A Review of Urban Flood Disaster Chain Research: Causes, Identification, and Assessment

Abstract

Urban flood disasters have become one of the most significant natural hazards under the dual pressures of rapid urbanization and intensified climate change. With the increasing interconnection among urban subsystems, these disasters often evolve into urban flood disaster chains, characterized by cascading failures across infrastructure, environment, and society. Current research hotspots mainly focus on three key aspects: the formation mechanisms, identification methods, and risk assessment approaches of urban flood disaster chains. In terms of formation mechanisms, most studies qualitatively describe the triggering and transmission processes of cascading events, revealing how interactions among hazard-inducing factors, disaster-formative environments, and disaster receptor generate chain reactions. Identification methods are categorized into four paradigms: qualitative identification based on experiential reasoning, semantic identification driven by data, structural identification through model inference, and behavioral identification using simulation modeling. Risk assessment approaches include historical disaster analysis, indicator-based evaluation models, uncertainty models, numerical simulation models, and intelligent algorithm models that integrate machine learning with physical simulations. The review finds that, due to the scarcity and heterogeneity of disaster chain event data, existing studies lack a unified quantitative framework to represent the mechanisms of urban flood disaster chains, as well as dynamic identification and assessment methods that can adapt to their evolutionary processes. Future research should focus on developing integrated mathematical paradigms, enhancing multisource data fusion and causal reasoning, and constructing hybrid models to support real-time risk assessment for urban flooding disaster chains.

1. Introduction

Driven by rapid urbanization and intensified climate change, both the frequency and severity of urban flood disasters have increased markedly in recent decades [1,2]. The spatial concentration of assets within cities and the dense interconnections among functional zones, through transportation, power, communication, and industrial energy networks, have transformed cities into highly coupled and complex systems. Under such circumstances, urban flooding is no longer a single hydrometeorological event but a cascading disaster chain arising from the interactions between natural hazards and multiple layers of urban infrastructure and sociotechnical systems [2]. For instance, during the “7·20” Zhengzhou extreme rainfall event, floodwaters triggered a sequence of cascading effects, including explosions at an aluminum plant and the paralysis of the subway system, resulting in severe casualties and economic losses [3,4]. Such cases illustrate that urban flood disasters typically propagate through complex interdependent systems, generating secondary and derivative hazards that extend well beyond the initial flood impact.

In view of the complexity and high destructive potential of urban flood disaster chains, numerous scholars have conducted in-depth studies on their formation mechanisms, identification methods, and risk assessment techniques. At present, the academic community has reached a relatively consistent understanding of the basic concept of urban flood disaster chains, which are regarded as a specific form of disaster chains whose formation and evolution follow the tri-element framework of hazard-inducing factors, disaster-formative environment, and disaster receptors [5]. The hazard-inducing factors, such as extreme rainfall, riverine flooding, and the failure of hydraulic structures, initiate the chain by generating the initial driving forces. The disaster-formative environment, encompassing urban topography, drainage capacity, land-use configuration, and ecological infrastructure, defines the physical context and transmission pathways for flood propagation. Variations in these environmental characteristics lead to a pronounced spatial heterogeneity in flood impacts, even with similar hazard intensities [6]. Disaster receptors, including urban lifeline systems (transportation, power supply, communication, and water networks), underground spaces, and residential and industrial zones, determine the extent to which flood impacts are amplified or transmitted. Their high exposure and vulnerability often serve as the key nodes through which disaster effects cascade and spread. In typical events, extreme rainfall first interacts with the disaster-formative environment to generate overland flow and overwhelm the drainage system, initiating flooding. This process subsequently disrupts exposed infrastructure systems, leading to traffic paralysis, power outages, communication failures, and even chemical explosions—forming temporally continuous and causally linked cascading disaster processes [7]. However, such understanding remains largely conceptual, lacking quantitative models that can explicitly capture the dynamic formation and propagation of urban flood disaster chains. Research on the methodological aspects of urban flood disaster chains remains in a preliminary stage. Key challenges include how to quantitatively represent the formation and evolution processes of disaster chains, how to identify their structural connections within urban systems, and how to effectively assess the associated risks. In response to these issues, numerous scholars have conducted extensive studies based on various urban flood disaster chain events and have proposed a range of methodological approaches and solutions. Therefore, a systematic synthesis of existing knowledge is essential to clarify the causal linkages among flood-induced disasters, summarize advances in identification and modeling techniques, and outline the methodological foundations for future work.

Accordingly, this paper provides a comprehensive review of the existing research on urban flood disaster chains, focusing on the latest advances in their formation mechanisms, identification methods, and risk assessment techniques. It further summarizes and analyzes the potential directions for future development, aiming to advance the theoretical and methodological understanding of urban flood disaster chains and to provide a solid reference framework for future research and practical disaster management.

2. Research Progress on the Formation Mechanisms of Urban Flood Disaster Chains

Urban flood disaster chains are rarely composed of isolated events. Rather, they evolve through multi-stage, multi-system coupling processes driven by the interactions between hazard-inducing factors and disaster receptors, leading to progressive transmission and amplification effects. A typical disaster chain exhibits a hierarchical progression of primary–secondary–derivative hazards. Significant causal links exist between these stages; for example, under extreme rainfall conditions, intensified surface runoff exceeds the design capacity of the drainage system, leading to widespread inundation of subway tunnels; the resulting power system short circuits and failures cause large-scale transportation paralysis, which subsequently hampers the emergency response and undermines overall social stability [7]. Such cascading relationships reflect the strong interdependence and coupling among urban subsystems, particularly within lifeline systems such as electricity, communication and transportation, which frequently act as critical conduits for multi-hazard propagation [8]. Existing studies on the formation mechanisms of urban flood disaster chains have primarily focused on process-based case analyses of specific events, while systematic and generalized interpretations remain limited. Synthesizing current research findings, the formation mechanisms of urban flood disaster chains can be categorized into two fundamental processes: disaster chain triggering and disaster chain transmission Figure 1.

2.1. Disaster Chain Triggering

Disaster chain triggering refers to the process in which an initial hazard-inducing factor exceeds the system’s defense threshold within a particular disaster-forming environment, leading to functional instability of disaster receptors and the initiation of secondary disasters. The core mechanism lies in the coupling between the hazard-inducing factor and the disaster-forming environment, whereby disaster effects breach local constraints and spread outward, indicating the onset of cascading hazards. Typical triggering modes for urban flood disaster chains include the following four types:

(1) Rainstorm-induced waterlogging. When the intensity or duration of extreme rainfall exceeds the design capacity of the urban drainage system, surface runoff accumulates rapidly and produces widespread ponds. This process damages roads, underground spaces, and buildings and may trigger traffic paralysis and power outages. It is jointly controlled by surface imperviousness and drainage capacity, exhibiting short duration, strong suddenness, and spatially fragmented patterns.

(2) River overflow. During regional rainstorms or upstream floods, when the river stages exceed warning levels, an overflow occurs and directly impacts the flood protection systems and exposed assets along the riverfront. The source of triggering may originate from external inflows or be compounded by local runoff. Because rivers possess inherent storage and conveyance capacity, such processes tend to evolve more slowly, with fixed spatial locations and greater predictability.

(3) Instability of flood control structures. When critical hydraulic infrastructure, such as reservoirs or levees, fails under prolonged high water levels or structural damage, the resulting breach flood exerts concentrated impacts on downstream areas, producing severe secondary disasters. This type of triggering is characterized by extreme destructiveness, high suddenness, low predictability, and rapid cascading reactions.

(4) Infiltration by mountain floods. In mountainous or piedmont cities, storm-generated flash floods can rapidly converge to urban areas, overwhelming drainage systems, inundating large areas, and damaging infrastructure. This type of triggering displays the highest uncertainty and suddenness, as well as the greatest potential destructiveness.

The triggering of urban flood disaster chains depends on whether the initial hazard intensity surpasses the system’s defense threshold, a critical condition marking the transition from a latent state to cascading reactions. This threshold is jointly determined by the hazard intensity, the defense capacity of the disaster-forming environment, and the resistance of disaster receptors. Most existing studies emphasize hydraulic or hydrodynamic thresholds, such as flow velocity, discharge, water depth, and inundation duration, while comprehensive assessments that incorporate environmental factors (e.g., drainage configuration, topographic convergence) and infrastructure attributes (e.g., protection level, redundancy) remain insufficient.

Moreover, the triggering of urban flood disaster chains exhibits pronounced spatiotemporal compounding characteristics. Their formation is jointly influenced by the spatial distribution and temporal structure of rainfall as well as by the functional configuration of urban systems. Different subsystems display markedly distinct sensitivities to flood impacts. The transportation network, for instance, is extremely sensitive to water depth, as inundation of only a few tens of centimeters can paralyze traffic operations [9]. The failure of the power system is largely dependent on the layout and protection standards of the critical facilities. Flooding of substations or cable vaults can result in large-scale power outages. In a similar vein, the concentrated spatial arrangement of telecommunication base stations renders them exceedingly susceptible to submersion. Upon inundation, these facilities precipitate the formation of communication blind zones, which significantly impede the operational efficacy of urban environments [10]. The urban spatial structure also serves as a key controlling factor. Physical and functional interdependencies between different urban zones mean that the failure of a single subsystem can spread through transportation, power, or communication networks, triggering cascading failures [11]. Old urban districts and areas with dense underground spaces are particularly vulnerable: the former often suffer from combined aging sewer systems, low pipe density, and poor maintenance, which promote recurrent waterlogging; the latter, including subways, underground garages, and commercial complexes, are typically located in low-lying catchment areas where surface flooding can easily backflow once overtopped. The coupling between rainfall timing and urban operational rhythms further intensifies the risk. When extreme rainfall coincides with rush hours or peak drainage loads, both system exposure and vulnerability increase simultaneously, increasing the likelihood of infrastructure failures and secondary disasters [12].

In general, the triggering of urban flood disaster chains is the result of the joint spatio-temporal interaction of multiple sources of hazards that collectively exceed the defensive thresholds of the system, manifesting strong nonlinearity and pronounced regional heterogeneity.

2.2. Disaster Chain Transmission

The propagation of the disaster chain refers to the process in which, following the occurrence of secondary disasters, the effects of the disaster are transmitted and amplified between different spatial units and systems through physical, functional, or informational pathways [13,14,15]. Compared with the triggering phase, the propagation phase emphasizes the diffusivity, interconnectedness, and systemic nature of cascading effects [16,17,18]. In urban systems, flood hazards not only propagate spatially through hydrodynamic processes but also spread across subsystems through infrastructure failures and functional dependencies, leading to broader socioeconomic disturbances and functional collapses [11,19].

Mechanically, the propagation of the disaster chain requires the simultaneous fulfilment of two fundamental conditions: disaster overflow [20] and propagation pathways [16,21]. Disaster overflow is the prerequisite, which describes the process by which the energy, material, or information of a disaster breaches the defensive boundary of the original system and spills into the external environment. Only when disaster effects exceed the tolerance threshold of a disaster receptor, causing functional instability, can the impact extend outward and enter the cascading phase. For example, overloaded drainage systems, levee breaches, and short circuits in the power system can break physical or functional boundaries, triggering chain reactions in adjacent systems. The propagation pathways constitute the necessary condition, representing the physical, functional, or informational channels through which the effects of the disaster are transmitted across space or between systems. In the context of urban flooding, these pathways include natural geographic conduits (e.g., rivers, gullies, and low-lying terrain) and engineered infrastructure networks (e.g., transportation, power, communication, and water supply systems). These channels form the medium through which energy and functional dependencies propagate, enabling disasters to transmit and amplify across multiple systems and spatial scales. If the intensity of overflow disaster is insufficient or transmission pathways are blocked, cascading effects terminate; conversely, when both coexist and reinforce each other, the disaster chain evolves into a systemic catastrophe.

The propagation of the disaster chain exhibits pronounced spatio-temporal dynamics and system dependence. The temporal causality among disaster events, the spatial spread of impacts, and the dynamic evolution of the disaster-formative environment and disaster receptors together shape the core mechanism of propagation. During this process, floods often carry material, energy and information flows that, through frequent and high-intensity interactions, trigger multiple types of secondary and derivative hazards, producing multidimensional amplification effects [22]. These cascading impacts can occur sequentially in time or overlap spatially, forming compound disasters [23]. Such responses not only alter the magnitude and extent of hazards, but also, through inter-system coupling, exacerbate vulnerability and response delays, amplifying overall risk [24]. Some propagation processes involve chemical or biological media; for instance, floodwaters can erode and mobilize pollutants, causing environmental contamination, while post-flood conditions can favor vector proliferation and epidemic outbreaks. In addition, socioeconomic disruptions such as infrastructure damage, supply chain disruption, and functional paralysis also constitute critical manifestations of disaster chain transmission [25].

The propagation pathways of urban flood disaster chains can be further categorized into natural geographic conduits and infrastructure networks [6]. The former refers to naturally formed physical channels such as rivers, gullies, and terrain corridors through which disaster effects are transmitted to downstream or low-lying areas. For example, during the extreme rainfall event on 23 July 2023 in Mentougou District, Beijing, intense rainfall combined with inappropriate land use caused 353 geological hazards, including collapses, landslides and debris flows, forming a typical cascading chain ’rainstorm-flood-geological hazards’ that destroyed houses, disrupted traffic, and caused casualties. The latter category includes man-made urban channels such as roads, railways, power grids, and communication pipelines, which possess both physical and functional propagation characteristics. Once critical nodes fail, disasters can rapidly spread along these networks, resulting in multi-system cascading collapse. The subway system, as a nexus of transportation and underground space, is an example of a major propagation hub: once inundated, it not only paralyzes transport, but also triggers power, communication, and emergency system failures, rapidly extending the effects of disasters in lifeline infrastructures and public service systems [26,27,28].

From the perspective of spatiotemporal structure and coupling mode, the propagation of urban flood disaster chains can be classified into sequential, parallel, and compound types. Sequential chains involve disasters occurring successively with strong causal dependence; parallel chains feature multiple disasters triggered within the same temporal window by a common initiating factor; and compound chains emerge when multiple pathways intersect and interact, forming feedback loops characterized by high nonlinearity and uncertainty [29]. The propagation process also exhibits notable time-lag effects, as the onset of secondary hazards can vary from several hours to many years [30]. For example, earthquake-induced landslides that dam rivers can form barrier lakes whose eventual breaching is delayed by days, or even centuries, while pipe ruptures and pollution events caused by heavy rainfall may manifest only days later. Fan et al. [31] emphasized that such delay effects impart long-term persistence and recurrence to disaster risks, implying that post-disaster recovery and subsequent risk assessments must account for prolonged chain feedback mechanisms.

In summary, the propagation of urban flood disaster chains constitutes a cross-system, multiscale cascading diffusion process, representing both the extension of physical hydrodynamic interactions and the manifestation of socio-technical system failures. Energy and information are continuously transmitted and superimposed across interconnected systems, allowing risks to accumulate and amplify during propagation, ultimately transforming localized disturbances into large-scale systemic crises. Understanding these propagation dynamics reveals the inherently non-linear nature of disaster diffusion within complex urban systems and provides a theoretical foundation for disaster chain early warning and interruption strategies.

3. Research Progress on the Identification of Urban Flood Disaster Chains

The identification of flood disaster chains aims to reveal causal relationships, spatio-temporal correlations, and cascading evolution patterns among different disaster events. By clarifying key nodes, propagation pathways, and critical conditions of disaster chains, such studies provide a fundamental basis for risk assessment, loss estimation, and disaster prevention and control decision making [32].

Currently, research on the identification of urban flood disaster chains can be broadly categorized into two methodological paradigms. The first is qualitative inference and deductive analysis based on historical experience and expert knowledge, which identifies disaster chains by summarizing their common patterns and typical evolution mechanisms [33,34]. The second is quantitative reasoning and intelligent identification driven by data and models, which leverages multi-source information, knowledge representation, and computational modeling to automatically extract the structural relationships among disaster events [7]. From the perspective of recognition logic, existing approaches to urban flood disaster chain identification can be further classified into four categories: (1) qualitative identification based on experiential reasoning, (2) semantic identification driven by data, (3) structural identification based on model inference, and (4) behavioral identification based on simulation modeling. The following sections elaborate on these four categories in detail.

3.1. Qualitative Identification Based on Experiential Reasoning

The qualitative identification method based on experiential reasoning is primarily based on historical disaster observations, documentary records, and expert knowledge. By systematically analyzing past disaster events, it summarizes the triggering conditions, propagation pathways, and consequence characteristics of disaster chains, thus forming a deductive conceptual framework for chain evolution [35]. Due to its minimal data requirements, this approach remains a fundamental tool for identifying flood disaster chains, particularly in contexts where systematic data collection and modeling capabilities are limited.

Event Tree Analysis (ETA) [36] represents the most typical form of qualitative identification based on experiential reasoning. The general procedure involves first selecting an initial event node (e.g., extreme rainfall, river overflow, or levee overtopping) and then progressively determining the branching pathways and intermediate events along possible propagation routes of the disaster chain, ultimately constructing different sequences of outcomes [37]. In the context of urban flooding, researchers commonly collect representative case studies and enumerate potential events, such as heavy rainfall, inundation, geological hazards, and infrastructure damage, before building tree-based events or evolutionary networks based on expert judgment [38,39]. The ETA method offers clear structure, intuitive logic, and operational simplicity, making it well suited for scenario construction and preliminary mapping of disaster chains. It also helps to identify key nodes and the main propagation pathways. Frequently, ETA serves as a structural prototype for subsequent complex-network or Bayesian-network modeling, providing a prior framework for quantitative analysis.

For example, Guo et al. [8] analyzed 87 cases of urban rainstorm disasters and extracted 16 representative nodes, including rainfall, landslides, flooding, pipeline failures, and traffic congestion, to construct an evolutionary network ’rainstorm-urban lifeline system’ consisting of 16 nodes and 29 edges. Their analysis revealed that rainstorms can trigger secondary hazards such as landslides, flooding, and debris flows, which in turn lead to pipeline damage, water pollution, and power outages. Similarly, Li et al. [39] developed an event tree for urban waterlogging, designing urban flooding as the main event and including unsafe human behavior, unsafe urban environment, and management deficiencies as intermediate events. Using logical gates, “AND” and “OR”, they decomposed causal relationships and further performed a structural importance analysis, identifying “insufficient drainage capacity” and “inadequate emergency response” as the most influential basic events affecting the probability of the event.

In summary, the experiential reasoning-based qualitative identification method offers the advantages of low data dependency, strong interpretability, and intuitive logic, which makes it particularly effective for prototype identification and preliminary screening of key links in disaster chain studies. However, its high subjectivity and limited capacity to quantify complex interdependencies among disaster events constrain its application. Consequently, this method is often integrated with data-driven or model-based approaches to enhance the objectivity, dynamism, and comprehensiveness of disaster chain identification.

3.2. Semantic Identification Driven by Data

The semantic identification approach uses unstructured data sources, including social media posts, news reports, remote sensing images, government bulletins, and field investigation reports, to automatically extract disaster events and their interrelationships through natural language processing (NLP) and semantic analysis techniques, thereby enabling the dynamic identification of disaster chains [7].

In practical applications, researchers typically crawl and analyze large-scale text streams from social media and news platforms. Using lexical analysis, Named Entity Recognition (NER), and dependency parsing, researchers automatically extract structured event–time–location–relation quadruples from unstructured text sources. Recent studies have further enhanced the robustness and granularity of event extraction by combining deep learning architectures (e.g., BERT, LSTM-CRF) with Convolutional Neural Networks (CNNs) for multimodal text–image fusion [40,41]. Once extracted, these structured events are temporally aligned through timestamp synchronization and spatially positioned via geocoding, forming event chains with both spatial trajectories and temporal sequences. Such event representations enable dynamic reconstruction of disaster evolution, particularly suited for identifying localized, short-duration, and rapidly cascading processes typical of urban flood disasters [42,43].

Based on structured event information, researchers have integrated semantic data into Knowledge Graph (KG) frameworks to represent complex disaster causality. In these systems, disasters are described through triplet structures (entity–attribute–relation), representing logical relationships among hazard-inducing factors, disaster-formative environments, disaster receptors, and consequential impacts [44,45]. Through rule-based reasoning or embedding-based learning approaches (e.g., TransE, RotatE) [46], these models can perform link prediction and causal inference, thus automatically identifying critical nodes such as trigger points, explosion points, and diffusion points within the disaster chain. To improve both precision and timeliness, recent studies have adopted cross-modal data fusion frameworks that integrate remote sensing imagery, IoT sensor data, and social media streams [47,48]. By aligning event evidence across image–text–sensor modalities and constructing event timelines and spatial topological grids, these models extend disaster chain perception from single time slices to multiscale spatio-temporal dynamics. This enables continuous monitoring and retrospective analysis of rapidly evolving urban flood chains, providing a foundation for early intervention and adaptive emergency response.

Although the semantic identification method fully exploits multisource and multimodal data, it also faces challenges such as high information noise and uneven data quality. Therefore, it is often necessary to perform cross-validation with structured datasets to enhance the robustness and reliability of disaster chain identification.

3.3. Structural Identification Based on Model Inference

The structural identification approach focuses on uncovering causal dependencies between disaster events and employs complex graph-theoretic network models to represent and infer the structural organization of disaster chains [49,50]. Within this framework, disasters, infrastructure components, or functional units of urban systems are abstracted as nodes, while the transmission or dependency relationships between them are modeled as directed edges, forming a network representation of the disaster chain.

Complex network analysis and Bayesian network (BN) modeling are the two most widely used techniques in this domain [51,52]. In complex network analysis, topological indicators such as degree centrality, betweenness, proximity, and vulnerability are calculated to identify critical nodes and bottleneck links within the disaster propagation network. Furthermore, cascade failure models, threshold models, and multilayer network models have been developed to simulate how disasters propagate across interconnected systems and to capture amplification effects resulting from system interdependencies [53]. These models are particularly effective in revealing cascading transmission characteristics of urban flood disasters within lifeline systems, such as transportation, power, and communication networks, and are well suited for studying multihazard coupling mechanisms and cross-system interactions.

Compared with other approaches, structural identification offers strong systematicity, quantitative interpretability, and network-level explanatory power, allowing researchers to depict the spatial functional organization of complex disaster systems. However, its performance heavily depends on the availability of comprehensive data sets and reliable prior structural knowledge. The accurate determination of network topology and the precision of parameter learning within probabilistic inference frameworks, particularly Bayesian networks, remain key methodological challenges that constrain its large-scale application and real-time implementation.

3.4. Behavioral Identification Based on Simulation Modeling

The behavioral identification approach builds on the theoretical foundation of system dynamics and the human–object–environment interaction framework, employing simulation-based modeling techniques, such as Agent-Based Models (ABM) to reproduce spatiotemporal evolution and cascading propagation of urban flood disaster chains [54,55]. In this framework, entities such as residents, vehicles, pumping stations, emergency teams, and substations are represented as agents that interact under hydrodynamic, infrastructural, and behavioral coupling rules [56]. These models capture the coevolution of human decision-making, infrastructure functionality, and environmental processes, revealing the feedback loops, nonlinear dependencies, and adaptive patterns that govern cascading flood impacts [57].

Building upon these foundational concepts, recent research has advanced the integration of behavioral modeling with physical system simulations to represent the coupled dynamics of human, infrastructure, and environmental components in urban flood disaster chains. Refined sociohydrological frameworks, such as the Flood-Agent-Institution (FAI) model, incorporate behavioral responses, institutional interventions, and flood diffusion processes to illustrate how governance and public decision-making interact with hydrological systems [58,59]. Currently, studies have focused on identifying dynamic exposure and feedback mechanisms by embedding cascading processes, such as traffic congestion, emergency delays, and infrastructure failures, into simulation models, thus reproducing realistic pathways of evolution of disasters and temporal dependencies between subsystems [60,61,62]. Data-driven calibration and the integration of IoT, remote sensing, and in situ monitoring data have further strengthened the accuracy of the model and reduced epistemic uncertainty [63,64]. Multi-agent coordination models have also been developed to explore interactions between communities, self-organized groups, and government agencies under temporal and resource constraints, revealing collaboration and conflict mechanisms during the response and recovery phases [65,66,67]. In addition, behavioral simulations have been increasingly applied to multiscenario evaluation and policy testing in climate and urbanization scenarios, supporting adaptive flood risk management and resilience planning.

In general, simulation-based behavioral identification provides a robust framework that bridges physical hydrodynamics and social behavioral responses. It enables dynamic, predictive, and scenario-based analysis of the evolution of the disaster chain, offering new insights into the resilience of the urban system. Despite current challenges related to parameter calibration, data integration, and computational demands, the convergence of ABM, SDM, and data-driven intelligence represents a promising direction for developing next-generation urban flood disaster chain identification and management systems.

4. Research Progress on Risk Assessment of Urban Flood Disaster Chains

The risk of chain-induced flood disasters in urban areas is determined by the vulnerability, exposure, and hazard of the affected elements. The assessment of urban flood disaster chains faces not only the common challenges encountered in traditional disaster chain studies, but also additional complexities arising from the diverse types and intricate attributes of urban exposed elements. Unlike previous research that focused mainly on the evolution of disaster processes in natural environments, urban flood disaster chains involve a wide range of affected systems, such as residential areas, transportation networks, power infrastructure, and underground spaces characterized by high heterogeneity in spatial distribution, functional properties, and disaster resilience. This complexity significantly alters the spatiotemporal patterns of flood risk and imposes higher demands on the assessment of urban flood disaster chains [68]. At the same time, the increasing variety and availability of disaster-related data have laid a solid analytical foundation for urban flood risk assessment. Based on different theoretical foundations, current approaches to evaluating risks associated with urban chain-induced flood disasters can be broadly categorized into five types: historical disaster analysis, indicator-based models, uncertainty models, numerical simulation models, and intelligent algorithm models [49,69]. The details of these methods are shown in

Table 1. Comparison of Urban Flood Disaster Chain Risk Assessment Methods.

Assessment Method	Assessment Process	Advantages	Limitations
Historical Disaster Analysis	Based on historical flood disaster data and regional characteristics; statistical methods are used to identify the flood disaster chain risk characteristics.	Simple analysis process; suitable for regional-scale flood risk studies; capable of identifying historical flood chain risks.	Inadequate for forecasting future flood risks; exhibits limited capability in characterizing flood-risk variations over different locations and periods.
Indicator-Based Evaluation Model	Select key indicators related to flood risk; construct an evaluation dataset; determine the relative weights; calculate a composite evaluation index.	Captures the driving mechanisms of flood risk in a systematic manner; efficient for quantitative evaluation of flood risk at multiple scales.	Requires extensive data collection and preprocessing; subjectivity in indicator selection and weighting may affect the objectivity of results.
Uncertainty Model	Conduct feature analysis of disaster chain nodes; construct network structure; compute node probabilities to calculate overall flood risk characteristics.	Considers the uncertainty and complexity of disaster chain evolution; suitable for expert-driven probabilistic inference and risk assessment.	Relies heavily on historical data or expert judgment; model building is complex and may introduce subjectivity; results may lack generalizability.
Numerical Simulation Model	Develop hydrological and hydrodynamic models; simulate flood evolution under different scenarios; analyze the flooding process.	Accurately simulates physical flood processes; allows for scenario-based assessments under varying rainfall or infrastructure conditions.	Constructing the model requires considerable time and resources; the simulation may become less efficient in complex and large environments.
Intelligent Algorithm Model	Collect and generate training samples; optimize model parameters; predict spatial variation in disaster chain risk based on key drivers.	Effectively mines patterns in historical data; suitable for dynamic and large-scale prediction of flood chain risks.	Requires large computational resources and sufficient training data; has “black-box” nature with relatively low interpretability of results.

.

4.1. Historical Disaster Analysis Method

The historical disaster analysis method aims to identify the risk characteristics and evolutionary patterns of urban flood disaster chains through retrospective investigations of past flood events [70]. By integrating multi-source historical data—such as hydrometeorological observations, disaster loss statistics, remote sensing imagery, and documentary archives—this approach reconstructs the spatiotemporal propagation pathways and cascading impacts of historical disasters, thereby revealing the dominant causal mechanisms and transformation rules of disaster chain evolution.

Methodologically, this approach commonly involves multi-event comparative analysis, temporal correlation analysis, and loss evolution modeling to extract the common evolution patterns and critical chain transitions across historical flood events. One research direction focuses on establishing statistical association models linking hazard drivers, disaster processes, and final losses to identify transmission pathways and cascading amplification mechanisms [71]. Another research direction emphasizes spatial reconstruction of disaster expansion trajectories and infrastructure performance degradation using geospatial analysis and remote sensing techniques, thereby revealing how urbanization and the spatial configuration of infrastructure influence propagation patterns [72]. In addition, long-term historical disaster databases are often mined through regression analysis, clustering, and frequency analysis to examine changes in exposure, failure frequencies of key nodes, and vulnerability evolution, enabling regional-scale disaster chain risk identification [5,73].

Compared with forward simulation-based methods, historical disaster analysis is highly data-grounded, computationally efficient, and well suited for capturing long-term and regional flood risk shifts. It helps identify recurrent propagation pathways and risk transition characteristics under different socioeconomic development stages. However, its performance depends heavily on the completeness, spatial granularity, and temporal resolution of historical datasets, limiting the ability to capture newly emerging risk drivers under climate change and compact urbanization [74].

To address these limitations, recent research has begun integrating historical analysis results with network inference, knowledge graph construction, and predictive simulation frameworks. By transforming empirical event-derived knowledge into structured and inferable forms, researchers aim to enhance forward-looking disaster chain identification and support proactive, resilience-oriented urban flood risk management.

4.2. Indicator-Based Assessment Models

Indicator-based assessment methods characterize urban flood disaster chain risks through the development of representative indicators. These indicators reflect hazard-prone environments, hazard drivers, exposure conditions, and system vulnerabilities, which are then integrated to form a comprehensive risk profile. Through the construction of a multi-dimensional index system—typically involving risk, exposure, vulnerability, and resilience—this approach synthesizes heterogeneous influencing factors into a composite disaster chain risk index, enabling regional comparative analysis and hierarchical risk classification.

In terms of methodological procedures, this approach generally includes: (1) indicator screening based on disaster chain elements, expert knowledge, statistical correlations, or sensitivity analysis; (2) weight determination through Analytic Hierarchy Process, Entropy Weight Method, or multi-criteria decision-making techniques; and (3) risk aggregation using weighted summation, fuzzy logic, or spatial overlay analysis to generate integrated disaster chain risk patterns [2,75,76]. By mapping the spatial differences in hazard-bearing environments, functional infrastructure, and emergency response capability, this method allows intuitive visualization of risk hotspots and potential cascading impact areas across different spatial scales.

The indicator-based method features high interpretability, transparent structure, and efficient implementation. It is especially suited for regional planning and rapid evaluation where computational complexity and data processing requirements must remain manageable [77,78]. Additionally, it provides a practical means for early-stage disaster chain assessment in data-limited environments. However, the indicator-based approach often suffers from subjectivity in the selection of representative indicators and determination of indicator weights [79]. The method also depends on the availability of diverse socio-environmental datasets, and discrepancies in data resolution may lead to bias in risk assessment results [80,81].

To improve objectivity and dynamic representation, recent research efforts have focused on integrating indicator-based frameworks with machine learning optimization, network-based causality modeling, and scenario simulation approaches. This hybridization allows the inclusion of disaster chain triggering thresholds, functional degradation processes, and temporal evolution characteristics, enhancing the reliability and predictive capability of indicator-based flood disaster chain assessments under changing climate and urbanization conditions.

4.3. Uncertainty Models

Uncertainty-based modeling focuses on capturing the inherent randomness, ambiguity, and complexity within urban flood disaster chains by leveraging statistical inference and conditional probability theory. This approach represents disaster events as probabilistic nodes and their cascading interactions as conditional dependencies, thereby establishing a stochastic evolution framework of flood-induced disaster chains for risk analysis under uncertain conditions [82,83].

Methodologically, uncertainty-based models commonly include Bayesian networks, probabilistic graphical models, stochastic simulation, and other statistical inference tools [84,85,86]. Model construction typically consists of: (1) structural learning, which determines dependency relationships among disaster events through data-driven algorithms or expert knowledge; (2) parameter learning, which estimates conditional probability distributions using historical datasets; and (3) probabilistic inference, which calculates evolving disaster states and cascading failure probabilities based on posterior probability updates. By quantifying the likelihood of disaster propagation under multifactorial interactions, these models support scenario reasoning, failure pathway identification, and uncertainty-informed decision making.

Compared with deterministic physics-based models, uncertainty-based approaches provide strong causal interpretability, sensitivity to probabilistic dependencies, and good adaptability to incomplete or noisy data [87]. They are particularly capable of characterizing stochastic triggering mechanisms and cascading amplification effects in complex urban systems, supporting real-time risk prediction and dynamic warning applications. However, the reliability of uncertainty modeling largely depends on the availability and credibility of historical data and expert inputs, which may introduce subjectivity into both structural specification and conditional parameter estimation [88,89]. In addition, traditional uncertainty models struggle to represent spatiotemporal correlations and cross-system couplings that are common within flood disaster chains, which may limit their ability to accurately predict evolution under rapidly changing environmental and socioeconomic conditions.

To overcome these challenges, emerging research trends integrate uncertainty-based models with machine learning, physical hydrodynamic simulations, and knowledge graph reasoning, enabling hybrid representations of both mechanistic processes and probabilistic dependencies [90,91]. This convergence is expected to enhance the predictive stability, generalization capability, and real-time applicability of uncertainty modeling in urban flood disaster chain management.

4.4. Numerical Simulation Models

Numerical simulation models based on physical mechanisms play a central role in quantifying and understanding the evolution of urban flood disaster chains. These models can dynamically reproduce the interactions among the hydrological, geomorphological, and engineering components within complex urban environments. By constructing multiple simulation scenarios, they are able to trace the initiation, propagation, and transformation of disaster chains under specific boundary and forcing conditions, thereby facilitating the assessment of chain-induced flood risks at multiple spatial and temporal scales [92,93].

In recent years, hydrodynamic models, such as 1D/2D coupled shallow-water solvers, physically based rainfall–runoff models, and integrated hydrology-hydrodynamics platforms, have been widely applied to simulate the cascading effects of floods across interconnected urban subsystems. For example, Fan et al. [94] developed a physically based modeling framework to simulate the spatiotemporal evolution of landslide-induced dam-break floods, capturing the downstream propagation dynamics under different trigger scenarios. Similarly, Zhang et al. [95] applied a high-resolution hydrodynamic model to simulate urban flood development and integrated the simulation outputs with spatial network analysis, successfully identifying spatial patterns and evolution pathways of urban flood disaster chains. Other studies have coupled hydraulic models with geotechnical and infrastructure modules to represent the propagation of embankment failure, culverts, and drainage networks, thus revealing the inter-system links between natural processes and engineered systems [96,97].

Compared with empirical or statistical models, numerical simulation approaches are grounded in well-defined physical principles, enabling explicit representation of the governing processes and feedback mechanisms in flood evolution. They offer the advantage of high interpretability and transferability, allowing scenario-based assessment of disaster propagation, functional degradation of infrastructure, and spatial heterogeneity of cascading risks. In addition, recent advances in data assimilation [98,99], remote sensing integration [100,101,102], and high-performance computing (HPC) [103,104] have significantly improved the precision, spatial fidelity, and computational efficiency of such models. Hybrid modeling frameworks that integrate hydrodynamic solvers with data-driven modules, such as machine learning-based boundary condition estimation or surrogate modeling, are emerging as powerful tools for near-real-time disaster chain simulation [105].

However, the development and application of physically based numerical models remain computationally demanding and data intensive. Model calibration and validation require detailed spatial data on topography, land use, and infrastructure networks, as well as high-resolution temporal records of rainfall, flow, and flooding. When applied to large-scale or high-resolution urban domains, simulation efficiency can be constrained by the complexity of model coupling and the volume of computation required. Furthermore, most existing models still struggle to represent human-infrastructure interactions, sociotechnical feedbacks, and behavioral adaptations, which are critical components of disaster chain dynamics in urban settings.

4.5. Intelligent Algorithm Models

Intelligent algorithm models, represented by machine learning and artificial intelligence (AI) techniques, have become an increasingly important tool for assessing the risks associated with urban flood disaster chains. Unlike traditional physically based models that rely on explicit hydrodynamic equations, intelligent models learn the complex non-linear relationships between hazard-inducing factors, environmental conditions, and secondary disaster impacts directly from data. Using large-scale historical disaster records, environmental indicators, and socioeconomic variables, these models enable efficient and high-dimensional risk characterization even when physical data or real-time observations are limited.

Recent research has demonstrated a wide range of algorithmic frameworks for the assessment of flood-related disaster chains. Zhao et al. [106] used a random forest model coupled with a multi-criteria decision analysis framework to evaluate the spatiotemporal characteristics of flood risk in Zhengzhou, China, over a 15-year period (2005–2020). They further proposed a hybrid Bayesian network–land use simulation approach to project future flood probabilities under changing urbanization scenarios. Liang et al. [107] introduced a hybrid graph convolutional neural network (GCN) and spiking neural network (SNN) model that extracts spatiotemporal dependencies from multi-source datasets, substantially improving prediction accuracy in urban flood risk analysis. Nachappa et al. [108] employed support vector machines (SVMS) and random forest (RF) algorithms to generate flood and landslide exposure maps across Salzburg, Austria, integrating 13 influencing factors to support spatial planning in high-risk regions. In addition, deep neural networks (DNN) [109], convolutional neural networks (CNN) [110], and long short-term memory (LSTM) [111] architectures have been increasingly applied to capture rainfall–runoff responses and infrastructure vulnerabilities in urban flooding situations.

Beyond single-model applications, recent advances have focused on hybrid and ensemble learning frameworks that combine the strengths of different algorithmic paradigms. For instance, machine learning models have been coupled with hydrodynamic simulations or indicator-based systems to improve both interpretability and physical consistency. Lin et al. [112] integrated key flood characteristics derived from physically based models into an indicator-driven evaluation framework, enhancing the spatial precision of urban flood risk mapping. Similarly, Hui et al. [113] developed a one- and two-dimensional coupled flood model combined with spatial correlation analysis to assess the cascading risk of power equipment failure under flood conditions. Such multi-method integration reflects an emerging trend toward hybrid modeling, where physical, statistical, and intelligent models are jointly employed to capture the multi-scale, cross-domain characteristics of disaster chain evolution.

The key advantage of intelligent algorithm models lies in their capability to process high-dimensional, heterogeneous data, uncover latent correlations among multiple hazard factors, and enable near-real-time prediction and decision support. By integrating multisource data—ranging from remote sensing imagery and IoT sensor networks to social media feeds—these models can dynamically update risk assessments and provide early warning signals with increasing precision. Moreover, the incorporation of graph-based learning and spatiotemporal attention mechanisms allows for the representation of disaster propagation pathways and inter-system dependencies within urban environments.

However, the application of AI-driven models still faces several challenges. The “black-box” nature of deep learning algorithms limits their interpretability and hinders their acceptance in risk governance and operational decision making. Model generalization remains sensitive to data imbalance and regional heterogeneity, while the scarcity of labeled disaster chain datasets restricts transferability across cities and climates. Furthermore, current AI models often lack explicit representation of physical constraints and causal logic, leading to potential inconsistencies between statistical predictions and physical realities. Future research should therefore emphasize interpretable and physically informed machine learning frameworks, integrating causal inference, attention visualization, and knowledge-guided regularization to improve model transparency and scientific validity. The combination of AI models with physical simulations, graph-based causal networks, and dynamic risk propagation models offers a promising pathway towards intelligent adaptive flood disaster chain assessment systems. Ultimately, the convergence of data-driven learning and mechanism-based reasoning will enable real-time, explainable, and scalable solutions to manage complex urban flood risks in the era of climate uncertainty and rapid urbanization.

5. Conclusions and Future Perspectives

Existing studies on urban flood disaster chains have explored their structural components, triggering factors, and transmission pathways, yet most remain at a qualitative and conceptual level. Quantitative descriptions and unified mathematical representations of their formation and evolution are still lacking, particularly regarding key parameters such as nodal spillover thresholds, risk transfer intensities, and their relationships with nodal states and pathway attributes. This review synthesizes and compares these studies, revealing that the current research on the formation mechanisms of urban flood disaster chains has not yet achieved a consistent theoretical framework capable of quantitatively depicting dynamic interactions among multiple hazard-inducing factors and disaster-bearing elements. At the same time, the identification of urban flood disaster chains has evolved from empirical reasoning toward data-driven and intelligent approaches, moving from static descriptions to dynamic modeling and from qualitative inference to causal reasoning. The application of big data analytics, deep learning, and knowledge graph techniques has enhanced the automation and accuracy of disaster chain identification. Nevertheless, the availability and quality of relevant data remain limited, with deficiencies in spatiotemporal resolution, completeness, and inter-element correlation. As a result, most studies focus on macro-level causal relationships between disaster events, while systematic identification of spatiotemporal continuity and intrinsic coupling mechanisms within specific chains remains underdeveloped. In terms of risk assessment, methodologies have progressed from direct statistical analyses and empirical models to hybrid approaches combining physical mechanism models with intelligent algorithms. These developments have improved the comprehensiveness and precision of assessments and promoted a shift from single-factor analysis toward multidimensional, system-level evaluation. However, existing studies still tend to emphasize direct correlations between initial hazards and final losses while neglecting the temporal sequencing, staged evolution, and internal coupling that characterize disaster chain development. Consequently, current assessments struggle to capture the processes of energy transfer, functional degradation, and cascading amplification that drive disaster evolution. Future research should therefore emphasize several directions. First, efforts are needed to clarify the dynamic response mechanisms of disaster-bearing elements under the combined influence of multiple hazards and complex environmental conditions, and to establish quantitative models that describe the functional degradation and instability evolution of infrastructure systems. Such models should allow for the identification of critical thresholds and spillover intensities marking the transition from functional failure to secondary disaster initiation, and formulate transmission functions that represent the multipath, nonlinear, and cross-system nature of disaster evolution. Second, advances in data mining and fusion techniques should be pursued to construct heterogeneous, multisource datasets with improved spatial and temporal resolution and semantic consistency. Integrating intelligent algorithms and causal inference frameworks will enable more accurate, fine-grained identification of disaster chain structures and dynamics. Finally, future work should focus on developing dynamic, stage-based risk assessment models that integrate physical mechanisms, real-time simulations, and intelligent algorithms. These hybrid frameworks will combine mechanistic rigor with adaptive intelligence, supporting real-time prediction, early warning, and decision-making for flood prevention and emergency management. In summary, the future development of urban flood disaster chain research depends on the coordinated advancement of mechanism understanding, identification methodologies, and risk assessment frameworks. These aspects are closely interlinked: mechanistic understanding provides the theoretical basis for accurate identification; identification offers structural and data support for dynamic risk assessment; and robust assessment enables adaptive management and resilience enhancement. Establishing a unified theoretical and computational paradigm that integrates these three dimensions will promote a transition from qualitative description to quantitative, predictive, and adaptive management of urban flood disaster chains, thereby strengthening urban resilience under growing climatic and anthropogenic pressures (Figure 2).