Urban Flood Resilience in a Megacity Context: Multidimensional Assessment and Spatial Differentiation in Shenzhen

Abstract

This study assesses urban flood resilience at the subdistrict scale in Shenzhen, China, addressing the lack of fine-grained spatial analysis in existing city-level models. A multidimensional framework integrating natural geography, infrastructure, socioeconomics, emergency management, and risk exposure was constructed, with indicator weights derived from a hybrid Analytic Hierarchy Process–Entropy Weight Method. Spatial autocorrelation analysis (Moran’s I = 0.475, p < 0.001) revealed distinct “resilience fault lines,” with high-resilience clusters in central districts and low-resilience clusters in peripheral industrial belts. Geodetector identified economic intensity (q = 0.46), elevation (q = 0.39), and emergency shelter density (q = 0.37) as dominant drivers, with strong interaction effects. These findings highlight significant resilience inequality, emphasizing the need for targeted, multidimensional interventions to enhance adaptive capacity and inform climate adaptation strategies in rapidly urbanizing coastal megacities.

1. Introduction

In recent years, influenced by global warming and the rapid development of urbanization, extreme climate events have occurred frequently [1,2,3]. Climate disasters such as heavy rain and floods have already had a significant impact on public safety in cities. Data from the “2024 China Flood and Drought Defense Bulletin” shows that, from 2014 to 2023, an average of 63.4028 million people were affected by floods each year in China, with an average of 445 deaths and disappearances due to floods annually. The area of crops affected by floods reached 59.7961 million hectares, and the average direct economic loss caused by floods was CNY 214.1 billion (USD 29.8 billion) per year. To reduce the losses caused by floods in cities, research on urban flood resilience has increasingly become a topic of concern in the academic community.

The concept of “resilience” was first introduced by American ecologist Holling, who defined ecological resilience as the capacity of an ecosystem to absorb disturbances, adjust, and return to a balanced state after a disaster [4]. Subsequently, researchers began to apply this concept to the fields of disaster management and urban studies. Timmerman was the first to incorporate the idea of resilience into disaster research, defining it as the ability of human communities to withstand external shocks or disruptions to their infrastructure and to recover from them [5]. Walker et al. further described disaster resilience as a system’s ability to absorb the energy of disturbances, undergo structural reorganization, and maintain its core functions and infrastructure without experiencing disruptive transformations during and after the impact of a disaster [6]. In the context of resilient cities, Huang Hong et al. defined them as cities capable of effectively responding to internal and external shocks and stresses across economic, social, technological, and infrastructural systems. These cities can maintain basic functions, structures, and systems in the face of major emergencies and are able to quickly recover, adapt, and pursue sustainable development [7].

At present, a growing body of research has explored urban resilience to flooding. From various perspectives, scholars have developed indicator systems to quantitatively assess cities’ capacity to cope with flood disasters. These systems often incorporate multidimensional indicators across environmental, economic, social, and infrastructural domains. Based on the BRIC model, Burton constructed a flood resilience assessment framework for storm and flood disasters, analyzing resilience across social, economic, institutional, infrastructural, community, and environmental dimensions [8]. Kotzee et al., using data from South Africa, selected 24 resilience indicators covering social, ecological, infrastructural, and economic dimensions to evaluate spatial differences in urban flood resilience [9]. Tayyab et al. developed a GIS-based urban flood resilience model to assess the flood sensitivity and coping capacity of Peshawar, Pakistan, through the lenses of urban flood hazards, exposure, vulnerability, and adaptive capacity [10]. Chen et al. screened indicators based on technological, organizational, social, and economic aspects and used grey relational analysis to evaluate urban flood resilience in Chongqing as a case study [11]. Aroca-Jiménez et al. constructed a flood resilience evaluation system from six dimensions—social, economic, environmental, physical, institutional, and cultural—and analyzed urban flood resilience characteristics in flood-prone areas [12]. Lyu et al. examined underground spaces, focusing on the risks and resilience of infrastructure such as subways during flood events [13,14,15,16].

A wide range of methodologies have also been applied in the study of urban flood resilience. Liu et al. developed an urban-scale flood resilience indicator system based on the “Pressure-State-Response” (PSR) model and determined indicator weights using a combination of the Analytic Hierarchy Process (AHP) and the Entropy Weight Method (EWM) [17]. Wang et al. integrated structural and non-structural measures with the Internet of Things (IoT) and machine learning technologies to enhance resilience in flood prevention and disaster mitigation [18]. Gao et al. introduced hydrodynamic-inundation modeling, combining simulated flood scenarios using the MIKE model with socioeconomic and natural indicators to construct a comprehensive flood resilience index for Nanjing [19]. Xu et al. measured flood resilience levels in three Chinese cities by integrating D-numbers theory, AHP, and the VIKOR method [20]. Zhu et al. combined multiple objective Multi-Criteria Decision-Making (MCDM) weighting methods with stochastic acceptability analysis to generate a comprehensive flood resilience index, enabling cross-city comparability of resilience among 41 cities along the Yangtze River [21].

Cities, especially megacities, are inherently complex systems. However, most existing studies focus primarily on assessing flood resilience at the city-wide level, with relatively limited attention given to the internal spatial heterogeneity of urban resilience [22]. Current evaluations tend to rely on traditional domains, such as natural, economic, and social factors, while key indicators like urban blue–green infrastructure and emergency response capacity have received comparatively less attention in understanding their impact on flood resilience. In particular, urban blue–green infrastructure and emergency response capacity play a critical role in shaping flood resilience but have often been underrepresented in previous assessments. Blue–green infrastructure, such as wetlands, green roofs, permeable pavements, and urban parks, enhances infiltration, regulates runoff, and delays flood peaks, thereby providing both ecological and hydrological benefits that grey infrastructure alone cannot achieve. It not only contributes to reducing pluvial flooding risks but also improves urban livability and climate adaptation capacity [23]. Meanwhile, emergency response capacity directly determines the speed and effectiveness of disaster relief. A lack of adequate emergency shelters, hospitals, or coordinated rescue forces can significantly amplify the social and economic losses caused by floods, even in areas with strong physical defenses [24]. Therefore, incorporating these two dimensions is essential for developing a comprehensive and realistic framework of urban flood resilience.

Based on the above analysis, this study selects 74 subdistricts of Shenzhen as the research area and constructs a flood resilience assessment system using a combination of the Analytic Hierarchy Process (AHP) and the Entropy Weight Method (EWM), incorporating key factors affecting urban flood resilience. This study analyzes the flood resilience characteristics of different subdistricts within Shenzhen. On this basis, spatial autocorrelation analysis is employed to explore the spatial differentiation of flood resilience across the city, while the Geodetector model is used to identify the contribution of various influencing factors. This study aims to reveal the phenomenon of “resilience inequality” within the city and provide practical recommendations for enhancing urban flood resilience. Figure 1 shows the overall workflow of the research.

2. Materials and Methods

2.1. Overview of the Study Area

Shenzhen City (22°27′ N–22°52′ N, 113°46′ E–114°37′ E) is located in the southern part of Guangdong Province on the eastern shore of the Pearl River Estuary, adjacent to the Hong Kong Special Administrative Region. Following the establishment of the Special Economic Zone and the advancement of reform and opening-up policies, Shenzhen rapidly transformed from a small fishing village to a global megacity. By the end of 2024, the total area of the city reached 1997.47 square kilometers, comprising 9 administrative districts and 1 new area (excluding the Shenshan Special Cooperation Zone), with a total of 74 subdistricts and a permanent population of 17.99 million.

Shenzhen features complex terrain and diverse landforms, forming a composite geomorphic zone dominated by hills and characterized by a combination of low mountains, hills, terraces, stepped landforms, and plains. The city has a subtropical monsoon climate with distinct wet and dry seasons and abundant rainfall. The average annual precipitation is 1932.9 mm, with 86% of the total rainfall occurring during the flood season (April to September) under the influence of weather systems such as frontal troughs, tropical cyclones, and monsoonal cloud clusters. These conditions make Shenzhen particularly prone to flood disasters. The rapid pace of urbanization has led to a significant increase in impervious surfaces and high-density concentrations of population and assets, further exacerbating the risk of urban flooding. The extreme torrential rainfall event in September 2023 had a profound impact on Shenzhen, causing severe economic and social losses. Therefore, studying urban flood resilience at the subdistrict level can help identify current challenges in flood prevention and control and provide effective support for enhancing Shenzhen’s disaster resilience capacity. Figure 2 shows an overview of Shenzhen.

2.2. Data Sources

The data used in this study to evaluate urban flood resilience in Shenzhen covers all 74 subdistricts across the city’s 9 administrative districts and 1 functional area. This study integrates multi-source datasets to comprehensively assess urban flood resilience. The data and sources are shown in

Table 1. Urban flood resilience data categories and resources.

Data Category	Specific Criterium	Resource	Time	Resolution
DEM	Elevation (m)	Copernicus	2015	30 m
	Slope (°)
River	Water Network Density (km/km2)	OpenStreetMap	2022	-
Road	Road Network Density (km/km2)	OpenStreetMap	2022	-
Land use	Land use	AIEC	2022	10 m
NDVI	NDVI	NASA	2020	0.05°
Population	Population Density	National Population Census	2020	-
	Proportion of population by age group			-
POI	POI	Baidu and Shenzhen Municipal Government Data Open Platform	2022	-
GDP	GDP (yuan/km2)	IGSNRR, CAS	2020	1 km
Historical flood and precipitation	Historical flood and precipitation data	Shenzhen Meteorological Bureau	2023	-

. Topographic data, including elevation and slope, were obtained from Copernicus. Hydrological features, represented by water network density, and transportation infrastructure, represented by road network density, were extracted from OpenStreetMap. Land use data were sourced from AIEC, while vegetation coverage was derived from NASA. Demographic information, including population density and the proportion of population by age group, was collected from the National Population Census. Point of Interest (POI) data, such as the locations of emergency hospitals, are sourced from Baidu and the Shenzhen Municipal Government Data Open Platform. Economic data, represented by GDP density, were provided by IGSNRR, CAS. Historical flood and precipitation records were acquired from the Shenzhen Meteorological Bureau. These datasets collectively support a multidimensional analysis of spatial patterns and influencing factors of flood resilience at the street scale in Shenzhen.

To provide a clearer understanding of the spatial heterogeneity of the underlying data, representative maps of key input parameters used in the flood resilience assessment are presented in Figure 3.

2.3. Construction of the Indicator System

Urban flood resilience is a complex concept, and there has been extensive research on how to measure and evaluate it through appropriate indicator systems. This study draws upon the “Resistance–Restoration–Adaptability” three-dimensional theoretical framework of urban flood resilience proposed by Serre et al. [25]. Based on this framework, an evaluation indicator system for flood resilience at the subdistrict level is constructed from five dimensions: natural geography, infrastructure, socioeconomic factors, emergency management, and risk exposure. The specific indicators are shown in

Table 2. Urban flood resilience indicator system.

Dimension	Criterion Layer	Indicator Layer	Indicator Description
Resistance	Natural Geography (A)	Terrain Slope (A1)	Average ground slope
		Elevation (A2)	Average elevation above sea level
		Rainstorm Intensity (A3)	1 h rainfall during a 10-year return period storm
		Water Network Density (A4)	Length of rivers per square kilometer
	Infrastructure (B)	Road Network Density (B1)	Length of roads per square kilometer
		Impervious Surface Ratio (B2)	Proportion of impermeable surface area
		Blue–Green Space Ratio (B3)	Proportion of lakes, parks, and green spaces capable of water retention
Restoration	Socioeconomic (C)	Population Density (C1)	Permanent resident population per square kilometer
		Economic Intensity (C2)	GDP per square kilometer
		Population Vulnerability (C3)	Proportion of population under age 14 years and over age 65 years
		Vulnerable Group Facilities (C4)	Number of primary schools and nursing homes per square kilometer
Adaptability	Emergency Management (D)	Emergency Shelter Density (D1)	Shelter area per capita
		Medical Emergency Coverage (D2)	Coverage area within 3 km of Level I/II emergency hospitals
		Fire Rescue Response (D3)	Distance to the nearest special-duty fire station
		Emergency Supplies Access (D4)	Distance to the nearest comprehensive emergency supply warehouse
	Risk Exposure (E)	Historical Waterlogging Density (E1)	Number of historical flooding points
		Critical Rainfall Threshold (E2)	Hourly rainfall threshold that triggers water accumulation

.

(1) Resistance: This primarily refers to the system’s ability to withstand disaster impacts and to prevent or mitigate damages. At the level of physical geography, the city’s slope, elevation [26], storm intensity [27], and water network density [28] reflect the extent of natural flood risk and the inherent resistance capacity to such hazards. Regarding infrastructure, road network density functions both as a route for disaster response and population evacuation [29] and is also closely related to the density of drainage pipelines [30]. The hardened surface coverage rate and the proportion of blue–green space affect flood resistance by influencing surface runoff during flooding events [31,32].

(2) Restoration: This refers mainly to the speed and efficiency with which a system can return to its basic functional state after a disaster. Socioeconomic indicators effectively reflect this capacity. Whether it is population density [33], the presence of vulnerable groups [34], or economic intensity [35], all indicate how much external disturbance the area can endure and thus influence its resilience performance.

(3) Adaptability: This includes key indicators of emergency management resilience, which directly determine the capacity for rapid evacuation, medical response, containment of secondary disasters, and maintenance of basic living conditions following a flood [36,37]. In terms of risk exposure, this dimension reflects the extent to which a street is exposed to hazards induced by flood disasters. It is represented by the density of historical waterlogging points and the critical rainfall threshold. The former reflects the frequency and spatial distribution of past flood events, while the latter quantifies the rainfall level that may trigger water accumulation [38]. Together, these indicators capture both the long-term exposure of the area to flood disasters and its acute potential hazards.

2.4. Weight Determination and Model Construction

The choice of the hybrid Analytic Hierarchy Process (AHP) and Entropy Weight Method (EWM) in this study is motivated not only by their wide application in previous resilience research but also by the specific needs of the Shenzhen case. The Entropy Weight Method (EWM), which calculates weights based on the degree of data dispersion, is relatively objective. In contrast, the Analytic Hierarchy Process (AHP), which relies on expert judgment, effectively integrates domain knowledge and reflects policy orientation and management priorities in resilience building.

The assessment framework involves multiple dimensions—natural geography, infrastructure, socioeconomic conditions, emergency management, and risk exposure—with seventeen heterogeneous indicators. Relying solely on subjective expert judgment (AHP) may introduce bias, while using only objective statistical dispersion (EWM) risks overlooking policy priorities and local governance experience. The hybrid approach ensures complementarity: AHP incorporates domain expertise and reflects management emphasis, whereas EWM captures intrinsic variability and reduces subjectivity. This combination is particularly important at the subdistrict scale of Shenzhen, where fine-grained differences require both contextual expert knowledge and objective data-driven weighting to improve robustness and interpretability of the results. The composite weight of the i-th indicator is calculated as follows:

$$w_i=\mu w_E+(1-\mu )w_A$$

(1)

where $w_i$ represents the composite weight of the i-th indicator and $w_E$ and $w_A$ denote the weights calculated via the entropy weight method and the analytic hierarchy process, respectively. $\mu$ is the combination coefficient, which is assigned a value of 0.5 in this study, indicating that equal importance is given to both subjective and objective methods.

The urban flood resilience of each street is calculated as follows:

$$F_j=\sum _{i=1}^N{w}_i{y}_{ij}$$

(2)

where $F_j$ denotes the urban flood resilience score of street j, $w_i$ is the weight of the i-th indicator, and $y_{ij}$ represents the normalized value of the i-th indicator for street j.

2.4.1. Entropy Weight Method

(1) Data Normalization

Since the original data are unordered and dimensionally inconsistent, this study adopts the range method to normalize the data and eliminate dimensional differences.

For positive indicators,

$$y_{ij}={\displaystyle \frac{x_{ij}-x_{min}}{x_{max}-x_{min}}}$$

(3)

For negative indicators,

$$y_{ij}={\displaystyle \frac{x_{max}-x_{ij}}{x_{max}-x_{min}}}$$

(4)

Here, $y_{ij}$ is the normalized value of the j-th indicator in region i, $x_{ij}$ is the original value, and $x_{max}$ and $x_{min}$ denote the maximum and minimum values of the j-th indicator across all regions, respectively.

(2) Information Entropy Calculation

$$p_{ij}={\displaystyle \frac{y_{ij}}{\sum _{i=1}^m{y}_{ij}}}$$

(5)

$$e_j=-{\displaystyle \frac{1}{\ln m}}\sum _{i=1}^m{p}_{ij}\mathrm{l}\mathrm{n}(p_{ij})(0\le e_j\le 1)$$

(6)

where $p_{ij}$ represents the proportion of the j-th indicator’s value in region i and $e_j$ denotes the information entropy of the j-th indicator.

(3) Weight Determination

$${\mathrm{w}}_j={\displaystyle \frac{1-e_j}{\sum _{j=1}^{n}1-e_j}}$$

(7)

where ${\mathrm{w}}_j$ represents the final weight of the j-th indicator in the entropy weight method.

2.4.2. Analytic Hierarchy Process

(1) Hierarchical Structure Construction

A hierarchical structure is established based on three layers:

▪

Goal layer: Urban flood resilience of city streets;

▪

Criterion layer: Natural geography/Infrastructure/Socioeconomic/Emergency management/Risk exposure;

▪

Indicator layer: 17 specific indicators.

(2) Expert Judgment Matrix

Eighteen experts in hydrology, urban planning, emergency management, and related fields were invited to conduct pairwise comparisons of indicators at each level using Saaty’s 1–9 scale, forming a judgment matrix:

$$A=[\begin{array}{ccc}{a}_{11}& \dots & a_{1m}\\ ⋮& \ddots & ⋮\\ a_{m1}& \dots & a_{mm}\end{array}], a_{ij}={\displaystyle \frac{1}{a_{ji}}}, a_{ii}=1$$

(8)

where $a_{ij}$ represents the relative importance of the i-th indicator compared to the j-th.

(3) Weight Calculation and Consistency Test

The eigenvector method is applied to derive weights $w_j$ using the following:

$$Aw={\lambda }_{max}w$$

(9)

$$w_j={\displaystyle \frac{w_j}{\sum _{j=1}^m{w}_j}}$$

(10)

The validity of the judgment matrix is tested using the consistency ratio (CR), calculated as follows:

$$CR={\displaystyle \frac{CI}{RI}}$$

(11)

where CR is the consistency ratio, CI the consistency index, and RI the random index. If CR < 0.1, the matrix is considered to pass the consistency test; otherwise, it must be revised.

2.5. Spatial Autocorrelation Analysis

To investigate the spatial patterns of flood resilience across Shenzhen, both Global Moran’s I and Local Indicators of Spatial Association (LISA) analyses were conducted. The input parameter for these analyses was the composite flood resilience score of each of the 74 subdistricts, calculated from the weighted aggregation of the five resilience dimensions.

The spatial relationships among subdistricts were represented by a Queen contiguity-based spatial weight matrix, in which $w_{ij}$ = 1 if two subdistricts share a common boundary or vertex and $w_{ij}$ = 0 otherwise. This approach ensures that all spatially adjacent units are considered neighbors. All subdistricts were included in the computation, and the same dataset and spatial weight matrix were used for both the global and local spatial autocorrelation analyses.

Global Moran’s I was employed to measure the overall degree of spatial autocorrelation in flood resilience, where positive values indicate clustering of similar resilience levels and negative values indicate spatial dispersion. The index was calculated using the following formula:

$$I={\displaystyle \frac{n{\sum }_{i=1}^n{\sum }_{j=1}^n{w}_{ij}(x_i-\overline{x})(x_j-\overline{x})}{S_0{\sum }_{I=1}^N{(X_i-\overline{x})}^2}}$$

(12)

where

• I: Global Moran’s I;
• n: Number of spatial units;
• $w_{ij}$: Spatial weight matrix defining adjacency between unit i and j;
• $x_i,x_j$: Flood resilience values of units i and j;
• $\overline{x}$: Mean flood resilience;
• $S_0$: Sum of spatial weights.

The LISA analysis was then applied to the same resilience score dataset and spatial weight matrix to identify localized clusters and outliers. LISA provides more refined analysis of clustering or dispersion at a local scale. It identifies hot spots (high-resilience clusters) and cold spots (low-resilience clusters), effectively uncovering spatial heterogeneity in flood response capacity.

2.6. Geodetector Analysis

Geodetector, proposed by Wang Jinfeng [39], is a spatial analysis method based on the principle that spatial heterogeneity of a geographical phenomenon can reveal its driving forces. It quantifies the explanatory power of influencing factors on the spatial variation in a dependent variable and assesses the interactions between factors. Widely applied in environmental science, public health, and socioeconomics, it is used here to evaluate the explanatory power of secondary indicators on urban flood resilience. The formula is as follows:

$$q=1-{\displaystyle \frac{1}{N{\sigma }^2}}\sum _{i=1}^L{N}_i{\sigma }_i^2$$

(13)

where

• $N{,\sigma }^2$: Total sample size and variance of the whole region;
• $N_i,{\sigma }_i^2$: Sample size and variance of subregion i;
• $L$: Number of subregions;
• $q$: Explanatory power of the factor (ranges from 0 to 1).

A higher $q$ value indicates a stronger explanatory power of the corresponding indicator on flood resilience.

3. Results

3.1. Urban Flood Resilience System

The weights presented in

Table 3. Final weights of urban flood resilience assessment indicators.

Criterion Layer	Weight	Indicator Layer	Weight
Natural Geography A	0.209167495	Topographic Slope A1	0.050502717
		Street Elevation A2	0.03053217
		Rainstorm Intensity A3	0.084135111
		Water Network Density A4	0.043997497
Infrastructure B	0.295713673	Road Network Density B1	0.059837886
		Impervious Surface Ratio B2	0.062261042
		Blue–Green Space Ratio B3	0.173614744
Socioeconomic C	0.139050615	Population Aggregation C1	0.02024131
		Economic IntensityC2	0.07843128
		Population Vulnerability C3	0.026645561
		Concentration of Vulnerable Groups C4	0.013732465
Emergency Management D	0.263413287	Emergency Shelter Density D1	0.145614305
		Medical Emergency Response Coverage D2	0.049594723
		Fire Rescue Response D3	0.033073476
		Emergency Supplies Support D4	0.035130783
Risk Exposure E	0.09265493	Historical Waterlogging Point Density E1	0.040145513
		Critical Rainfall Threshold E2	0.052509416

were determined using a hybrid weighting approach that integrates the Analytic Hierarchy Process (AHP) and the Entropy Weight Method (EWM). AHP captures expert judgment from hydrology, urban planning, and emergency management specialists, thereby reflecting policy priorities and domain knowledge. EWM calculates weights objectively based on the degree of data dispersion, reducing subjectivity. By assigning equal importance (μ = 0.5) to both methods, this approach balances subjective expertise and objective data characteristics, improving the robustness and reliability of the final weights.

Based on the constructed indicator system, the urban flood resilience scores for each subdistrict were calculated. To provide transparency regarding the contribution of each indicator, the results of all seventeen factor layers are presented in Appendix A 

Table A1. Results of AHP + EWM weight assignment and resilience values for all 17 indicators across five dimensions.

Dimension	Indicator Code	Indicator Name	AHP Weight	EWM Weight	Final Weight (Hybrid)	Weighted Indicator Score (Mean)
Natural Geography	A1	Average Slope	0.0405	0.0605	0.0505	0.3571
	A2	Elevation	0.0149	0.0461	0.0305	0.4985
	A3	Rainstorm Intensity	0.0555	0.1127	0.0841	0.6164
	A4	Water Network Density	0.0217	0.0663	0.0440	0.2296
Infrastructure	B1	Road Network Density	0.0583	0.0613	0.0598	0.3367
	B2	Impervious Surface Ratio	0.0717	0.0529	0.0623	0.4105
	B3	Blue–Green Space Ratio	0.2812	0.0660	0.1736	0.3690
Socioeconomic	C1	Population Density	0.0270	0.0134	0.0202	0.7399
	C2	Economic Intensity	0.0278	0.1290	0.0784	0.2929
	C3	Population Vulnerability	0.0194	0.0338	0.0266	0.4801
	C4	Vulnerable Group Facilities	0.0146	0.0128	0.0137	0.8061
Emergency Management	D1	Emergency Shelter Density	0.0879	0.2033	0.1456	0.1499
	D2	Medical Emergency Coverage	0.0574	0.0418	0.0496	0.6255
	D3	Fire Rescue Response	0.0420	0.0242	0.0331	0.7039
	D4	Emergency Supplies Access	0.0566	0.0136	0.0351	0.7495
Risk Exposure	E1	Historical Waterlogging Density	0.0651	0.0151	0.0401	0.8559
	E2	Critical Rainfall Threshold	0.0580	0.0470	0.0525	0.3320

. For ease of analysis and understanding, the subdistricts were classified into five categories based on resilience scores: High Resilience, Upper-Middle Resilience, Moderate Resilience, Lower-Middle Resilience, and Low Resilience.

In this classification process, the Natural Breaks (Jenks) method was adopted. This method identifies natural groupings in the data to maximize between-group variance and minimize within-group variance. Compared with other methods, the natural breaks approach avoids forcibly splitting similar areas, highlights the intrinsic structure of the data, and reduces the subjectivity of manual classification.

The score ranges for the five categories are presented in

Table 4. Classification criteria and area coverage of urban flood resilience categories.

Category	Score Range	Area (km2)	Percentage of Total Area
High Resilience	9.64–10.83	179.78	9.14%
Upper-Middle Resilience	8.82–9.64	406.53	20.66%
Moderate Resilience	8.03–8.82	560.47	28.49%
Lower-Middle Resilience	7.04–8.03	639.50	32.51%
Low Resilience	6.00–7.04	181.04	9.20%

, along with the corresponding area coverage in both square kilometers and percentage of the total study area.

3.2. Classification Results of Urban Flood Resilience at the Subdistrict Level

Based on the five primary indicators and seventeen secondary indicators, the urban flood resilience of each subdistrict in Shenzhen was evaluated. The comprehensive flood resilience results are shown in the figure below.

From Figure 4, it can be observed that subdistricts with high resilience are mainly concentrated in the urban centers, such as Nanshan, Futian, and Luohu, which are more developed and equipped with well-established infrastructure. In contrast, subdistricts with low resilience are primarily located in the northwestern industrial belt of Shenzhen, particularly in Bao’an, Guangming, Longhua, and surrounding areas. Furthermore, this study conducted a detailed analysis of the resilience performance under each primary indicator, and the results are illustrated as follows.

• Natural Geography: Subdistricts with low or moderately low resilience are primarily distributed in northern Shenzhen along the Guanlan River and Maozhou River basins (Bao’an, Guangming, Longhua), forming a continuous cluster. In contrast, subdistricts with high resilience are mainly distributed along the southern coastal areas of Shenzhen.
• Infrastructure: Low-resilience areas are scattered and mainly found in older urban areas such as Luohu. Moderately low resilience areas are generally adjacent to these old neighborhoods. High-resilience areas are primarily located in the southeastern parts of the city (Dapeng, Yantian), where there is an abundance of blue–green spaces and low levels of urban hardening.
• Socioeconomic Factors: Low-resilience subdistricts are relatively dispersed, often old communities with high population densities and large migrant populations, such as Jihua and Buji. High-resilience areas are concentrated in economically developed zones of Nanshan and Futian districts, such as Yuehai and Nantou.
• Emergency Management: Low-resilience regions are mainly situated in peripheral areas of the city with weaker emergency coverage, such as Nanao and Pingdi. In contrast, high-resilience regions are found in early-developed areas with better emergency infrastructure, such as Luohu.
• Risk Exposure: Subdistricts with low resilience, such as Fenghuang, Dalang, and Yuanshan, have shown poor performance in past flood events. High-resilience subdistricts, including Dapeng, Nanao, and Nanshan, have demonstrated stronger flood resilience.

3.3. Results of Spatial Autocorrelation and Clustering Pattern Analysis

To investigate whether there is spatial correlation in urban flood resilience, this study conducted a global Moran’s I analysis on the results of urban flood resilience. The results are shown in

Table 5. Global Moran’s I results of urban flood resilience in Shenzhen.

Global Analysis	Result
Moran’s I	0.474622
Expected Index	−0.013699
Variance	0.003390
z-score	8.386553
p-value	0.000000

.

The analysis shows that the global Moran’s I value of urban flood resilience across Shenzhen’s subdistricts is 0.474622, indicating a significant positive spatial autocorrelation. In other words, subdistricts with similar resilience levels tend to cluster spatially, and this pattern is statistically significant.

Subsequently, Local Indicators of Spatial Association (LISA) analysis was conducted to further explore the spatial heterogeneity of flood resilience in Shenzhen and to identify specific hotspot areas (high–high clusters) and coldspot areas (low–low clusters). The results of the LISA analysis are illustrated in the following figure.

As shown in Figure 5, the flood resilience of subdistricts exhibits various spatial association patterns at a significant level. Hotspot areas (high–high clusters) are mainly concentrated in the central and southern parts of Shenzhen, such as Huafu, Lianhua, and Liantang subdistricts. This indicates that these areas have relatively high levels of flood resilience and form significant high-value spatial clusters with their surrounding subdistricts, demonstrating strong synergy. Coldspot areas (low–low clusters) are primarily located in the northwest of Shenzhen, including parts of Guangming and Bao’an districts. This suggests that these subdistricts have relatively low flood resilience and are spatially associated with each other, likely due to poorer performance in natural geography, infrastructure, and emergency management. In addition, Hangcheng subdistrict exhibits a high–low clustering pattern, indicating that its flood resilience is significantly higher than that of the surrounding areas, forming a “resilience island,” which is likely closely related to the driving influence of the Bao’an Airport. On the other hand, Buji and Cuizhu subdistricts show a low–high clustering pattern, meaning that, although their own flood resilience is relatively low, they are surrounded by subdistricts with high resilience.

3.4. Results of Geodetector Analysis

The spatial autocorrelation analysis revealed significant spatial heterogeneity in urban flood resilience across Shenzhen’s subdistricts. To further explore the driving factors behind this spatial differentiation, the results of the Geodetector analysis are shown in Figure 6. Green indicates indicators with strong explanatory power for urban flood resilience, cyan indicates moderate explanatory power, and blue represents weak explanatory power.

The results show that, at the level of individual factors, the top six indicators with the highest explanatory power for urban flood resilience in descending order are economic intensity (C2), rainstormintensity (A3), street elevation (A2), shelter density (D1), and critical disaster-causing rainfall threshold (E2). These indicators span the domains of physical geography, socioeconomic factors, emergency management, and risk exposure, suggesting a relatively balanced distribution of key drivers of flood resilience in Shenzhen. It can also be inferred that economic intensity (C2) and street elevation (A2) exert the strongest positive influence on flood resilience.

In contrast, indicators with weak explanatory power mainly include water network density (A4) and population aggregation (C1). This may be due to the significant prevalence of culverted (underground) rivers in Shenzhen. According to statistics, culverted rivers account for 50.59% of the total number of rivers and 19.39% of the total river length. In Futian District, the proportion of culverted river length even reaches 51.47%, which diminishes the explanatory power of surface water network data for flood resilience [40]. Meanwhile, population aggregation may present a “bimodal” phenomenon. For instance, high-end population clusters in urban centers, such as Yuehai and Futian, exhibit high resilience due to well-developed infrastructure and strong emergency response capacity, while urban villages, such as Buji and Shajing, display lower resilience. This coexistence of high and low resilience weakens the overall explanatory power of population aggregation for urban flood resilience.

Next, an interaction analysis was conducted on the various influencing factors, and the results are shown in Figure 7.

At the level of factor interaction, the analysis indicates significant bivariate enhancement or nonlinear enhancement effects across all factors. That is, the q-values resulting from the interactions exceeded the maximum q-values of the individual factors, indicating an increase in explanatory power after the interaction. This demonstrates that the spatial distribution of urban flood resilience in Shenzhen is driven by the combined effects of multiple influencing factors. Among them, the two interactions with the strongest explanatory power were economic intensity (C2) and rainstorm intensity (A3) as well as economic intensity (C2) and density of emergency shelters (D1).

The results of the Geodetector analysis indicate that urban flood resilience in Shenzhen is the result of system coupling. It is jointly influenced by natural geography, infrastructure, socioeconomic development, emergency management, and risk exposure. No single factor can independently explain the spatial heterogeneity of resilience. Therefore, improving urban flood resilience in Shenzhen requires a shift from fragmented enhancements to systematic and collaborative governance.

4. Discussion

4.1. Spatial Patterns of Flood Resilience and the “Resilience Fault Line”

The classification results (Section 3.2) and spatial autocorrelation analysis (Section 3.3) reveal a distinct spatial polarization of flood resilience in Shenzhen, with high-resilience subdistricts concentrated in core urban areas (Nanshan, Futian, Luohu) and low-resilience subdistricts primarily located in peripheral industrial belts (Bao’an, Guangming). By comparing our findings with results from previous flood resilience studies conducted at other spatial scales in Shenzhen and overlaying the identified low-resilience areas with the flood-prone zones published by the Shenzhen Water Authority, its effectiveness has been preliminarily validated, as both show spatial consistency [41,42,43]. Shenzhen’s “resilience fault line” reflects not only historical development disparities but also imbalances in urban renewal.

Historically, the establishment of the “second-line” boundary in 1982 demarcated the city into two parts: inside and outside the Special Economic Zone (SEZ). Early development focused predominantly on the inner zone, leading to a substantial foundational gap in flood resilience between the two regions [44]. Although the SEZ was expanded to cover the entire city in 2010, the outer districts were mainly designated for industrial relocation, resulting in inadequate infrastructure and public services, with urbanization lagging behind industrialization. The delayed development of socioeconomic conditions, infrastructure, and emergency management has created persistent “resilience depressions” in these areas. Compounding this, government-led urban renewal projects—largely funded through public investment—have prioritized economically stronger inner-city districts. The Master Plan for Comprehensive Renovation of Urban Villages (2019–2025) allocates renovation to 75% of urban villages in central districts, far above the citywide average of 55%. This preferential focus risks widening the resilience gap, as already resilient areas continue to strengthen while low-income populations, driven by rising living costs, are displaced to less resilient peripheries—thereby deepening inequality in flood risk exposure.

Similar spatial inequality patterns have been observed in other megacities. In Monterrey, low-resilience areas are concentrated in low-lying regions susceptible to river impacts, where urban infrastructure is inadequate, population density is high, poverty levels are severe, and rapid urban growth has outpaced the provision of infrastructure [45]. In Tehran, resilience inequality is strongly associated with urban planning, infrastructure, and impermeable water surfaces [46]. New York City exhibits a core–periphery resilience divide, as demonstrated during hurricane Sandy, where well-resourced Manhattan neighborhoods recovered more rapidly than outer boroughs due to differences in coastal defenses, emergency preparedness, and socioeconomic resources [47].

Shenzhen’s resilience geography shares common mechanisms with Monterrey and Tehran in terms of infrastructure deficits but also parallels New York’s pattern in which wealthier, better-planned districts exhibit higher resilience. This suggests that resilience inequality is not unique to Shenzhen but is a recurring challenge in rapidly growing or globally significant cities.

4.2. Strategies for Enhancing Urban Flood Resilience

Geodetector analysis (Section 3.4) indicates that flood resilience in Shenzhen is driven by a multi-factor coupling mechanism. Similar multi-factor interactions have been identified in other studies [48,49]. To overcome the current “resilience fault line” dilemma, a holistic approach is required—one that builds a multidimensional and coordinated resilience framework. Accordingly, we propose targeted strategies for low-, medium-, and high-resilience streets as follows.

For low-resilience streets (e.g., parts of Bao’an and Guangming districts), natural geographical and emergency management resilience is notably weak, compounded by longstanding infrastructure deficits. It is, therefore, critical to promote sponge city development and integrate “grey and green” infrastructure to enhance flood resilience. This includes strengthening the ecological protection of mountains, rivers, forests, farmland, lakes, and grasslands to leverage natural surfaces for water infiltration as well as improving drainage systems and transportation infrastructure. Efforts should also focus on addressing deficiencies in emergency medical services and disaster response capabilities by leveraging Artificial Intelligence (AI) algorithms to optimize the allocation of emergency resources and accelerate disaster response as well as integrating IoT and related technologies to enhance early warning and real-time monitoring capabilities, thereby improving the overall efficiency and effectiveness of flood management.

For medium-resilience streets (e.g., parts of Longgang District), natural geographical resilience is relatively strong, but socioeconomic and emergency management resilience remains weaker. Future efforts should steadily advance flood resilience planning and construction, particularly in densely populated urban villages. These areas require enhanced emergency response forces and improved shelter facilities. Emphasis should be placed on mobilizing social forces and public participation in flood prevention and disaster relief, fostering synergistic resilience through collective action.

For high-resilience streets (e.g., parts of Nanshan and Futian districts), the focus should be on maintaining existing advantages by continuously optimizing infrastructure and enhancing emergency capabilities. These areas should also serve as demonstration zones, driving resilience improvements in neighboring streets.

In addition, as a whole, Shenzhen needs to strengthen top-level design and directly address regional disparities in flood resilience. Resources and funding should be directed toward underdeveloped streets and areas to avoid fragmented efforts by individual administrative districts. The current path dependence—“weak in defense, strong in response”—must be corrected. While emergency response capacity continues to improve, more emphasis should be placed on pre-disaster resilience, including road networks, drainage systems, and blue–green infrastructure.

4.3. Limitations of This Study

This study still has certain limitations that need to be addressed in future research. First, due to constraints in data availability, the scope of this study is limited to cross-sectional data at the street level in Shenzhen, making it impossible to conduct time-series analyses of urban flood resilience across different temporal scales. Second, the selection and classification of indicators may involve a certain degree of subjectivity, especially in the weighting process based on the Analytic Hierarchy Process (AHP), which depends on expert judgment. Although the combined AHP–Entropy Weight Method (AHP–EWM) helps to reduce subjectivity, potential biases cannot be completely eliminated. Third, the quality of certain datasets—such as historical waterlogging records and demographic data—may affect the accuracy of the results, particularly when the data are updated infrequently. Finally, the smallest unit of analysis in this study is the street, which means that variations within large streets (e.g., between urban villages and commercial areas) cannot be fully captured. Future research will integrate higher-resolution spatial data and multi-source dynamic monitoring (e.g., satellite remote sensing, IoT sensors) to achieve more accurate and real-time assessments of flood resilience.

5. Conclusions

This study developed a novel multidimensional flood resilience assessment framework at the subdistrict scale, integrating the Analytic Hierarchy Process (AHP) and the Entropy Weight Method (EWM) to balance expert knowledge and data-driven weighting. Combining spatial autocorrelation (Moran’s I, LISA) and the Geodetector model, the analysis revealed clear spatial disparities in Shenzhen’s flood resilience, with high-resilience clusters concentrated in urban cores and low-resilience clusters located in peripheral districts. These patterns reflect uneven distributions of infrastructure capacity, socioeconomic resources, and exposure to flood hazards.

In addition to its methodological contributions, this study’s findings have direct implications for flood management in Shenzhen. The results support the need for Shenzhen to develop coordinated, multidimensional strategies, such as enhancing drainage capacity and expanding blue–green infrastructure in low-resilience areas like Bao’an and Guangming. These recommendations align with the goals of the Shenzhen Climate Change Adaptation Plan (2023–2035) and will provide strong support for building Shenzhen into a climate-resilient and sustainable city.

The framework is not limited to Shenzhen. Its design—grounded in established resilience theory and adaptable to diverse urban contexts—makes it scalable to other rapidly growing coastal megacities in China and Southeast Asia, where fine-grained spatial disparities in resilience are often overlooked. With appropriate adjustments to reflect local hazard profiles, socioeconomic conditions, and governance structures, it can serve as a decision-support tool for urban climate adaptation and disaster risk reduction.

Future research could extend this work by incorporating multi-temporal data to capture resilience dynamics, integrating probabilistic hazard projections to improve long-term planning.