportion of high-risk areas and provides probability distributions of indicator weights, thereby reducing the influence of subjective bias. Validation against historical flood events confirms the framework’s predictive reliability. By explicitly incorporating uncertainty and extending analysis to planned infrastructure, this study advances the methodological rigor of urban flood risk assessment and provides generalizable insights for strengthening metro network resilience worldwide in the face of accelerating climate change and rapid urbanization. Introduction In recent decades, the frequency and intensity of urban flooding have increased significantly due to rapid urbanization and global climate change1, making floods one of the most economically destructive natural hazards worldwide2. Densely populated cities, especially those with aging infrastructure and inadequate drainage systems, are particularly vulnerable3. To address growing traffic demand and relieve surface congestion, metro network has become essential components of sustainable urban transportation4,5. These systems, however, are particularly susceptible to flood-related disruptions because of their underground nature6,7. Flooding of metro stations or tunnels can paralyze urban transit, endanger human lives, and cause substantial economic losses8. A notable example is the 2021 Zhengzhou metro flood disaster, where excessive rainfall inundated tunnels, causing 14 deaths and trapping hundreds of passengers9. This event underscores the vulnerability of rapidly expanding underground rail networks to extreme rainfall. Over the past decade, China has undergone one of the fastest metros build-outs worldwide5. By the end of 2024, nearly 50 cities operated metro network totalling more than 10,000 km. This rapid national expansion has greatly increased the spatial footprint and societal dependence on underground rail infrastructure, thereby raising both the likelihood that metro assets intersect flood-prone areas and the potential human/economic consequences of station or tunnel inundation. Consequently, flood risk assessment of metro network is essential for prioritizing retrofitting and early-warning measures. However, resilience investments have not kept pace with system expansion, particularly in regions prone to extreme precipitation10,11. Against this backdrop, Chengdu, a sprawling megacity in the Sichuan Basin with intricate watercourses, faces escalating flood risk as heavy rainfall intensifies alongside continued urban growth. To strengthen urban resilience and safeguard metro network, comprehensive flood risk evaluation methods are crucial12. These should account for hazard intensity, exposure levels, and system vulnerability while providing detailed spatial data to inform planning and crisis management13–16. Existing flood risk assessment methods fall into four broad categories: historical disaster statistics (HDMS), scenario-based simulations (SSA), remote sensing and GIS-based assessments, and multi-criteria decision analysis (MCDA)11,17–19. Integrating MCDA into GIS has advanced spatial decision support by combining geospatial analysis with both empirical data (e.g., rainfall, land use, slope) and expert judgment, enabling more comprehensive and value-informed assessments12,20,21. Among MCDA techniques, the Analytic Hierarchy Process (AHP) is widely applied due to its structured approach to factor weighting and its capacity to incorporate expert judgment22,23. However, AHP is constrained by its deterministic framework and susceptibility to subjective bias, reducing its reliability in decision-making under uncertainty and imperfect information24,25. To address this issue, researchers have introduced probabilistic extensions to AHP. One such approach is the Monte Carlo Analytic Hierarchy Process (MC-AHP), which uses probabilistic distributions, typically β-PERT to simulate variability in expert judgments and generate more robust weight estimates26,27. Compared to traditional AHP, MC-AHP provides a deeper understanding of uncertainty and reduces the influence of extreme or inconsistent evaluations28. Despite the growing attention to flood risks in urban infrastructure, existing research still presents several gaps. First, few studies have systematically applied advanced multi-criteria methods such as the Monte Carlo Analytic Hierarchy Process (MC-AHP) to metro network risk assessment. Second, many assessments lack spatial precision or fail to incorporate comprehensive vulnerability indicators such as passenger flow, population density, and road proximity. Moreover, existing studies often overlook the future risks faced by infrastructure under construction. To address these limitations, this study develops this study develops a probabilistic GIS-based flood risk assessment framework that integrates AHP with its Monte Carlo extension (MC-AHP). Specifically, a multi-level indicator system is established based on the Intergovernmental Panel on Climate Change29 risk framework, incorporating 16 indicators across the three dimensions of hazard, exposure, and vulnerability. The framework integrates traditional AHP with MC-AHP to evaluate and compare the spatial distribution of flood risks across metro lines and stations. This study advances existing research in three key aspects, enhancing both the methodological value and practical relevance of metro flood risk assessment. First, it introduces MC-AHP to explicitly quantify uncertainty in expert judgments, a critical limitation often overlooked in conventional AHP-based studies. Second, by integrating the triad of disaster formation mechanics—hazard-causing factors, hazard-pregnant environment, and hazard-bearing carriers—the framework enables a more comprehensive evaluation. Third, by simultaneously assessing both operational and under-construction metro lines and stations, this study extends the scope of prior research and provides forward-looking insights for infrastructure planning and resilience building. The objectives of this study are fourfold: (1) to construct a spatially explicit and probabilistic flood risk assessment model for metro network using GIS, AHP, and MC-AHP; (2) to identify high-risk metro lines and stations in Chengdu and examine their spatial distribution; (3) to compare the performance of deterministic AHP and probabilistic MC-AHP in capturing uncertainty in urban flood risk evaluation; and (4) to provide practical recommendations for infrastructure planning, early-warning systems, and resilience enhancement. By combining methodological innovation with empirical validation in Chengdu, this study not only addresses the limitations of conventional flood risk assessments but also offers generalizable insights for strengthening metro network resilience in rapidly urbanizing and climate-stressed cities worldwide. Methodology Guided by the research objectives, we integrated AHP, MC-AHP, and GIS to evaluate flood risk in Chengdu’s metro network. Figure 1 illustrates the assessment flowchart. The evaluation covers both operational and under-construction lines and comprises three components: (i) constructing a hierarchical indicator system; (ii) deriving indicator weights via AHP and MC-AHP; and (iii) performing GIS-based spatial analysis and visualization. Fig. 1Flow chart of flood risk assessment of metro network. Flow chart of flood risk assessment of metro network. Assessment structural system System construction The assessment structure consists of the goal layer, criteria layer, and indicator layer. Following the IPCC29 risk framework and disaster system theory, flood risk is defined as a function of hazard (H), exposure (E), and vulnerability (V), as shown in Eq. (1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:R=f(H,E,V)=H+E+V$$\end{document} where ‘+’ represents overlay analysis in GIS, R denotes the metro flood disaster risk. Considering that different factors contribute differently to flood risk, the evaluation of flood risk should employ different evaluation factors and their corresponding weights. Therefore, metro flood risk can be redefined using Eq. (2).2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:R=\alpha\:\sum\:_{i=1}^{n}{H}_{i}{h}_{i}+\beta\:\sum\:_{j=1}^{n}{E}_{j}{e}_{j}+\gamma\:\sum\:_{k=1}^{n}{V}_{k}{v}_{k}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha\:$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\beta\:$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\gamma\:$$\end{document} are the weights for hazard, exposure, and vulnerability, respectively. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{H}_{i}$$\end{document}、\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{E}_{j}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{V}_{k}$$\end{document} represent the values of the hazard, exposure, and vulnerability indices, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{i}$$\end{document}、\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}_{j}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{v}_{k}$$\end{document} denote the weights of each respective indicator within the three factors of hazard, exposure, and vulnerability. Index system According to the formation mechanism of flood disasters, a regional flood disaster system consists of three components: disaster-causing factors, disaster-pregnant environment, and disaster-bearing carriers30,31. Based on this framework, a metro flood risk assessment index system is developed by incorporating the dimensions of hazard, exposure, and vulnerability. The index system includes: (1) Hazard index. Hazard refers to the potential occurrence of a natural or human-induced physical event, trend, or impact that may cause loss of life, damage to property, ecosystems, or disruption of services29,32. In this study, hazard denotes the probability, frequency and intensity of flood-producing processes (e.g., heavy rainfall, river overflow) that can cause inundation. Hazard of disaster-causing factors index include annual average rainfall (H11), annual number of heavy rain days (H12), and maximum daily rainfall (H13). And Hazard of disaster-pregnant environment index include elevation (H21), slope (H22), river density (H23), and river proximity (H24). River density (H23) refers to the length of rivers per unit area. River proximity (H24) refers to the spatial relationship between rivers and surrounding environments. (2) Exposure index. Exposure refers to the presence and distribution of people, assets, and ecosystems in areas that may be adversely affected by natural hazards, reflecting the extent to which these receptors are subject to the frequency, intensity, and spatiotemporal characteristics of hazardous events29,33. In this study, the exposure of disaster-bearing carriers index includes the number of metro exits (E1), metro density (E2), metro proximity (E3), and the number of transfer stations (E4). (3) Vulnerability index. Vulnerability refers to the inherent characteristics of receptors that determine their susceptibility to harm and their capacity to cope with, resist, and recover from the impacts of natural hazards29,34. The vulnerability of disaster-bearing carriers index includes five indicators: passenger flow intensity (V1), station intensity (V2), population density (V3), road network density (V4), and road network proximity (V5). Weight calculation Analytic hierarchy process (AHP) AHP uses pairwise comparisons to evaluate the relative importance of indicators within each layer23,35. This method helps break down complex problems into a hierarchical structure for multi-level comparison and weight determination, followed by qualitative and quantitative assessments, ultimately leading to decision results. The main steps are:Constructing the Judgment Matrix: According to basic principles of AHP22, the judgment matrix is constructed using pairwise comparisons, as shown in Eq. (3).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{u}={\left({a}_{ij}\right)}_{m\times\:n}=\left(\begin{array}{ccc}{a}_{11}&\:\cdots\:&\:{a}_{1n}\\\: \ldots &\:\ddots\:&\: \ldots \\\:{a}_{n1}&\:\cdots\:&\:{a}_{nn}\end{array}\right)$$\end{document}(2)Calculation of weights by normalizing the matrix and solving for the principal eigenvector.(3)Consistency check via the consistency ratio (CR). A matrix is valid if CR < 0.1. Constructing the Judgment Matrix: According to basic principles of AHP22, the judgment matrix is constructed using pairwise comparisons, as shown in Eq. (3).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{u}={\left({a}_{ij}\right)}_{m\times\:n}=\left(\begin{array}{ccc}{a}_{11}&\:\cdots\:&\:{a}_{1n}\\\: \ldots &\:\ddots\:&\: \ldots \\\:{a}_{n1}&\:\cdots\:&\:{a}_{nn}\end{array}\right)$$\end{document} Constructing the Judgment Matrix: According to basic principles of AHP22, the judgment matrix is constructed using pairwise comparisons, as shown in Eq. (3).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{A}_{u}={\left({a}_{ij}\right)}_{m\times\:n}=\left(\begin{array}{ccc}{a}_{11}&\:\cdots\:&\:{a}_{1n}\\\: \ldots &\:\ddots\:&\: \ldots \\\:{a}_{n1}&\:\cdots\:&\:{a}_{nn}\end{array}\right)$$\end{document} (2)Calculation of weights by normalizing the matrix and solving for the principal eigenvector. Calculation of weights by normalizing the matrix and solving for the principal eigenvector. (3)Consistency check via the consistency ratio (CR). A matrix is valid if CR < 0.1. Consistency check via the consistency ratio (CR). A matrix is valid if CR < 0.1. Monte Carlo analytic hierarchy process (MC-AHP) Compared to the AHP method, the application of MC-AHP demonstrates two major advantages. First, MC-AHP integrates the uncertainty inherent in expert judgments during the decision-making process36. To incorporate uncertainty in expert judgment, MC-AHP treats pairwise comparisons as β-PERT distributions, using minimum, most likely, and maximum values. Second, MC-AHP provides more detailed output, helping decision-makers better understand the uncertainty behind rankings28. The main steps are as follows:Generate corresponding β distributions for the off-diagonal elements of each judgment matrix.(2)Randomly generate numbers following the β-PERT distribution for the off-diagonal elements of each judgment matrix, and generate the reciprocals of the corresponding random numbers. The definition is given in Eq. (4).4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\rm A}}_{ij} = min + random\:beta(\alpha, \beta) \cdot\: (max-min)$$\end{document}(3)Calculate the maximum eigenvalue and its corresponding eigenvector for each randomly generated judgment matrix. The resulting eigenvector is further normalized to determine the relative weights of each indicator.(4)Measure the consistency of the judgment matrix by calculating the consistency index (CI), similar to the traditional AHP method.(5)Repeat steps 4 to 6 using Monte Carlo simulations until 1000 judgment matrices are obtained for each indicator group. If a judgment matrix does not meet the consistency requirement (CR > 0.1), it is excluded from the study until the consistency ratio (CR) for all obtained judgment matrices is less than 0.1. Finally, the relative weights of each factor are determined using the 1000 judgment matrices and presented as probability density functions rather than deterministic values. Generate corresponding β distributions for the off-diagonal elements of each judgment matrix. Generate corresponding β distributions for the off-diagonal elements of each judgment matrix. (2)Randomly generate numbers following the β-PERT distribution for the off-diagonal elements of each judgment matrix, and generate the reciprocals of the corresponding random numbers. The definition is given in Eq. (4).4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\rm A}}_{ij} = min + random\:beta(\alpha, \beta) \cdot\: (max-min)$$\end{document} Randomly generate numbers following the β-PERT distribution for the off-diagonal elements of each judgment matrix, and generate the reciprocals of the corresponding random numbers. The definition is given in Eq. (4).4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\rm A}}_{ij} = min + random\:beta(\alpha, \beta) \cdot\: (max-min)$$\end{document} (3)Calculate the maximum eigenvalue and its corresponding eigenvector for each randomly generated judgment matrix. The resulting eigenvector is further normalized to determine the relative weights of each indicator. Calculate the maximum eigenvalue and its corresponding eigenvector for each randomly generated judgment matrix. The resulting eigenvector is further normalized to determine the relative weights of each indicator. (4)Measure the consistency of the judgment matrix by calculating the consistency index (CI), similar to the traditional AHP method. Measure the consistency of the judgment matrix by calculating the consistency index (CI), similar to the traditional AHP method. (5)Repeat steps 4 to 6 using Monte Carlo simulations until 1000 judgment matrices are obtained for each indicator group. If a judgment matrix does not meet the consistency requirement (CR > 0.1), it is excluded from the study until the consistency ratio (CR) for all obtained judgment matrices is less than 0.1. Finally, the relative weights of each factor are determined using the 1000 judgment matrices and presented as probability density functions rather than deterministic values. Repeat steps 4 to 6 using Monte Carlo simulations until 1000 judgment matrices are obtained for each indicator group. If a judgment matrix does not meet the consistency requirement (CR > 0.1), it is excluded from the study until the consistency ratio (CR) for all obtained judgment matrices is less than 0.1. Finally, the relative weights of each factor are determined using the 1000 judgment matrices and presented as probability density functions rather than deterministic values. GIS-based spatial analysis Index normalization To ensure the accuracy of the evaluation results, all evaluation factors must be normalized before conducting overlay analysis. Based on the spatial characteristics of the factors, the maximum-minimum normalization technique is employed on the GIS platform37. Elevation and slope are considered negative indicators and are normalized using Eq. (5). Other factors, whose increase is proportional to flood risk, are considered positive indicators and normalized using Eq. (6).5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Y}_{ij}=\frac{{{X}_{max}-X}_{ij}}{{X}_{max}-{X}_{min}}$$\end{document}6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{ij}=\frac{{X}_{ij}-{X}_{min}}{{X}_{max}-{X}_{min}}$$\end{document} Where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{Y}_{ij}\:$$\end{document}represents the normalized value of the indicator, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{X}_{max}\:$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{X}_{min}$$\end{document} are the maximum and minimum values of the indicator, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{X}_{ij}$$\end{document} is the actual value of the evaluation indicator. Weighted linear combination (WLC) Weighted Linear Combination (WLC) linearly combines multiple geographical raster layers according to predefined weights to obtain an overall evaluation result, as defined in Eq. (7).7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:R=\sum\:{W}_{i}\times\:{X}_{i}$$\end{document} Where R is the flood disaster risk value, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{W}_{i}$$\end{document} is the weight of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:i$$\end{document} criterion layer factor, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{X}_{i}$$\end{document} is the normalized value of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:i$$\end{document} criterion layer factor. Risk level visualization In this study, the evaluation factors of rainfall events are visualized using GIS tools. The Kriging interpolation method is widely used to reproduce rainfall events, as it accurately represents the spatial distribution patterns of rainfall events38.8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:Z\left(s\right)=\mu\:\left(s\right)+\epsilon\:\left(s\right)$$\end{document} The principle of Kriging interpolation is expressed in Eq. (8), where s represents the spatial coordinates, Z(s) is the variable in the coordinate s, and the Z(s) is composed of a, trend µ(s) and random deviations ε(s). Case study: chengdu Overview of the study area Chengdu is located in the western part of the Sichuan Basin, situated between longitudes 102°54′ to 104°53′ and latitudes 30°05′ to 31°26′39, the geographical location is shown in Fig. 2. The topography is higher in the northwest and lower in the south. The city mainly consists of plains, located in the middle reaches of the Min River, where rivers are crisscrossed, forming a multi-river plain characteristic. As of 2023, Chengdu governs 12 urban districts, 3 counties, and 5 county-level cities, covering a total area of 14,335 square kilometers, with a permanent population of 21.403 million. Since the opening of the Chengdu Metro in 2010, it has become an integral part of the city’s transportation system. By December 2024, the Chengdu Metro operated 14 lines, with a total operating length of 594.14 km (excluding trams and suburban railways). There are 5 metro lines under construction, with a total length of 269.9 km. In 2024, the average daily ridership of the Chengdu Metro reached 6.0025 million trips. These figures reflecting a large spatial footprint and substantial human exposure. Figure 3 shows the metro network of the study area. Fig. 2The geographical location of the study area (Created by the authors using QGIS 3.40, https://qgis.org/). The geographical location of the study area (Created by the authors using QGIS 3.40, https://qgis.org/). Fig. 3Chengdu metro line map (The letters are the abbreviations of districts, counties and county-level cities in Chengdu: JJ: Jinjiang; QY: Qingyang; JN: Jinniu; WH: Wuhou; CH: Chenghua; LQY: Longquanyi; QBJ: Qingbaijiang; XD: XiNdu; WJ: Wenjiang; SL: Shuangliu; PD: Pidu; JT: Jintang; DY: Dayi; PJ: Pujiang; XJ: Xinjin; DJY: Dujiangyan; PZ: Pengzhou; QL: Qionglai; CZ: Chongzhou; JY: Jianyang) (Created by the authors using QGIS 3.40, https://qgis.org/). Chengdu metro line map (The letters are the abbreviations of districts, counties and county-level cities in Chengdu: JJ: Jinjiang; QY: Qingyang; JN: Jinniu; WH: Wuhou; CH: Chenghua; LQY: Longquanyi; QBJ: Qingbaijiang; XD: XiNdu; WJ: Wenjiang; SL: Shuangliu; PD: Pidu; JT: Jintang; DY: Dayi; PJ: Pujiang; XJ: Xinjin; DJY: Dujiangyan; PZ: Pengzhou; QL: Qionglai; CZ: Chongzhou; JY: Jianyang) (Created by the authors using QGIS 3.40, https://qgis.org/). Statistical data indicates that Chengdu experienced 30 major flood events between 2000 and 202031. During the flood season of 2020 (May to September), the city’s rainfall reached 1,110.9 mm, far exceeding the long-term average of 760.0 mm. In August alone, rainfall accounted for over 60% of the total flood season precipitation. Four heavy rainfall events occurred during this period (on August 7, 11, 16, and 30), characterized by frequent heavy rain, long duration, high intensity, wide coverage, and high hourly rainfall. These events also caused high water levels and severe disaster losses. For example, the ‘8.16’ rainfall event caused extensive flooding at the Shufang Station of Metro Line 2, trapping many passengers. And the datasets used to quantify hazard, exposure, and vulnerability are summarized in Table 1. Table 1Research data sources.Data typeData nameMain usageData sourcesRainfall characteristic dataannual average rainfallhazard assessment of disaster-causing factorsHuiju Environmental Platformannual rainstorm daysmaximum daily precipitationNatural geographic dataelevationhazard assessment of disaster-pregnant environmentThe geospatial data cloud platformslopeDEM generationriver systemOpenStreetMapriver densityGIS generationriver proximityGIS generationadministrative divisiondisaster risk mappingNational Geomatics Center of Chinasubway vector datamapSocio-economic datapopulation densityvulnerability assessment of disaster-bearing carriersChengdu Statistical Yearbookurban main roadsOpenStreetMaproad network densityGIS generationroad network ProximityGIS generationSubway attribute datanumber of subway exportsexposure assessment of disaster-bearing carriersmapnumber of transfer stationsChengdu Metro officialsubway densityGIS generationsubway proximityGIS generationpassenger flow intensityvulnerability assessment of disaster-bearing carriersGenerated based on official datastation intensity Table 1Research data sources.Data typeData nameMain usageData sourcesRainfall characteristic dataannual average rainfallhazard assessment of disaster-causing factorsHuiju Environmental Platformannual rainstorm daysmaximum daily precipitationNatural geographic dataelevationhazard assessment of disaster-pregnant environmentThe geospatial data cloud platformslopeDEM generationriver systemOpenStreetMapriver densityGIS generationriver proximityGIS generationadministrative divisiondisaster risk mappingNational Geomatics Center of Chinasubway vector datamapSocio-economic datapopulation densityvulnerability assessment of disaster-bearing carriersChengdu Statistical Yearbookurban main roadsOpenStreetMaproad network densityGIS generationroad network ProximityGIS generationSubway attribute datanumber of subway exportsexposure assessment of disaster-bearing carriersmapnumber of transfer stationsChengdu Metro officialsubway densityGIS generationsubway proximityGIS generationpassenger flow intensityvulnerability assessment of disaster-bearing carriersGenerated based on official datastation intensity Research data sources. Risk assessment based on AHP Weight calculation of indicators Based on the constructed indicator system and the results of expert surveys, as well as the weight calculation process introduced in Sect. 2.2.1 using the Analytic Hierarchy Process (AHP), judgment matrices and weight calculations were constructed using Excel. The comprehensive weight calculation results of the evaluation indicator system are shown in Table 2. From the table, it can be seen that the most influential factor affecting the metro flood risk is the vulnerability of the disaster-bearing carriers. This aligns with the findings of Lyu et al.40, who argued that in metro network flood risk assessments, vulnerability is more important than hazard and exposure. Table 2AHP comprehensive weight of metro flood disaster risk.Goal layerCriterion layerWeight of criterion layerSub-criterion layerWeight of sub-criteria layerIndicator layerIndex layer weightFlood risk of metro networkHazard(H)0.3119disaster-causing factors(H1)0.6667H110.1638H120.539H130.2972disaster-pregnant environment(H2)0.3333H210.3111H220.4222H230.1611H240.1056Exposure(E)0.1976disaster-bearing carriersE10.4333E20.3E30.15E40.1167Vulnerability(V)0.4905V10.3865V20.2574V30.1697V40.112V50.0744 Table 2AHP comprehensive weight of metro flood disaster risk.Goal layerCriterion layerWeight of criterion layerSub-criterion layerWeight of sub-criteria layerIndicator layerIndex layer weightFlood risk of metro networkHazard(H)0.3119disaster-causing factors(H1)0.6667H110.1638H120.539H130.2972disaster-pregnant environment(H2)0.3333H210.3111H220.4222H230.1611H240.1056Exposure(E)0.1976disaster-bearing carriersE10.4333E20.3E30.15E40.1167Vulnerability(V)0.4905V10.3865V20.2574V30.1697V40.112V50.0744 AHP comprehensive weight of metro flood disaster risk. Hazard analysis Based on publicly available data, this study collected annual average rainfall data from 2001 to 2023, as well as the number of annual heavy rainfall days and the maximum daily rainfall from 2001 to 2022. These data were visualized using Kriging interpolation in GIS, resulting in spatial distribution maps for these three indicators, as shown in Figs. 4(a-c). Using the normalization method described in Sect. 2.3.2, GIS tools were applied before performing overlay analysis. Based on the hazard index weights for disaster-causing factors, the normalized spatial distribution of each disaster-causing factors was combined with the corresponding weights of the hazard indices using the raster calculator in QGIS, resulting in the risk levels for hazard of disaster-causing factors, as shown in Fig. 4(d). From the figure, it can be observed that high-risk areas are concentrated in the northwest of Chengdu, such as Dujiangyan. Areas such as Pengzhou, Chongzhou, Dayi, Qionglai, and Pujiang are associated with higher levels of disaster risk. Fig. 4Spatial distribution of hazard of disaster-causing in Chengdu. (a) annual average rainfall (H11); (b) Annual rainstorm days (H12); (c) maximum daily precipitation (H13); (d) hazard of disaster-causing factors (H1) (Created by the authors using QGIS 3.40, https://qgis.org/). Spatial distribution of hazard of disaster-causing in Chengdu. (a) annual average rainfall (H11); (b) Annual rainstorm days (H12); (c) maximum daily precipitation (H13); (d) hazard of disaster-causing factors (H1) (Created by the authors using QGIS 3.40, https://qgis.org/). The elevation (H21) and slope (H22) raster layers were generated from the Digital Elevation Model (DEM) using the raster terrain analysis tools in the QGIS Processing Toolbox. River density refers to the length of rivers per unit area, calculated using the line density function with a search radius of 1 km. River network proximity refers to the distance to the nearest river, generated using the buffer tool in QGIS by specifying multiple concentric distances (200 m, 400 m, 600 m, 800 m, and 1000 m)41. Figures 5(a-d) show the spatial distribution of each hazard of disaster-pregnant environment index, including elevation, slope, river density, and river network proximity. Using the same method, the hazard of disaster-pregnant environment level distribution map for the metro network flood disaster is shown in Fig. 5(e). It can be observed that high-risk areas are located in the low-elevation, flat central areas and regions close to river systems. Based on the above analysis, Fig. 4(d). and Fig. 5(e). were subjected to a weighted linear combination in GIS to generate the hazard risk level map, as shown in Fig. 5(f). The map reveals that Dujiangyan exhibits the highest level of flood hazard risk. Areas with relatively high hazard levels include Pengzhou, Chongzhou, Dayi, Qionglai, and Pujiang. The main urban area demonstrates a lower level of hazard risk, while the eastern region of Chengdu shows the lowest level of hazard risk. Fig. 5Spatial distribution of hazard in Chengdu. (a)DEM(H21); (b)slope(H22); (c)river network density(H23); (d)river proximity(H24); (e) hazard of disaster-pregnant environment (H2); (f)Hazard index(H) (Created by the authors using QGIS 3.40, https://qgis.org/). Spatial distribution of hazard in Chengdu. (a)DEM(H21); (b)slope(H22); (c)river network density(H23); (d)river proximity(H24); (e) hazard of disaster-pregnant environment (H2); (f)Hazard index(H) (Created by the authors using QGIS 3.40, https://qgis.org/). Exposure analysis of the disaster-bearing carriers The exposure level of the metro network to flood disasters is assessed using the same method as the hazard analysis. The density of metro lines is defined as the length of metro lines per square kilometer, obtained using a radius of 1,000 m. Metro line proximity is defined as the distance to the nearest metro line, generated using the buffer tool in QGIS by specifying multiple concentric distances (200 m, 400 m, 600 m, 800 m, and 1000 m). Figure 6 shows the spatial distribution of each exposure index. Fig. 6Spatial distribution of exposure of disaster-bearing carriers of Chengdu metro. (a) number of metro exports (E1); (b) metro density (E2); (c)metro proximity (E3); (d) number of transfer stations (E4) (Created by the authors using QGIS 3.40, https://qgis.org/). Spatial distribution of exposure of disaster-bearing carriers of Chengdu metro. (a) number of metro exports (E1); (b) metro density (E2); (c)metro proximity (E3); (d) number of transfer stations (E4) (Created by the authors using QGIS 3.40, https://qgis.org/). Based on the above analysis, the spatial distribution map of the exposure risk level of the disaster-bearing carriers was obtained by weighted combination using the raster calculator in QGIS, incorporating the weights of each indicator and the normalized spatial distribution map, as shown in Fig. 7. It can be observed that the metro lines are more densely distributed in the city center, resulting in higher exposure levels, primarily involving metro lines traversing the main urban area, including Metro Line 1, Line 2, Line 4, Line 7, and Line 8, as well as the under-construction sections of Line 18 Phase 3, Line 13 Phase 1, and Line 17 Phase 2. Fig. 7Spatial distribution of exposure of Chengdu metro (Created by the authors using QGIS 3.40, https://qgis.org/). Spatial distribution of exposure of Chengdu metro (Created by the authors using QGIS 3.40, https://qgis.org/). Vulnerability analysis of the Disaster-bearing carriers Road network density reflects the length of the road network per unit area, while road network proximity refers to the distance to the nearest road. In the analysis process, the same method as for the river network and metro network was used. The road network density was calculated using the line density function, with a search radius of 1,000 m. The buffer tool in QGIS was employed by specifying multiple concentric distances (200 m, 400 m, 600 m, 800 m, and 1000 m) for road network proximity. Figures 8(a-e) show the spatial distribution of each vulnerability index. Based on the above analysis, each indicator was normalized, and the normalized evaluation indicators were combined with their respective weights using the raster calculator in QGIS to derive the spatial distribution of the vulnerability of the disaster-bearing carriers, as shown in Fig. 8(f). It can be observed that the main urban area exhibits the highest level of vulnerability risk, followed by the surrounding regions of Wenjiang, Shuangliu, Longquanyi, Xindu, and Pidu. Additionally, areas along the main road distribution in Chengdu also show relatively high vulnerability. Fig. 8Spatial distribution of vulnerability index of disaster-bearing carriers in Chengdu. (a) passenger flow intensity (V1); (b) station intensity (V2); (c) population density (V3); (d) road network density (V4); (e) road network proximity (V5); (f) vulnerability (V) (Created by the authors using QGIS 3.40, https://qgis.org/). Spatial distribution of vulnerability index of disaster-bearing carriers in Chengdu. (a) passenger flow intensity (V1); (b) station intensity (V2); (c) population density (V3); (d) road network density (V4); (e) road network proximity (V5); (f) vulnerability (V) (Created by the authors using QGIS 3.40, https://qgis.org/). AHP flood risk assessment results Based on the analysis above, the regional flood risk in Chengdu can be obtained by overlay analysis using the Raster Calculator in QGIS. Figure 9(a) shows the spatial distribution of regional flood risk obtained using the AHP method. As shown in the figure, the high-risk areas are mainly located in the central urban area, and the risk follows a ring-like distribution in the central urban area. The distribution of metro lines and the road network has a significant impact on the flood risk in Chengdu, with the metro line distribution having a larger impact. The medium-risk areas are mainly located in the northern and western parts of Chengdu, with lower risks in the east. Fig. 9Regional flood disaster risk and flood risk of metro network based on AHP (Created by the authors using QGIS 3.40, https://qgis.org/). Regional flood disaster risk and flood risk of metro network based on AHP (Created by the authors using QGIS 3.40, https://qgis.org/). After assessing the regional flood risk, the risk level along the metro lines can be obtained by extracting the raster values within a 500 m buffer zone of the metro lines using the ‘Clip Raster by Mask Layer’ tool in QGIS, which is used to evaluate the flood risk for the metro network40. To determine the risk level of the metro network, the flood risk raster was reclassified into five levels using the ‘Reclassify by table’ tool in QGIS: very high, high, medium, low, and very low. Figure 9(b) shows the spatial distribution of flood risk levels for the Chengdu metro network based on the AHP method. As shown in the figure, the central urban area with dense metro lines exhibits a higher flood risk. MC-AHP-based risk assessment Traditional AHP has limitations in handling fuzziness and probabilistic uncertainties. In this study, the evaluation was enhanced by integrating β-PERT distributions, judgment matrices, and Monte Carlo simulations. Multiple case studies have verified the feasibility and effectiveness of this method in handling uncertainty in decision-making assessments42. Indicator weight calculation The calculation of indicator weights is carried out using the method introduced in Sect. 2.2.2. By using 1000 consistent judgment matrices, the relative weights of each factor can be obtained, and their probability density distribution is generated. For example, the probability density distribution of the criterion layer is shown in Fig. 10(a), and the probability density distribution of vulnerability is shown in Fig. 10(b). Table 3 presents the MC-AHP overall score statistics. The results indicate that in the probability density distribution of criterion layer, the vulnerability probability density function carries the largest weight, which suggests that the vulnerability of the disaster-bearing carriers significantly increases metro flood risk, especially in lines with large passenger flows. This is consistent with the conclusions drawn from the AHP method. Based on the MC-AHP average weight values of each indicator, the spatial distribution map of flood disaster risk in the metro network is drawn. Fig. 10Probability density distributions of weights derived from MC-AHP: (a) criterion layer; (b) vulnerability index. Probability density distributions of weights derived from MC-AHP: (a) criterion layer; (b) vulnerability index. Table 3MC-AHP overall score statistics.IndexMean valueMinimum valueMaximum valueCriterion layerhazard(H)0.28340.16240.49exposure(E)0.22740.14820.3986vulnerability(V)0.48920.27980.6212Sub-criterion layerH10.6560.54960.7069H20.3440.29310.4504Hazard of Disaster-causing factor index layerH110.16470.09340.3869H120.53650.2570.6847H130.29880.13510.5467Hazard of Disaster-pregnant environmental index layerH210.26410.14010.4635H220.37680.16080.5793H230.18840.07880.3629H240.17070.06650.3412Exposure of disaster-bearing carriers index layerE10.2880.16330.4504E20.36460.21970.4962E30.20590.10410.392E40.14150.08160.2197Vulnerability of disaster-bearing carriers index layerV10.38660.24180.4866V20.2520.14970.3638V30.17370.09850.2912V40.1130.06730.184V50.07470.04850.1207 Table 3MC-AHP overall score statistics.IndexMean valueMinimum valueMaximum valueCriterion layerhazard(H)0.28340.16240.49exposure(E)0.22740.14820.3986vulnerability(V)0.48920.27980.6212Sub-criterion layerH10.6560.54960.7069H20.3440.29310.4504Hazard of Disaster-causing factor index layerH110.16470.09340.3869H120.53650.2570.6847H130.29880.13510.5467Hazard of Disaster-pregnant environmental index layerH210.26410.14010.4635H220.37680.16080.5793H230.18840.07880.3629H240.17070.06650.3412Exposure of disaster-bearing carriers index layerE10.2880.16330.4504E20.36460.21970.4962E30.20590.10410.392E40.14150.08160.2197Vulnerability of disaster-bearing carriers index layerV10.38660.24180.4866V20.2520.14970.3638V30.17370.09850.2912V40.1130.06730.184V50.07470.04850.1207 MC-AHP overall score statistics. MC-AHP-based metro network flood disaster risk analysis By applying the same method, the spatial distribution map of risk levels for each criterion layer is obtained using GIS overlay analysis, as shown in Fig. 11. Figure 11(a) shows the spatial distribution of integrated hazard risk levels. It can be observed that Dujiangyan has a higher hazard risk level than other areas, with high hazard risks in the northern and western parts, and the lowest risk in the east. Figure 11 (b) shows the spatial distribution of exposure risk levels for the disaster-bearing carriers. It can be seen that high-risk areas are mainly located within the central urban area, mainly due to the dense distribution of the metro network in this area. Figure 11 (c) shows the spatial distribution of vulnerability risk levels for the disaster-bearing carriers. From the figure, it can be seen that the vulnerability risk in the five districts of the central urban area, Shuangliu, and Xindu is higher than in other areas, especially in Qingyang, Wuhou, and Jinjiang, where the vulnerability risk levels are the highest. Outside these areas, high vulnerability follows a belt-like distribution influenced by the road network. Medium-risk areas are distributed around the very high and high-risk areas, covering the largest area. High vulnerability risk areas are mainly located in densely populated areas and regions with a dense metro network. Fig. 11Risk level of each factor in MC-AHP (a) Hazard index (H); (b) Exposure of disaster-bearing carriers(E); (c) Vulnerability of disaster-bearing carriers (V) (Created by the authors using QGIS 3.40, https://qgis.org/). Risk level of each factor in MC-AHP (a) Hazard index (H); (b) Exposure of disaster-bearing carriers(E); (c) Vulnerability of disaster-bearing carriers (V) (Created by the authors using QGIS 3.40, https://qgis.org/). MC-AHP risk assessment results Using the same method, the spatial distribution of regional flood risk obtained through MC-AHP is shown in Fig. 12(a); the spatial distribution of flood risk levels for the Chengdu metro network based on MC-AHP is shown in Fig. 12(b). As illustrated, the highest flood risk is in the central urban area, significantly higher than in other areas, followed by Dujiangyan. The risk in the eastern part of Chengdu is lower. In the flood risk distribution map for the metro network, high-risk areas are mainly concentrated in the central urban area, with the risk level decreasing outward from the center. Fig. 12Regional flood disaster risk and flood risk of metro network based on MC-AHP (Created by the authors using QGIS 3.40, https://qgis.org/). Regional flood disaster risk and flood risk of metro network based on MC-AHP (Created by the authors using QGIS 3.40, https://qgis.org/). Similar to the results obtained by AHP, the flood risk for the metro network based on MC-AHP follows a ring-like distribution, with the areas of very high risk to very low risk being 42.78 km² (8%), 67.57 km² (13%), 115.13 km² (21%), 133.25 km² (25%), and 177.42 km² (33%). The overlapping areas between metro lines are recalculated due to the different metro lines crossing. Fig. 13Distribution of different risk levels of different metro lines. Distribution of different risk levels of different metro lines. Figure 13 shows the area distribution of different flood risk levels for various metro lines in Chengdu. It can be observed that the largest and most widespread risk levels are medium and high-risk levels. Five metro lines exhibit particularly elevated risk profiles, with high and very high-risk zones comprising > 50% of their total coverage area: Line 1 (Wugensong direction, 54.24%), Line 2 (52.39%), Line 9 (52.03%), Line 10 Phase III (62.25%), and Line 18 North Extension Phase III (94.38%). These lines are mainly located in or pass through the central urban area. Fig. 14Number distribution of metro stations with different risk levels. Number distribution of metro stations with different risk levels. For metro station flood risk, the QGIS platform identified the flood risk levels of 484 metro stations (including some under-construction stations). Figure 14 shows the number distribution of metro stations with different flood risk levels for each metro line in Chengdu. After removing 137 duplicate transfer stations, the actual number of stations in the Chengdu metro network is 354 (including 7 under-construction stations). Very high-risk stations are 54, and high-risk stations are the most widely distributed, with 124 stations, accounting for 15.25% and 35.03% of the total, respectively. Stations with medium or higher risk (including medium risk) account for 72.60%. This proportion indicates that the Chengdu metro network has a high potential for flood risk. As seen in Fig. 14, the very high-risk stations are mainly located on Line 7 (19 stations), Line 4 (11 stations), Line 3 (7 stations), Line 2 (11 stations), and Line 1 (11 stations), with these stations mostly located in the central urban area, mainly transfer stations. Identifying the risk levels of metro lines and stations helps make timely emergency responses and implement flood prevention measures before heavy rains and floods, reducing casualties and losses. Validation of the assessment results To verify the practical effectiveness and accuracy of the flood risk assessment model, a spatial overlay analysis was conducted using historical flood event records. Information on ten metro stations that experienced waterlogging during past extreme rainfall events (e.g., the rainstorms on July 2, 2018; August 16, 2020; and July 3, 2025) was collected from official reports and reliable media sources. These historically affected stations served as reference points for validation. Using the QGIS platform, the spatial locations of these ten stations were overlaid onto the flood risk classification map of the metro network (Fig. 12(b)), which was generated based on the Monte Carlo Analytic Hierarchy Process (MC-AHP). The corresponding risk level for each historical flood point was then extracted and statistically analyzed. The results show that most of the historically affected metro stations fall within the medium- to very high-risk zones identified in this study, with five stations located in the very high-risk zone and four stations located in the high-risk zone. For example, Shufang Station, which experienced waterlogging during the August 16, 2020 event, lies in a high-risk area on the risk map. Similarly, Fuhuayuan Station, which accumulated water during the July 3, 2025 rainstorm, is situated in a very high-risk area. Overall, the historical flood points align closely with the medium- and higher-risk areas delineated in this assessment, indicating strong consistency between observed disaster impacts and the model’s predictions. This confirms that the proposed assessment framework can effectively and accurately identify sections of the metro network that are genuinely susceptible to flooding. Comparison of AHP and MC-AHP results As shown in the Figs. 9 and 12, the spatial distribution of flood risk in the region calculated using AHP is similar to that obtained through MC-AHP. The areas with very high flood risk are primarily concentrated in the central urban area, followed by Longquanyi, Xindu, and Wenjiang, with the southeastern region exhibiting the lowest risk. The flood risk results for the metro network obtained by both methods are also quite similar, with the risk following a circular distribution pattern. Figure 15(a) presents the percentage of different flood risk levels for the metro network as derived from both AHP and MC-AHP. In the metro network, the percentage of very high-risk areas according to AHP is 7%, while for MC-AHP, it is 8%. The high-risk areas in AHP and MC-AHP account for 8% and 13%, respectively, and the percentage of areas with medium or higher risk (including medium) is 32% and 42%, respectively. The proportion of very high, high, and medium or higher risk levels in the MC-AHP method is higher than that in AHP, indicating that MC-AHP can identify more high-risk areas in the metro network. Fig. 15Comparison of AHP results with MC-AHP results. Comparison of AHP results with MC-AHP results. A comparative analysis was performed by calculating the overall scores for each evaluation indicator using both AHP and MC-AHP. Statistical data, including the mean, minimum, maximum, standard deviation (SD), 95% confidence interval (CI 95%), and the variation of the standard deviation (SD%) were calculated. Table 4 shows the statistical scores of SD and CI (95%) for AHP and MC-AHP. The analysis reveals that the β-PERT distribution adopted in MC-AHP significantly reduces the standard deviation by 42%−76% compared to the traditional AHP, while weakening the influence of extreme evaluations on the results. The results of the β-PERT distribution resemble a normal distribution, tending to concentrate around the mean, and its confidence intervals are relatively smaller (e.g., the comparison of confidence intervals for the criterion layer is shown in Fig. 15(b)), indicating that the β-PERT distribution performs excellently in handling the uncertainty in group decision-making processes. This result is consistent with the study by Yaraghi et al.28. Table 4Statistics on AHP and MC-AHP indicator scores.AHPMC-AHPfactorsSDCI (95%)SDCI (95%)SD%Criterion layerH0.08580.23640.38670.03050.27970.287264%E0.05950.14540.24980.02660.22510.229655%V0.0820.41860.56240.03030.48550.492963%Sub-criterion layerH10.07880.59760.73580.02830.02740.03164%H20.05230.28740.37910.03040.34210.345942%Hazard of Disaster-causing factorH110.05920.11190.21570.02020.16220.167266%H120.05480.49090.5870.02130.53210.540961%H130.04690.25620.33830.02510.29470.302846%Hazard of Disaster-pregnant environmentalH210.06630.25290.36920.0230.26080.267465%H220.06070.3690.47540.02040.37250.381266%H230.05520.11270.20950.02270.18510.191659%H240.05120.06070.15050.02510.16790.173551%Exposure of disaster-bearing carriersE10.06660.37480.49170.0290.28430.291656%E20.06470.24330.35670.02820.36160.367656%E30.04450.11090.1890.02430.20250.209245%E40.06090.06330.170.02510.140.143159%Vulnerability of disaster-bearing carriersV10.0890.30770.46360.02810.38420.388968%V20.0530.2110.30390.02340.24990.254153%V30.04830.12740.21210.01910.17190.175560%V40.05830.06090.16310.01840.11190.114168%V50.04420.03560.11320.01070.07410.075476% Table 4Statistics on AHP and MC-AHP indicator scores.AHPMC-AHPfactorsSDCI (95%)SDCI (95%)SD%Criterion layerH0.08580.23640.38670.03050.27970.287264%E0.05950.14540.24980.02660.22510.229655%V0.0820.41860.56240.03030.48550.492963%Sub-criterion layerH10.07880.59760.73580.02830.02740.03164%H20.05230.28740.37910.03040.34210.345942%Hazard of Disaster-causing factorH110.05920.11190.21570.02020.16220.167266%H120.05480.49090.5870.02130.53210.540961%H130.04690.25620.33830.02510.29470.302846%Hazard of Disaster-pregnant environmentalH210.06630.25290.36920.0230.26080.267465%H220.06070.3690.47540.02040.37250.381266%H230.05520.11270.20950.02270.18510.191659%H240.05120.06070.15050.02510.16790.173551%Exposure of disaster-bearing carriersE10.06660.37480.49170.0290.28430.291656%E20.06470.24330.35670.02820.36160.367656%E30.04450.11090.1890.02430.20250.209245%E40.06090.06330.170.02510.140.143159%Vulnerability of disaster-bearing carriersV10.0890.30770.46360.02810.38420.388968%V20.0530.2110.30390.02340.24990.254153%V30.04830.12740.21210.01910.17190.175560%V40.05830.06090.16310.01840.11190.114168%V50.04420.03560.11320.01070.07410.075476% Statistics on AHP and MC-AHP indicator scores. Discussion Interpretation of results and comparison with previous studies This assessment reveals a concentric spatial distribution of flood risk, radiating from the high-risk urban core toward the lower-risk suburbs. This pattern is consistent with observations in other megacities with extensive metro network, such as Beijing15 and Shanghai7, where urban cores are also identified as flood risk hotspots due to concentrated infrastructure and population. However, this study deepens that understanding by dissecting the contributing factors: in Chengdu’s city center, the extreme risk is predominantly driven by high vulnerability (e.g., immense passenger flow, high station density), even though the hazard level is relatively moderate. This contrasts with areas such as Dujiangyan, where high risk is mainly attributable to hazard factors (e.g., extreme rainfall). This dichotomy underscores the critical importance of integrating the three dimensions of hazard, exposure, and vulnerability in metro flood risk assessments. Assessments based solely on a single dimension may yield incomplete or potentially misleading conclusions. Uncertainty analysis Despite the establishment of a robust assessment framework, several sources of uncertainty should be acknowledged. First, the hazard analysis relies on historical rainfall data and static environmental factors. Future climate change scenarios, which are expected to increase the frequency and intensity of extreme rainfall events, were not incorporated, potentially leading to an underestimation of future hazards. Second, although the vulnerability indicators are comprehensive, more precise physical vulnerability data—such as the actual elevations of station entrances—were not included in this study, though such data could further enhance the model’s accuracy. Third, while the MC-AHP method effectively handles uncertainty in weight assignment, it remains dependent on the initial construction of the hierarchy and the selection of indicators, both of which are based on expert opinion and literature. Although uncertainty in the weighting process was quantified, the structural uncertainty of the model itself warrants further exploration. Future research Building upon this study, future research could explore more advanced computational techniques to further enhance the accuracy of flood risk assessments. Data-driven methods, including probabilistic frameworks and optimization algorithms, can be employed to mine complex nonlinear relationships between risk indicators and historical flood events, potentially uncovering hidden patterns that traditional MCDA methods might miss43. For instance, Sadeghfam et al.44 developed a probabilistic framework that integrated frequency analysis and Particle Swarm Optimization (PSO) to objectively delineate climate suitability for agriculture, significantly reducing subjectivity and demonstrating the potential of data-driven approaches to enhance assessment accuracy. Furthermore, more sophisticated uncertainty analysis frameworks, such as the Generalized Likelihood Uncertainty Estimation (GLUE) method —which has been successfully applied to quantify uncertainty and address equifinality in environmental models, such as groundwater vulnerability assessments45—could be introduced to the field of metro flood risk. Applying these advanced data-driven methods and GLUE techniques represents a promising frontier for future research. Practical implications and mitigation strategies The spatially explicit risk maps generated in this study have significant practical value, providing actionable decision-support information for urban planners and metro operators. The high-resolution risk classification facilitates the development of tiered emergency response strategies, enabling prioritized resource allocation based on the specific risk profile of each metro line and station. Grounded in these practical implications, the following concrete strategies are proposed according to operational domains: First, from the perspective of early warning systems, for metro lines and stations with very high flood risk, such as Line 1, Line 2, Line 4, Line 7, and stations like Shufang Station, Guangfu Station, Fuhuayuan Station, South Railway Station, Taipingyuan Station, and TCM University Provincial Hospital Station, where passenger flow is high and transfer coefficients are significant, the metro flood disaster early warning system should be further improved. This will ensure that disaster information is received promptly and effective warnings are issued to the public. Additionally, a rapid response mechanism should be established so that once early warning information is received, the metro network can quickly activate emergency plans and take necessary flood prevention measures. Second, from the perspective of facility upgrades, the construction of waterproof facilities at the entrances and tunnels of metro stations such as Shufang Station, which suffered severe water accumulation during the ‘8.16’ rainstorm, and Shengxian Lake Station, which was submerged during the ‘7.3’ rainstorm, should be strengthened. In addition to raising exit heights and installing waterstops or waterproof doors, consideration should be given to adding waterproof layers in tunnels and enhancing drainage facilities. Furthermore, for older metro lines and facilities, such as Line 1, regular inspections and maintenance should be conducted to ensure they remain in good working condition. Third, from the perspective of urban planning, flood prevention planning for underground spaces should be strengthened in high-risk areas such as Jinjiang District, Wuhou District, Jinniu District, Chenghua District, Qingyang District, Shuangliu District, Longquanyi District, Wenjiang District, and Pidu District. These areas have high annual rainfall or impermeable surface rates. In urban development, flood prevention requirements for underground spaces should be fully considered, and drainage systems and flood control facilities should be appropriately planned. It is important to avoid constructing metro stations and other critical infrastructure in flood-prone areas to reduce the impact of flood disasters on the metro network. By implementing this tiered strategy—ranging from advanced early warning responses and structural upgrades to forward-looking planning—urban planners, metro operators, and emergency management departments can move beyond reactive measures and systematically enhance the resilience of the Chengdu metro network. This study holds critical significance for maintaining urban operational continuity, ensuring passenger safety, and ultimately strengthening overall urban resilience. Conclusion This study developed a GIS-based integrated framework combining AHP and MC-AHP methodologies to assess flood risk in the Chengdu metro network. By incorporating 16 indicators across hazard, exposure, and vulnerability dimensions, the framework effectively identified high-risk zones and provided a robust tool for urban infrastructure resilience planning. The main findings include four aspects. First, the flood risk distribution exhibits a concentric pattern, diminishing from the high-risk urban core to the lower-risk suburbs. Second, the MC-AHP method proved superior to the traditional AHP, identifying a greater proportion of high-risk areas (8% vs. 7% for very high-risk) and effectively quantifying uncertainties in expert judgment through β-PERT distributions, thereby yielding more reliable and accurate risk maps. Third, the assessment of 14 operational and 5 under-construction lines revealed that Lines 7, 4, and 2 contain the highest concentrations of very high-risk stations, necessitating prioritized mitigation efforts. Fourth, At the station level, 54 stations (≈ 15.3%) are classified as very-high risk and 124 stations (≈ 35.0%) as high risk, with ~ 72.6% of stations at medium or higher risk, indicating substantial system vulnerability. The validation of results against historical flood events confirmed the model’s practical accuracy. The findings offer critical insights for policymakers and urban planners, highlighting the urgency of implementing tiered emergency response strategies, upgrading waterproofing infrastructure in high-risk stations, and integrating flood resilience into future urban planning. The framework can serve as a methodological benchmark for flood risk assessment in other metropolises facing similar challenges of rapid underground expansion and escalating climate risks.