As urbanization progresses and the impacts of climate change become more pronounced, urban flooding has emerged as a critical challenge for resilient cities, particularly concerning urban underground spaces where flooding can lead to significant loss of life and property. Drawing upon a comprehensive review of global research on underground space flood simulation and evacuation, this paper undertakes the modelling of inundation in a substantial underground area during the extraordinary rainfall event on 7 September 2023, in Shenzhen, China. Specifically, it introduces a two-step method to simulate the coupled surface-underground inundation process with high accuracy. The study simulates the inflow processes in three types of underground spaces: parking lots, metro stations, and underpasses. Utilizing the specific force per unit width evaluation, the research examines how varying flood barrier heights influence evacuation time and inundation risk. Subsequently, the paper proposes corresponding evacuation strategies based on the obtained findings. By highlighting the vulnerability of urban underground spaces to flooding, the study underscores the urgent need for further research in this domain.

  • This study conducts a systematic review of existing research findings on urban underground space inundation, categorizing them into four types.

  • This article employs an authentic case to simulate the water ingress process and utilizes specific force per unit width evaluation to assess the associated risks.

  • Tailored strategies for addressing diverse underground spaces are proposed in this paper.

As industrialization accelerates, regions worldwide are experiencing unprecedented climate change impacts manifested by an increasing frequency of extreme rainfall events. In addition, urbanization is progressing at an unprecedented rate. As the number of structures in urban centers reaches saturation, cities are expanding not only in the two-dimensional plane but also vertically, both upward and downward. Underground spaces are increasingly being utilized for underground infrastructure, including metros, shopping malls, parking facilities, and underground complexes.

From the perspective of resilient cities, underground spaces exhibit an inverse correlation between their vulnerability and scale, ranging from structure levels to system network levels. The larger the complexity of underground facilities, the more susceptible the system becomes to external shocks (Huang et al. 2022); large-scale underground spaces that have been constructed and put into operation are highly sensitive to the profound impact of multiple disasters, including fires, earthquakes, floods, and more (Sterling & Nelson 2013).

The relatively enclosed ambiance of underground spaces, connected to the external environment by a limited number of entrances and exits, poses a significant risk during instances of underground inundation. Underground spaces such as shopping malls, metro stations, and basements are highly susceptible to inundation, leading to substantial economic losses and casualties (Zheng et al. 2019). Furthermore, various facilities and equipment installed in underground spaces, such as substations and water pumps, have the potential to cause cascading disasters during underground flooding, including electric shock hazards, power outages, water supply interruptions, building collapse, and ground subsidence.

In September 2001, heavy rainfall led to the complete suspension of the Taipei Metro for several months (Tso et al. 2011). From 1992 to 2003, the London Underground system experienced over 1,200 instances of flooding, with more than 200 incidents resulting in subway service disruptions (Arkell & Darch 2006). In August 2002, a rapid rise in river levels caused severe flooding on three metro lines in Prague, Czech Republic, leading to a prolonged suspension of subway services for 6 months (Compton et al. 2009). In May 2010, heavy rainfall in Guangzhou resulted in the inundation of multiple underground parking facilities, causing damage to 1,409 vehicles (Zheng et al. 2019). In October 2013, Typhoon Fitow caused an overflow in water channels in Shanghai. Multiple underground interchanges and tunnels were inundated, and four underground parking facilities were completely submerged (He et al. 2024). In 2020, heavy rainfall on 22 May in Guangzhou resulted in flooding on Metro Line 13 (Wang et al. 2021). In 2021, the unprecedented heavy rainfall in Zhengzhou led to 380 fatalities. Flooding occurred in Subway Line 5 and the B-G Expressway North Tunnel, causing significant inundation for numerous vehicles (Yin et al. 2023). In 2023, 88 subway stations in New York City sustained some form of flooding, and 22 stations have been identified as problematic and requiring major upgrades (Siff 2024).

Asia, indisputably, stands as the global epicenter of urbanization (Kundu & Pandey 2020), and the scale of urban underground spaces is rapidly expanding, with a particularly noticeable trend toward complex spatial structures. Concurrently, as a hotspot for climate change, Asia stands out as the most vulnerable continent, with approximately 100 million people residing in flood-prone areas. China and India account for the largest share, collectively representing nearly half of the global population affected by flooding (Devitt et al. 2023). The combined impacts of climate change and urban expansion have already indicated that flood risks in many regions may continue to escalate (Muis et al. 2015). From publicly available literature, it is evident that scholars from Japan have made significant contributions to the research on underground space flooding, and legislative measures pertaining to this issue have been established. As of March 2019, under the guidance of the Ministry of Land, Infrastructure, Transport and Tourism, Japan has incorporated 1,260 underground spaces into municipal or district-level disaster prevention plans. Among these, 890 have completed the formulation of flood prevention and evacuation plans (MLIT 2020).

In this context, there is a pressing need to prioritize research on simulation and assessment of urban underground space inundation during extreme rainfall events.

For the assessment of tools to analyze inundation risk in urban underground spaces, based on published research outcomes, we have summarized representative methods, categorizing them into four types: (1) statistical methods, (2) physical modelling approaches, (3) multi-criteria analysis (MCA), and (4) computational hydraulic methods. The following sections will discuss each of these four methods in detail.

Statistical methods

Statistical methods require long-term data records and operate under the premise that the past is the key to the future, assuming that historical floods can be used to predict future floods (Nott 2006). This method is characterized by its extremely simple calculations. Although the assessment results can to some extent reflect the risks in the study area and have been adopted by standards such as the American Society of Civil Engineers' Flood Resistant Design and Construction (ASCE/SEI 24-14), such approaches may require a significant amount of sensitive data. Moreover, they may face issues of hydrological non-stationarity in the context of climate change (Wasko et al. 2021), studies have already revealed that under the backdrop of climate change, the current once-in-a-century flood events in coastal areas may occur with increased frequency, potentially reaching intervals of 9–15 years by the year 2050 (Boumis et al. 2023). Currently, this method is primarily employed for parameter calibration. For instance, Forero-Ortiz et al. (2020) validated the assessment of inundation risk in the Barcelona subway using two decades of observed inundation data. Thang et al. (2004) validated the accuracy of a numerical model by measuring the flood process.

Physical modelling approaches

Underground space inundation is primarily caused by external water ingress. Due to the diverse pathways and rapid diffusion of flood ingress, coupled with the intricate internal spatial connectivity of underground complexes, the hydraulic characteristics of the system are complex, lacking a unified pattern (Chen et al. 2018); hence, some scholars resort to physical models to investigate the hydraulic characteristics of flood ingress in underground spaces. Liu et al. (2014) conducted a study on the flow patterns of water during urban underground space inundation using a physical model. Shin et al. (2012) employed a scaled physical model to simulate the underground space inundation situation in a specific area of Busan. Gotoh et al. (2006) utilized computational fluid dynamics (CFD) methods to assess pedestrian walking resistance during water ingress in underground spaces. Onishi et al. (2008) conducted a study involving real human subjects to investigate pedestrian evacuation capabilities during water ingress in underground spaces. They proposed the use of specific force per unit width as a criterion for safe evacuation. Later, this method was applied to vulnerable groups such as the elderly (Ishigaki et al. 2010). Shin et al. (2021) argued that although three-dimensional hydrodynamic models can accurately predict the complex inundation processes in underground spaces, constructing a numerical grid or elements throughout the entire space in three dimensions leads to underutilized grids. Therefore, they proposed that a two-dimensional flood modelling approach considering the interconnected structure is a method that balances reflecting the characteristics of flow field changes on each floor and computational efficiency (Shin et al. 2021).

Multi-criteria analysis

MCA is a method for analyzing complex decision problems. Su & Tung (2014) conducted a detailed study on the application of MCA in urban flood vulnerability assessment; the latest MCA methods are often combined with fuzzy analytic hierarchy process and geographic information system (GIS) technology to enhance the accuracy of this tool. Wu et al. (2018) employed a comprehensive Bayesian network framework to rapidly and dynamically assess the evolution processes and consequences of underground space inundation. Han et al. (2019) simultaneously considered flood intensity and evacuation difficulty, utilizing a consequence probability matrix to assess the risk of underground space inundation. Higo et al. (2017) developed the Vitae model to assess strategies for evacuation and rescue during underground space inundation events. Some scholars have also assessed pedestrian evacuation issues during underground flood diffusion based on cellular automaton models (Zheng et al. 2019). It can be observed that the MCA approach has achieved considerable success in assessing the impacts of flooding. However, it should be noted that this method has certain limitations in determining subjective factors, as it heavily relies on expert judgment. Expert knowledge can be siloed, incomplete, variable, and biased due to gender and background, as demonstrated in the case of Puget Sound, USA (O'Connor & Levin 2023).

Computational hydraulic methods

Computational hydraulic methods offer a sophisticated means of assessing flood risks across various scenarios, accounting for spatial and temporal nuances. This approach capitalizes on topographical data alongside insights into urban drainage systems, culminating in a robust quantitative framework that integrates diverse data sources, thereby accommodating a wide spectrum of situations. Augmented by advancements in cutting-edge technologies such as light detection and ranging (LiDAR), GIS, and deep learning, these methods enable precise modelling of urban inundation processes, bolstering our ability to understand and manage flood events.

While research on surface inundation predominantly operates at the watershed scale, focusing on flood analysis, drainage modelling, and the evaluation of flood control infrastructure, it often overlooks the impact on underground spaces. Conversely, although hydraulic software exists that can simulate water flow diffusion in subterranean environments, there remains a scarcity of specialized tools and algorithms tailored for modelling waterlogging in underground settings. Moreover, existing tools can usually only simulate the flooding process in the basement of a single building and are not suitable for simulating underground complexes in larger areas. This disparity results in data silos across different platforms and requires numerous assumptions and estimations. Consequently, a key challenge in current studies lies in effectively integrating surface and underground inundation models.

Early efforts to couple surface and underground inundation models include the employment of the plane binary runoff method for surface runoff calculations used by Toda et al. (1999), coupled with a culvert and reservoir generalized model for the subsurface. Herath & Dutta (2004) utilized a 2D model integrating river, terrain, and pipe networks for surface inundation, while generalizing underground space water ingress using experimentally validated rectangular weir formulas. Toda (2008) discussed surface inundation via a two-dimensional unsteady flow model based on an unstructured grid, representing the underground shopping mall as a combination of three-dimensional reservoirs. Ishigaki et al. (2016) generalized underground space entrances as rectangular weirs, employing a 1D–2D coupling approach to calculate water ingress under heavy rain, flood, storm surge, and tsunami scenarios. Son et al. (2016) compared two underground space generalization methods: boundary method and storage pond method, concluding the boundary method as computationally efficient and robust. Yuan et al. (2023) subsequently termed the boundary method as the ‘hydraulic connectivity method’, utilizing it to assess underground inundation during an extreme rainfall event in Zhengzhou. For Barcelona's metro risk assessment, the subway was simulated using pipes of equivalent dimensions, with inflow points generalized as gratings while ignoring entrance tunnel variability between stations (Forero-Ortiz et al. 2020). While generalization methods like reservoirs and culverts can meet general modelling requirements for overall underground flood risk, further research is needed for objective evacuation risk assessments in these spaces.

Study area and data

Luohu District, situated within the Shenzhen Special Economic Zone in Guangdong, China, exhibits a diverse topography with higher elevations in the northeast gradually descending to lower elevations in the southwest. Its landscape is characterized by hilly terrain intermingled with small alluvial plains. In 2023, an exceptional rainfall event on 7 September in Shenzhen resulted in the inundation of at least 136 residential underground spaces, alongside flooding on multiple metro lines. Some affected stations resumed operations only after a week-long interruption. Luohu bore the brunt of this disaster, experiencing significant inundation. The study area, located in the central city of Luohu, is bounded by rivers to the west and south. Encompassing 695.48 hectares, all fall within the same watershed.

The study utilized elevation data from the FABDEM dataset provided by the University of Bristol (Hawker et al. 2022); this dataset is the first worldwide bare-earth topographic product, which has removed buildings and trees with 30 × 30 m spatial resolution. To ensure the accuracy of the data, elevation data from relevant surveys were used for grid refinement, and local adjustments were made through on-site measurements, achieving a synergistic local-focal-zonal coordination in terms of map algebra (Mennis 2010).

The study utilized recorded rainfall data spanning from 00:00 on 7 September to 24:00 on 8 September. Over this 48-h period, the total precipitation amounted to 436.83 mm, with the highest 3-h rainfall recorded at 142.87 mm between 21:10 on 7 September and 00:05 on 8 September. Given the study area's proximity to the estuary and the consequent susceptibility of its drainage system to tides, the model incorporated monitored tide level data, as depicted in Figure 1. Remarkably, despite the record-breaking rainfall, the tide level remained normal throughout the disaster. Therefore, it can be assumed that this disaster was driven solely by precipitation. Although it can be deemed fortunate for this time, it underscores the potential for greater losses in the future against the backdrop of rising sea levels and coastal subsidence, a challenge already faced by many coastal cities (Ohenhen et al. 2024).
Figure 1

Rainfall and tide data.

Figure 1

Rainfall and tide data.

Close modal

Surface inundation model

For this study, the ICM software in the Autodesk Water suite was employed. The infiltration process utilized the Horton model, and the surface runoff process adopted a nonlinear reservoir model. Hydraulic aspects were achieved by fully solving the Saint-Venant equations, represented by the following formula:
formula
(1)
formula
(2)
where is the flow rate, is the cross-sectional area, h is the local water depth, is the friction slope, is the local channel bottom slope, and is the local lateral net inflow.
Based on existing pipeline data, a 1D–2D coupled surface runoff model was constructed. The final model system includes 423 nodes, 435 pipelines, 414 sub-catchments, and 36,110 mesh grids, as illustrated in Figure 2.
Figure 2

Overview of the study area.

Figure 2

Overview of the study area.

Close modal

The study utilized the SinoLC-1 dataset, a land cover dataset based on artificial intelligence, developed by China University of Geosciences (Li et al. 2022); the spatial resolution is 1 × 1 m. In accordance with local standards, runoff coefficients were set, with no consideration for reduction due to low impact development (LID) during extreme rainfall events. The final runoff coefficient proportions for different land covers are detailed in Table 1.

Table 1

Surface runoff coefficient settings

Land coverRunoff coefficientPercentage of areaLand coverRunoff coefficientPercentage of area (%)
Traffic route 0.90 14.46 Building 0.85 68.3 
Tree cover 0.45 8.81 Barren 0.60 4.23 
Grassland 0.50 0.01 Water 1.00 2.90 
Cropland 0.55 1.23    
Land coverRunoff coefficientPercentage of areaLand coverRunoff coefficientPercentage of area (%)
Traffic route 0.90 14.46 Building 0.85 68.3 
Tree cover 0.45 8.81 Barren 0.60 4.23 
Grassland 0.50 0.01 Water 1.00 2.90 
Cropland 0.55 1.23    

To validate the effectiveness of the model, the measured data from the flowmeter are compared with the simulated data after each model run. The Nash–Sutcliffe efficiency (NSE) coefficient is employed for this purpose. NSE serves as a standard metric for evaluating the predictive accuracy of hydrological models. NSE values can vary from −∞ to 1.0, with a score of 1.0 indicating perfect agreement between model predictions and observed data. Generally, NSE coefficient exceeding 0.75 suggests that the model provides a satisfactory approximation of reality. The specific formula is outlined as follows:
formula
(3)
where is the mean of observed discharges, is the modelled discharge, and is the observed discharge at time t.

Underground inundation model

As previously mentioned, the ICM software lacks a dedicated module for underground spaces. Consequently, we rely on engineering expertise to make necessary assumptions and align our objectives accordingly. Building upon the existing surface hydrological models outlined in the preceding subsection, we propose a two-step methodology to simulate and analyses underground space inundation, as depicted in Figure 3.
Figure 3

Model steps.

In the first step, we augment the existing surface model by conducting on-site surveys to adjust the elevations of roadway segments within the study area, facilitating the entry of rainwater from open sections into the underground space. Subsequently, we introduce ‘storage’ nodes to conceptualize the underground space, with the maximum storage volume equal to actual underground space volume, which represents complete submersion of the underground space. We utilize ‘weir’ elements as flood barriers in the model, setting the width and height to match the flood barrier at the underground space entrance. Although underground spaces typically integrate drainage pumps in practical settings, the current study excludes them due to this disaster event rendering them nonfunctional. To couple those elements with the surface model, we employ ‘break’ nodes, facilitating interaction between water accumulation in the 1D pipe network model and the 2D surface grid. Flow data from the weirs are recorded during subsequent model runs.

In the second step, a separate underground topography model is constructed based on on-site surveys, with further grid refinement. Flow data obtained from weirs in the first step are incorporated into the underground space model as inflow events, enabling pure topographic diffuse flow simulations.

We will investigate underground space inundation by directly adjusting the height of weirs to simulate the installation of flood barriers at various heights. In addition, we define the scenario where the weir height is set to 0 as the most adverse scenario, assuming the absence of any mitigation measures for flooding.

The underground inundation process employed in this study is based on the MULFLOOD model (Alcrudo & Mulet 2005), which utilizes the shallow water equations (SWEs). These equations represent the depth-averaged version of the Navier–Stokes equations and are employed for the mathematical depiction of two-dimensional flow. The SWEs assume predominantly horizontal flow and neglect variations in velocity along the vertical coordinate. Consequently, Manning's n is utilized to model the roughness of underground spaces, while the terrain grid exclusively governs slope and uniformity. Permeability is not considered in this study due to its focus on underground environments.

Underground spaces

This study encompasses a total of four underground spaces, spanning three distinct types: underground parking lots, underpass, and metro station, as detailed in Table 2 and Figure 4. In practice, not all entrances experience water ingress; thus, this study exclusively models those entrances susceptible to flooding impacts.
Table 2

Overview of underground spaces

SiteTypeArea (m2)Max volume (m3)Note
Underground parking lots 3,048 8,000  
Underground parking lots 19,530 49,000 Two floors 
Metro station 12,213 30,000 Two floors 
Underpass 2,843 5,000  
SiteTypeArea (m2)Max volume (m3)Note
Underground parking lots 3,048 8,000  
Underground parking lots 19,530 49,000 Two floors 
Metro station 12,213 30,000 Two floors 
Underpass 2,843 5,000  
Figure 4

Underground spaces.

Figure 4

Underground spaces.

Close modal

Surface inundation risk assessment

As illustrated in Figure 5, the final surface inundation simulation reveals the following outcomes: an area exceeding 0.5 m in water depth spans 144.79 hectares, while an area with depths between 0.3 and 0.5 m encompasses 21.70 hectares. In addition, a region with depths ranging from 0.15 to 0.3 m covers an area of 27.86 hectares. Inundation predominantly occurs in areas bordering the river, coinciding with regions densely with underground spaces as identified in this study. These simulation results underscore the vulnerability of highly urbanized old city areas to extreme rainfall events and emphasize potential risks at the urban-water interface and emphasize potential risks at the waterfront area in cities.
Figure 5

Flood map.

Furthermore, the obtained results demonstrate the satisfactory stability and accuracy of the model, with an overall NSE coefficient reaching as high as 0.87. Moreover, when compared with inundation depth at designated flooding points on the ground, the simulated results exhibit a high level of accuracy, as depicted in Table 3.

Table 3

Inundation depths and simulated results

LocationInundation depths (m)Simulated results (m)Bias (%)
East Gate North Road 0.17 0.18 5.88 
Shennan East Road 1.92 1.84 4.17 
Luofang Interchange 0.72 0.70 2.86 
LocationInundation depths (m)Simulated results (m)Bias (%)
East Gate North Road 0.17 0.18 5.88 
Shennan East Road 1.92 1.84 4.17 
Luofang Interchange 0.72 0.70 2.86 

Underground inundation risk assessment

Scenario without flood barriers

Under typical circumstances, when the inundation depth in flat areas reaches 0.5–0.7 m, both adult females and males may encounter difficulties wading through the water. In addition, the standard height of a car's exhaust port above the ground is typically 0.2–0.3 m. Once the exhaust port is submerged, there is a potential risk of engine stalling and further obstruction of evacuation routes. A survey conducted among merchants in underground malls in Japan revealed that 86% of users would experience fear and initiate evacuation when the inundation depth in the underground space reaches 0.1 m (Ishigaki 2006).

While using inundation depth as a measure to evaluate evacuation risk seems intuitive and straightforward, a single depth measurement may not objectively and scientifically assess the evacuation risk, especially in areas like staircases where pedestrians often face strong water flows, making evacuation challenging. To evaluate such risks, Japanese scholars conducted studies involving real individuals during subsurface flooding events to assess pedestrian evacuation capabilities. They proposed the use of specific force per unit width () as a criterion for safe evacuation (Onishi et al. 2008). This method considers both the impact of water flow and the depth of inundation on evacuation. The difficulties of evacuation for different groups of people are summarized in Table 4.

Table 4

Criteria of safe evacuation

Limit of safe evacuation (m3/m)Difficult without any help (m3/m)
Adult males 0.125 0.250 
Elderly males 0.100 0.200 
Adult females 0.100 0.200 
Elderly females 0.080 0.160 
Limit of safe evacuation (m3/m)Difficult without any help (m3/m)
Adult males 0.125 0.250 
Elderly males 0.100 0.200 
Adult females 0.100 0.200 
Elderly females 0.080 0.160 

In this study, we will adopt the specific force per unit width evaluation method, with the specific calculation formula as follows:
formula
(4)

In the formula, is specific force per unit width, u is the flow velocity of water in the area, h is the depth of inundation, and g is gravity.

The assessment begins with an examination of the inundation risk in each underground area under the most adverse scenario. At Site A, water ingress commences at 22:05, with regions exhibiting values surpassing 0.08 emerging by 23:00, signifying challenges in proactive evacuation. By 23:20, areas with exceeding 0.20 indicate difficulties in rescue evacuation within this vicinity. By 23:55, widespread difficulties in rescue evacuation are observed, leading to disruptions in essential evacuation routes, as depicted in Figure 6.
Figure 6

Underground evacuation difficulty map – Site A.

Figure 6

Underground evacuation difficulty map – Site A.

Close modal
Site B experiences water ingress from 17:38. By 20:03, areas on the underground second floor with values surpassing 0.08 indicate challenges in proactive evacuation. By 23:34, widespread difficulties in rescue evacuation are noted on the underground second floor, with disruptions to essential evacuation routes. Difficulties in proactive evacuation emerge on the underground first floor by 23:23, and by 23:47, widespread difficulties in rescue evacuation are observed on this level, as illustrated in Figure 7.
Figure 7

Underground evacuation difficulty map – Site B.

Figure 7

Underground evacuation difficulty map – Site B.

Close modal
At Site C, water ingress begins at 17:36. By 19:24, difficulties in proactive evacuation are observed on the underground second floor, and by 19:51, widespread challenges in rescue evacuation are noted on this floor, disrupting essential evacuation routes. Difficulties in proactive evacuation on the underground first floor emerge by 22:01, with widespread challenges in rescue evacuation observed by 22:22, as shown in Figure 8.
Figure 8

Underground evacuation difficulty map – Site C.

Figure 8

Underground evacuation difficulty map – Site C.

Close modal
Site D experiences water ingress starting at 17:06. By 17:33, regions with difficulties in proactive evacuation emerge, and by 17:47, widespread challenges in rescue evacuation are observed, as depicted in Figure 9.
Figure 9

Underground evacuation difficulty map – Site D.

Figure 9

Underground evacuation difficulty map – Site D.

Close modal

The model employed in this study effectively demonstrates the flood risks inherent in underground spaces with multiple floors. Evacuation from the lowest level consistently proves to be of the shortest duration, consistent with common experience, as water naturally gravitates toward the lowest. This hypothesis is supported by the fundamental thin-wall weir formula, which suggests that as long as lower spaces remain partially submerged, water accumulation on upper levels remains relatively shallow. This occurs due to the proportional relationship between the flow rate of the weir and the head of water before the weir. Consequently, higher floors may achieve a certain evacuation time before the complete inundation of lower floors.

It is important to note that the simulation does not consider the impact of structural columns, parked vehicles, and stored items in underground spaces on the inundation process. The occupied volumes of these elements could exacerbate water level rise and flow velocity, thereby diminishing the available time for safe evacuation.

Scenarios with flood barrier

Ensuring the safety of life and property in underground spaces necessitates employing various methods, among which the installation of flood barriers or adjustments to entrance elevations are deemed the most effective and cost-efficient approaches. This section is dedicated to scrutinizing the inundation risk of each underground space under different flood barrier heights and evaluating evacuation times. As previously noted, the lowest level of underground spaces consistently proves to be the most vulnerable to inundation. Hence, our primary focus will revolve around discussing the situation from the perspective of the lowest level.

In evaluating evacuation times, this study defines the initiation of evacuation as the moment when groundwater depth exceeds 0.1 m. Proactive evacuation is deemed terminated when areas with specific force per unit width greater than 0.08 emerge, while the conclusion of rescue evacuation is marked by the point when one-third of the area in the region has a specific force per unit width greater than 0.20 or obstructs essential evacuation routes. It is crucial to note that Site D, designated as an underpass, precludes the installation of flood barriers, and hence remains unaddressed in this section. Figure 10 illustrates the water ingress volume in each underground space under varying flood barrier heights.
Figure 10

Water ingress volume under varying flood barrier heights.

Figure 10

Water ingress volume under varying flood barrier heights.

Close modal

In specific terms, for Site A, the absence of flood barriers results in the commencement of underground flooding at 22:05, with water levels reaching 0.1 m within a mere 18 min, marking the begin of evacuation. Both proactive and rescue evacuation times are notably short, standing at 37 and 55 min, respectively, underscoring the susceptibility of underground spaces to inundation. However, the introduction of a flood barrier, standing at a height of 0.3 m, brings about significant improvements. The onset of flooding across the entire underground area is delayed by 3 h, affording ample time for the execution of pertinent emergency protocols. Furthermore, there is a substantial reduction in actual water ingress, leading to considerable extensions in both proactive and rescue evacuation durations. Elevating the height of the water-blocking barrier to 0.6 m ensures the complete prevention of inundation throughout the site, underscoring the efficacy of flood barriers as a straightforward and efficient means of mitigating inundation in underground spaces, as illustrated in Table 5.

Table 5

Evacuation time under different scenarios – Site A

Flood barrier height (m)Begin of ingressBegin of evacuationEnd of proactive evacuationEnd of rescue evacuationProactive evacuation timeRescue evacuation time
0.0 22:05 22:23 23:00 23:55 0 h 37 min 0 h 55 min 
0.3 1:26 3:07 4:19 12:13 1 h 12 min 8 h 54 min 
0.6 NA NA NA NA NA NA 
Flood barrier height (m)Begin of ingressBegin of evacuationEnd of proactive evacuationEnd of rescue evacuationProactive evacuation timeRescue evacuation time
0.0 22:05 22:23 23:00 23:55 0 h 37 min 0 h 55 min 
0.3 1:26 3:07 4:19 12:13 1 h 12 min 8 h 54 min 
0.6 NA NA NA NA NA NA 

For Site B, the absence of flood barriers triggers underground flooding at 17:38, with water depths in the underground second floor reaching 0.1 m by 19:15. Proactive evacuation and rescue times are notably short, at 48 and 31 min, respectively. However, installing a flood barrier standing at a height of 0.3 m delays the onset of flooding throughout the underground area by nearly an hour. Proactive evacuation time experiences a slight extension. Elevating the barrier to 0.6 m further delays both flooding onset and evacuation initiation times, with no significant increase in proactive and rescue evacuation durations. This suggests that emphasizing rescue evacuation time is less meaningful when flood flow remains largely unabated. In the scenario featuring a 1.2-m flood barrier, water inflow diminishes significantly; consequently, evacuation time becomes relatively adequate, as detailed in Table 6.

Table 6

Evacuation time under different scenarios – Site B: underground second floor

Flood barrier height (m)Begin of ingressBegin of evacuationEnd of proactive evacuationEnd of rescue evacuationProactive evacuation timeRescue evacuation time
0.0 17:38 19:15 20:03 20:34 0 h 48 min 0 h 31 min 
0.3 18:33 19:56 20:59 21:35 1 h 03 min 0 h 36 min 
0.6 19:22 20:55 22:01 22:33 1 h 06 min 0 h 32 min 
1.2 22:04 0:34 1:08 2:09 1 h 34 min 1 h 01 min 
Flood barrier height (m)Begin of ingressBegin of evacuationEnd of proactive evacuationEnd of rescue evacuationProactive evacuation timeRescue evacuation time
0.0 17:38 19:15 20:03 20:34 0 h 48 min 0 h 31 min 
0.3 18:33 19:56 20:59 21:35 1 h 03 min 0 h 36 min 
0.6 19:22 20:55 22:01 22:33 1 h 06 min 0 h 32 min 
1.2 22:04 0:34 1:08 2:09 1 h 34 min 1 h 01 min 

At Site C, in the absence of flood barriers, underground flooding begins at 17:36, prompting evacuation for the underground second floor by 18:10, culminating in an overall evacuation duration of approximately 18:40. As flood barrier height increases incrementally, both flooding onset and evacuation initiation times demonstrate improvements for both the underground second and first floors. Nonetheless, similarly, in cases where there is no notable reduction in flooding, there appears to be no significant correlation between rescue evacuation time and flood barrier height, particularly concerning the underground second floor, as illustrated in Table 7.

Table 7

Evacuation time under different scenarios – Site C: underground second floor

Flood barrier height (m)Begin of ingressBegin of evacuationEnd of proactive evacuationEnd of rescue evacuationProactive evacuation timeRescue evacuation time
0.0 17:36 18:10 19:24 19:51 1 h 14 min 0 h 27 min 
0.3 18:18 19:44 20:35 21:04 0 h 51 min 0 h 29 min 
0.6 19:35 20:54 21:46 22:12 0 h 52 min 0 h 26 min 
1.2 22:25 23:53 0:57 1:52 1 h 04 min 0 h 55 min 
Flood barrier height (m)Begin of ingressBegin of evacuationEnd of proactive evacuationEnd of rescue evacuationProactive evacuation timeRescue evacuation time
0.0 17:36 18:10 19:24 19:51 1 h 14 min 0 h 27 min 
0.3 18:18 19:44 20:35 21:04 0 h 51 min 0 h 29 min 
0.6 19:35 20:54 21:46 22:12 0 h 52 min 0 h 26 min 
1.2 22:25 23:53 0:57 1:52 1 h 04 min 0 h 55 min 

Based on the simulations outlined in the preceding sections, we have conducted an analysis of underground space inundation scenarios in various underground spaces under differing flood barrier heights and have formulated the corresponding recommendations:

For Site A, we recommend the preparation of flood barrier modules measuring 0.6 m in height for direct installation during rainfall events.

Regarding Sites B and C, due to their inferior locations, even the installation of 1.2 m flood barriers may not entirely resolve the issue (higher flood barriers are conceivable but not practically feasible for implementation). Therefore, we propose adjusting the elevations of these sites in the long term. In the short term, we suggest prioritizing the development of evacuation plans and the relocation of high-value assets from the underground spaces, building upon the foundation of 1.2-m flood barriers.

As for Site D, given its nature as a roadway, flood barriers cannot be feasibly installed. Consequently, we urge the transportation department to enhance signage indicating flood risks and to install online water level monitors. This measure will facilitate the immediate cessation of traffic upon the onset of flooding.

This study examines the inundation process of various underground spaces in a specific area of Shenzhen, China, during an extreme rainfall event utilizing the ICM software. It proposes a two-step method to simulate the coupled surface-underground inundation process, which is relatively straightforward, computationally efficient, and exhibits high accuracy. To assess the inundation risk of underground spaces, the study employs the specific force per unit width evaluation method and discusses the influence of different flood barrier heights on underground inundation and evacuation times.

Hydraulically, the study identifies that in multilevel underground spaces facing floods, the lowest level consistently proves the most vulnerable. Before the lower spaces become entirely submerged, the water level in upper spaces could remain at a lower level for a certain period, thus facilitating evacuation. This suggests that in extreme scenarios, sacrificing the lowest level to evacuate higher floors may be feasible, although it does not advocate for the construction of large-scale underground storage facilities, as their actual effectiveness may be debatable.

Furthermore, the study demonstrates that flood barriers can significantly delay the inundation of underground spaces, serving as a simple and cost-effective flood prevention measure. However, for areas with inferior locations, the effectiveness of prolonging evacuation times is limited unless flood volumes are substantially reduced by the barriers. Therefore, emphasis should be placed on developing evacuation plans in such instances.

Moreover, the research illuminates the inundation risk of densely developed urban underground spaces within the context of climate change. It underscores the urgent necessity for further research and attention from both the academic community and water management authorities on this matter.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.

Alcrudo
F.
&
Mulet
J. J. F.
2005
Urban inundation models based on the shallow water equations. Numerical and practical issues
. In: Benkhaldoun, F., Ouazar, D. & Raghay, S. (eds) Proceedings of Finite Volumes for Complex Applications IV. Problems and Perspectives.
Hermes Science Publishing
3
12
.
Arkell
B. P.
&
Darch
G.
2006
Impact of climate change on London's transport network
. In:
Proceedings of The Institution of Civil Engineers-Municipal Engineer
159 (4), 231–237. doi: https://doi.org/10.1680/muen.2006.159.4.231
Boumis
G.
,
Moftakhari
H. R.
&
Moradkhani
H.
2023
Coevolution of extreme sea levels and sea-level rise under global warming
.
Earth's Future
11
,
e2023EF003649
.
Chen
F.
,
Liu
S.-G.
&
Liu
W.-W.
2018
Causes and characteristics of flooding in urban underground space
.
Journal of Yangtze River Scientific Research Institute
35
,
38
.
Compton
K. L.
,
Faber
R.
,
Ermolieva
T. Y.
,
Linnerooth-Bayer
J.
&
Nachtnebel
H.-P.
2009
Uncertainty and Disaster Risk Management: Modeling the Flash Flood Risk to Vienna and its Subway System. IASA Research Report. IIASA, Laxenburg, Austria, RR-09-002
.
Devitt
L.
,
Neal
J.
,
Coxon
G.
,
Savage
J.
&
Wagener
T.
2023
Flood hazard potential reveals global floodplain settlement patterns
.
Nature Communications
14
,
2801
.
Forero-Ortiz
E.
,
Martínez-Gomariz
E.
,
Cañas Porcuna
M.
,
Locatelli
L.
&
Russo
B.
2020
Flood risk assessment in an underground railway system under the impact of climate change – A case study of the Barcelona Metro
.
Sustainability
12
,
5291
.
Gotoh
H.
,
Ikari
H.
,
Sakai
T.
&
Tanioka
H.
2006
Computational mechanics of a force on a human leg in flow over a underground staircase in urban flood
.
Proceedings of Hydraulic Engineering
50
,
865
870
.
Han
Y.-S.
,
Shin
E. T.
,
Eum
T. S.
&
Song
C. G.
2019
Inundation risk assessment of underground space using consequence-probability matrix
.
Applied Sciences
9
,
1196
.
Hawker
L.
,
Uhe
P.
,
Paulo
L.
,
Sosa
J.
,
Savage
J.
,
Sampson
C.
&
Neal
J.
2022
A 30 m global map of elevation with forests and buildings removed
.
Environmental Research Letters
17
,
024016
.
He
R.
,
Tiong
R. L.
,
Yuan
Y.
,
Zhang
L. J. T.
&
Technology
U. S.
2024
Enhancing resilience of urban underground space under floods: Current status and future directions
.
Tunnelling and Underground Space Technology
147
,
105674
.
Herath
S.
&
Dutta
D.
2004
Modeling of urban flooding including underground space
. In
Proceedings of the Second International Conference of Asia-Pacific Hydrology and Water Resources Association
,
Tokyo, Japan
, pp.
55
63
.
Higo
E.
,
Okada
N.
,
Hipel
K. W.
&
Fang
L.
2017
Cooperative survival principles for underground flooding: Vitae system based multi-agent simulation
.
Expert Systems with Applications
83
,
379
395
.
Ishigaki
T.
2006
Difficulty of evacuation from underground space in urban flood
. In Kundu, D., Sietchiping, R. & Kinyanjui, M. (eds)
Proceedings of 7th International Conference on Hydroinformatics, HIC 2006
,
France
, pp.
614
620
.
Ishigaki
T.
,
Asai
Y.
,
Nakahata
Y.
,
Shimada
H.
,
Baba
Y.
&
Toda
K.
2010
Evacuation of aged persons from inundated underground space
.
Water Science and Technology
62
,
1807
1812
.
Ishigaki
T.
,
Kawanaka
R.
,
Ozaki
T.
&
Toda
K.
2016
Vulnerability to underground inundation and evacuation in densely urbanized area
.
Journal of Disaster Research
11
,
298
305
.
Kundu
D.
&
Pandey
A. K.
2020
World urbanisation: Trends and patterns
. In:
Developing National Urban Policies: Ways Forward to Green and Smart Cities
(Kundu, D., Sietchiping, R. & Kinyanjui, M. (eds)). Springer, Singapore, pp.
13
49
. https://doi.org/10.1007/978-981-15-3738-7_2.
Li
Z.
,
Zhang
H.
,
Lu
F.
,
Xue
R.
,
Yang
G.
&
Zhang
L.
2022
Breaking the resolution barrier: A low-to-high network for large-scale high-resolution land-cover mapping using low-resolution labels
.
ISPRS Journal of Photogrammetry and Remote Sensing
192
,
244
267
.
Liu
S. G.
,
Chen
F.
&
Zhong
G. H.
2014
Urban Underground Space Flood Control and Security
.
Tongji University Press
,
Shanghai, China
.
MLIT
2020
Guidelines for Safe Evacuation Measures for Underground Malls (revised in March 2020)
.
Ministry of Land, Infrastructure, Transport and Tourism
,
Tokyo
, Japan. www.mlit.go.jp/toshi/toshi_gairo_tk_000062.html.
Muis
S.
,
Güneralp
B.
,
Jongman
B.
,
Aerts
J. C.
&
Ward
P. J.
2015
Flood risk and adaptation strategies under climate change and urban expansion: A probabilistic analysis using global data
.
Science of the Total Environment
538
,
445
457
.
Nott
J.
2006
Extreme Events: A Physical Reconstruction and Risk Assessment
.
Cambridge University Press
, Cambridge, UK, pp. 1–310. https://doi.org/10.1017/CBO9780511606625.
Ohenhen
L. O.
,
Shirzaei
M.
,
Ojha
C.
,
Sherpa
S. F.
&
Nicholls
R. J.
2024
Disappearing cities on US coasts
.
Nature
627
,
108
115
.
Onishi
Y.
,
Ishigaki
T.
,
Baba
Y.
&
Toda
K.
2008
Criterion and its application for safety evacuation during underground flooding
.
Proceedings of Hydraulic Engineering
52
,
841
846
.
Shin
D. S.
,
Park
J. B.
,
Shon
T. S.
&
Jo
D. J.
2012
Preventive measures assessment to prevent flooding complex underground space through hydraulic model experiment
.
Journal of the Korean Society of Hazard Mitigation
12
,
59
66
.
Shin
E.
,
Kim
H.-J.
,
Rhee
D. S.
,
Eom
T.
&
Song
C. G.
2021
Spatiotemporal flood risk assessment of underground space considering flood intensity and escape route
.
Natural Hazards
109
,
1539
1555
.
Siff
A.
2024
200 NYC Subway Stations Have Flooded in Recent Storms, 22 of Which Need Major Fixes: MTA. NBC
. www.nbcnewyork.com/news/local/200-nyc-subway-stations-have-flooded-in-recent-storms-22-of-which-need-major-fixes-mta/5093801
Sterling
R.
&
Nelson
P.
2013
City resiliency and underground space use
.
Advances in Underground Space Development
7,
43
55
.
Su
H.-T.
&
Tung
Y.-K.
2014
Multi-criteria decision making under uncertainty for flood mitigation
.
Stochastic Environmental Research and Risk Assessment
28
,
1657
1670
.
Thang
N. T.
,
Inoue
K.
,
Toda
K.
&
Kawaike
K.
2004
A model for flood inundation analysis in urban area: Verification and application
.
DPRI Annuals B
47
,
303
315
.
Toda
K.
2008
Urban flood analysis considering underground space
.
Proceedings of Hydraulic Engineering
52
,
K7
K8
.
Toda
K.
,
Inoue
K.
,
Maeda
O.
&
Tanino
T.
1999
Analysis of overland flood flow intrusion into underground space in urban area
.
Proceedings of Hydraulic Engineering
43
,
539
544
.
Tso
Y.-E.
&
McEntire
D. A.
2011
Emergency Management in Taiwan: Learning From Past and Current Experiences. Organizations & Initiatives From Around the World, Federal Emergency Management Agency
.
Wasko
C.
,
Westra
S.
,
Nathan
R.
,
Orr
H. G.
,
Villarini
G.
,
Villalobos Herrera
R.
&
Fowler
H. J.
2021
Incorporating climate change in flood estimation guidance
.
Philosophical Transactions of the Royal Society A
379
,
20190548
.
Yin
L.
,
Ping
F.
,
Mao
J.
&
Jin
S.
2023
Analysis on precipitation efficiency of the ‘21.7’ Henan extremely heavy rainfall event
.
Advances in Atmospheric Sciences
40
,
374
392
.
Yuan
G.
,
Luyao
W.
,
Nengzhi
C.
&
Juxiang
J.
2023
Simulation of the flood process in urban surface-underground space under extreme rainfall
.
Advances in Water Science
34
,
209
217
.
Zheng
Y.
,
Li
X.-G.
,
Jia
B.
&
Jiang
R.
2019
Simulation of pedestrians' evacuation dynamics with underground flood spreading based on cellular automaton
.
Simulation Modelling Practice and Theory
94
,
149
161
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).