Exploring Socio-Spatial Inequalities in Flood Response Using Flood Simulation and Social Media Data: A Case Study of 2020 Flood in Nanjing, China

Abstract

Identifying socio-spatial inequalities in flood resilience is crucial for effective disaster risk management. This study integrates flood susceptibility simulations and Weibo activity data to construct a flood susceptibility index and incorporates socio-spatial differentiation to represent residents’ coping capacities. By combining flood risk awareness and coping capacity, we develop a comprehensive flood response capability model to examine the spatial patterns of flood resilience inequality. The findings reveal that (1) high flood risk awareness is concentrated near the Yangtze River and major lakes based on social media data and simulations; (2) coping capacity to floods exhibits a central–periphery pattern, with higher resilience in urban centers and gradually decreases gradually to the suburban and exurban areas; (3) communities are classified into four types based on the combination of flood risk awareness and coping capacities. Multiple linear regression analysis indicates that both natural and social factors significantly influence flood response capacity. This research provides critical insights into the spatial patterns of flood resilience, offering valuable guidance for formulating targeted adaptation strategies.

1. Introduction

Flooding is an increasingly frequent and serious natural hazard that significantly impacts the economy, society, and human safety. Urban flooding is especially destructive in densely populated areas and exhibits notable spatial disparities. These disparities are influenced by socio-economic conditions, land use patterns, and demographic factors. The uneven distribution of these elements affects coping capacity, risk perception, and preparedness for floods across various regions, contributing to spatial inequality [1,2,3,4,5]. Research indicates that economically advantaged areas recover more quickly from floods, while disadvantaged regions, which often contain vulnerable populations—such as migrants, the elderly, and children—face greater risks and slower recovery rates [6,7,8]. Differences in infrastructure, resource allocation, and access to information further exacerbate these disparities.

Under the dual influences of rapid urbanization and climate change, floods impact urban populations with significant heterogeneity. Disadvantaged groups disproportionately suffer from these events, exacerbating social inequality and poverty [5,9]. Research conducted by the Environment Agency of the United Kingdom indicates that London’s flood vulnerability exhibits a bimodal distribution, with both impoverished and affluent communities residing in high-risk areas [5]. The Hurricane Katrina disaster highlighted racial disparities in flood risk, as minority communities in New Orleans struggled to access recovery resources after this devastating event [10,11]. The severe rainstorm and flooding in Zhengzhou, Henan Province, in July 2021 resulted in substantial casualties and property damage, raising concerns about urban resilience to such disasters and underscoring the urgent need to enhance social response capabilities for urban flood events.

The concept of coping capacities has emerged as a fundamental paradigm for examining disaster resilience. While most researchers have focused on social capital, economic resources, institutional structures, and infrastructure, there has been a notable oversight regarding the utilization of big data to enhance coping capacities [12,13,14,15,16,17]. Specifically, there is a lack of exploration into the interactions between coping capacities and spatial inequality. Coping capacity refers to the ability of individuals, households, communities, and institutions to manage adverse conditions, emergencies, or disasters using available skills, resources, and strategies [18]. Initiatives from the Federal Emergency Management Agency (FEMA), such as the National Flood Insurance Program (NFIP) and the Community Rating System (CRS), incentivize communities to adopt mitigation measures that ultimately reduce disaster risks. These measures contribute to enhancing community coping capacity while alleviating spatial inequity.

Social media data, a form of spatio-temporal big data, are characterized by real-time and location-based services. Recently, these data been applied in flood risk management in countries such as the United States, Germany, Australia, and Japan [19,20,21,22]. Applications primarily focus on flood monitoring and early warning, spatio-temporal disaster analysis, sentiment and response behavior assessment, disaster loss evaluation, relief deployment strategies, and vulnerability and resilience assessments [23,24,25,26,27,28,29,30]. Although social media data enhance rapid disaster relief through real-time geospatial analysis and information dissemination, challenges such as inadequate geographical accuracy and credibility remain significant concerns [31,32]. Current research trends aim to optimize algorithm models for processing social media data and effectively integrate multi-source information throughout the entire flood risk management process.

This study contributes to the literature in the following ways. First, most prior studies have examined socio-spatial differentiation at the community level through the use of population variables. However, they seldom investigate socio-spatial differentiation from the perspective of hazard, which is vital for comprehending coping capacities during disasters. Socio-spatial differentiation derived from census data effectively reveals how spatial disparities contribute to variations in flood disaster impacts. Utilizing flood simulations, population census data, and real-time Weibo data, this study explores the Flood Susceptibility Index and Coping Capacity Index to assess flood response capabilities. This approach enhances our comprehension of the mechanisms underlying community-level responses to flood disasters. Ref. [33] developed the Social Vulnerability Index (SoVI) to analyze the factors influencing social vulnerability at the county level. Ref. [34] analyzed the Pressure and Release (PAR) model to measure global disaster vulnerability, positing that vulnerability results from root causes, dynamic pressures, and unsafe conditions. Ref. [35] developed Baseline Resilience (BRIC) Indicators for Communities to evaluate community coping capacity. Ref. [4] proposed an environmental justice framework for examining flood risk in the UK, revealing that vulnerable populations encounter increased flood risks due to inequalities in exposure and access to protective resources. Ref. [36] utilized a sustainable livelihood framework to assess climate change vulnerability and adaptation. Ref. [37] found that Indigenous knowledge enhances modern disaster risk reduction strategies. Ref. [13] examined social capital analysis as a method for assessing community resilience in post-disaster contexts; communities with higher levels of social capital typically recover more rapidly than those that rely solely on physical infrastructure. This study contributes to the existing literature by empirically analyzing response capacities in China, emphasizing the relationship between socio-spatial differentiation and coping capacities in flood situations.

Second, the impact of real-time social media on human behavior during disasters has been studied; however, few examine its effects on coping capacities in flood situations, particularly those involving flood simulations. Existing studies have focused on estimating potential damage, reducing disparities, and enhancing resilience [28,30,38,39]. However, access inequalities, misinformation, and the digital divide represent significant limitations of social media platforms. Therefore, it is essential to integrate social media with flood simulations for enhanced precision. Ref. [40] surveyed Japanese Twitter users to study network structure and community evolution during the 2011 earthquake and tsunami, finding that social media can support long-term recovery. Ref. [41] examined geotagged tweets during Hurricane Sandy and found that social media data were instrumental in crisis mapping, enabling targeted resource allocation. Ref. [20] employ text mining, sentiment analysis, and geospatial analysis to evaluate user responses during and after disasters. Real-time Twitter data can enhance disaster management strategies by providing actionable insights into public needs and resilience. Ref. [20] analyzed disaster severity through social media, using tweets from the South East Queensland Flood to demonstrate that empowering communities through social media enhances resilience. Ref. [24] analyzed Weibo posts about the Nanjing flood and found a significant correlation between online activity and actual damage levels in urban areas, highlighting social media’s role in assessing urban resilience. Ref. [30] employed text mining and sentiment analysis to understand Henan Deluge recovery and found that recovery timelines varied across urban and rural areas, reflecting disparities in resource allocation and response efficiency. Existing studies on disaster response using social media data primarily focus on quantitative methods, such as statistical analyses of information volume and keyword extraction. These studies often overlook the deeper exploration of social and economic factors, resident behavior patterns, and their connections to urban spatial structures. Our findings indicate that combining social media with flood simulations enhances our understanding of potential risks. Thus, we analyze flood susceptibility levels by integrating these two data sources.

Third, existing studies on disaster recovery assessments primarily focus on developed and a limited number of developing countries, with minimal research from an environmental justice perspective to explore the link between flood exposure and socio-spatial differentiation. Ref. [4] found that social, economic, and geographic factors contribute to uneven preparedness and responses to flood risks. Ref. [42] noted that flood risks in Miami disproportionately impact socially disadvantaged neighborhoods, thereby exacerbating existing inequalities. Ref. [7] developed a model linking environmental injustice to flood risk exposure. Ref. [43] revealed significant barriers faced by individuals with disabilities during Hurricane Harvey concerning evacuation, shelter access, and recovery services. Ref. [44] provide insights into how targeted interventions can reduce social inequalities in disaster risk exposure. In China, flood disaster research tends to concentrate on the events themselves, neglecting socio-economic factors such as income and age when assessing flood risk exposure. Existing research on socio-spatial differentiation is abundant; however, studies linking this field to disaster resilience and recovery remain scarce in China. Thus, we focus on analyzing the mechanisms and spatial differentiation of flood response capacity by integrating social media, demographic, and flood data.

This research examines 99 communities in Nanjing to develop a disaster recovery evaluation system for flood events. It integrates the Flood Susceptibility Index, Coping Capacity Index, and Response Capability Index using multi-sourced data, including social media platforms. The response capabilities are categorized into four groups. Finally, the study analyzes factors influencing these response capacities.

2. Research Methodology

2.1. Study Area

Nanjing is the capital city of Jiangsu Province and an important central city in the eastern region of China, covering a total area of 6587 km². In 2023, the population was 9.547 million, with an urbanization rate of 87.2%. Nanjing is one of the 31 national key flood control cities in China. It has experienced significant urban flooding, particularly during June and July, which are identified as the most frequent periods for such events. In July 2016, significant flooding occurred in the Hexi and Yuhuatai districts of Nanjing. On 18 July 2020, the Yangtze River flood set a new record in the history of Nanjing’s hydrology. Nanjing comprises 11 districts and 99 neighborhoods. The urban center encompasses four districts (Xuanwu, Qinhuai, Jianye, and Gulou), which contain 38 neighborhoods. Suburban areas consist of five districts (Pukou, Qixia, Jiangning, Yuhuatai, and Luhe), featuring 46 neighborhoods. Exurban areas comprise the Lishui and Gaochun districts, which contain 15 neighborhoods (Figure 1).

2.2. Methodology

2.2.1. Framework

Key dimensions of spatial inequality in urban flood disasters encompass three aspects: the flood risk index (FRI), coping capacity (CC), and flood response capability (FRC). The flood risk awareness (FRA) reflects residents’ awareness of potential flood disasters by integrating social media data with flood simulations. Coping capacity (CC) reflects the ability of individuals, organizations, and systems to utilize available skills and resources to effectively manage risks and disasters. flood response capability (FRC) is derived from a comprehensive assessment that integrates the Flood Risk Awareness Index and the Flood Response Capability Index. By constructing these indices to represent flood exposure and socio-economic disparities, we analyze the factors influencing socio-spatial differentiation in flooding within Nanjing. Furthermore, based on the interaction between flood risk awareness and coping capacity, we identify spatial types of flood response capabilities. The assessment steps are outlined as follows (Figure 2).

2.2.2. Data Collection and Cleaning

The datasets employed in this study (

Table 1. Evaluation index system of flood response capability in Nanjing.

	Data Type	Explanation	Data Source	Description
Flood Risk Awareness (FRA)	Social media data	The number of Weibo posts related to flood	http://weibo.com accessed on 23 June 2022	19,599 in total 966 in flooded
	Precipitation	Daily rainfall	Nanjing Meteoological Bureau	Daily precipitation data
	Soil Type	Nurturing soil, allite, pedocal, yellow-cinnamon soil, yellow brown soil, anthrosols	http://geodata.pku.edu.cn accessed on 23 June 2022	1 km raster data
	Land use	Urban land, farm land, forestry, water bodies	http://www.globallandcover.com accessed on 23 June 2022	30 m raster data
	Elevation	DEM	http://www.gscloud.cn accessed on 23 June 2022	30 m raster data
	POIs	Ponding points, supermarkets, companies, enterprises, and Park Plaza locations	http://www.amap.com accessed on 23 June 2022	18,333
	River Network	Main rivers, tributaries, minor tributaries	http://www.openstreetmap.org accessed on 23 June 2022	Vector data
Coping Capacities (CCs)	Rural	Proportion of rural population (%)	Seventh National Population Census of China	Statistical data
	Population	Proportion of local households (%)	Seventh National Population Census of China	Statistical data
		Proportion of urban population with separate household registration (%)		Statistical data
		Population density (persons/km2)		Statistical data
		Population growth rate over the past 10 years (%)		Statistical data
	Education	Proportion of illiterate people (%)	Seventh National Population Census of China	Statistical data
		Preschool education (%)		Statistical data
		Proportion of primary school (%)		Statistical data
		Proportion of middle schools (%)		Statistical data
		Proportion of high schools (%)		Statistical data
		Proportion of colleges (%)		Statistical data
		Proportion of undergraduates (%)		Statistical data
		Proportion of master students (%)		Statistical data
		Proportion of doctoral students (%)		Statistical data
	Gender	Proportion of women (%)	Seventh National Population Census of China	Statistical data
	Age structure	Proportion of aged 0–14 (%)	Seventh National Population Census of China	Statistical data
		Proportion of aged 15–59 years (%)		Statistical data
		Proportion of aged over 60 (%)		Statistical data
	Immigrant	Proportion of internal migrants in the province (%)	Seventh National Population Census of China	Statistical data
		Proportion of migrants from other provinces (%)		Statistical data
	households	Proportion of one-generation households (%)	Seventh National Population Census of China	Statistical data
		Proportion of second-generation households (%)		Statistical data
		Proportion of households with three generations or more (%)		Statistical data
		Proportion of households with an elderly population (%)		Statistical data
	Home value	Average household size (persons/household)	Seventh National Population Census of China	Statistical data
		Average number of rooms per household (rooms/household)		Statistical data
		Per capita housing floor space (square meters/person)		Statistical data

) include flood simulation inputs, social media activity records, and socio-spatial indicators. Flood simulation data comprise remote sensing imagery and rainfall records. Land use data were derived from the “GlobeLand30” remote sensing dataset for Nanjing. The 30 m Digital Elevation Model (DEM) was created from Landsat-8 satellite images taken in September 2018. Soil-type data originated from the “1:1,000,000 Soil Type Map of Jiangsu Province”, spatially corrected using the Mercator projection. Point of Interest (POI) data, collected from the 2018 Gaode Map, encompass four categories: ponding points, supermarkets, corporate enterprises, and public parks. Daily rainfall records from 10 June to 21 July 2020 were obtained from the National Meteorological Information Center (http://data.cma.cn/, accessed on 22 May 2023).

Social media data were collected using a Python-based web crawler within the Scrapy framework, utilizing the Requests library to retrieve webpage source codes and filter posts from 10 June to 21 July 2020. This process yielded 19,599 geotagged tweets. Flood-related content was identified through keyword searches (e.g., “rainstorm rescue”, “flooded”), resulting in 966 relevant tweets (4.9% of the total dataset). To address the small sample size, flood simulation outputs were integrated with social media data for cross-validation. This approach is consistent with methods used for sparse social media data analysis, including the work by [28], who examined Hurricane Irma tweets (0–40 posts per census tract) in Pinellas County, Florida, USA. Geotagged data were converted into coordinate pairs via the Gaode API for spatial visualization.

Coping capacity indicators were sourced from Nanjing’s 2010 and 2020 national population censuses. Twenty-seven variables capturing individual, household, and regional attributes were selected. Neighborhood-level population growth rates (2010–2020) were computed to quantify socio-spatial structural evolution.

The indicator index used in this study is presented in Table 1. The flood simulation data encompass various components, including remote sensing images and rainfall data. The land use information was derived from the “GlobeLand30” remote sensing imagery of Nanjing. The Digital Elevation Model (DEM) data were sourced from medium-resolution Landsat 8 satellite images acquired in September 2018. Soil-type data originated from the “1:1,000,000 Soil Type Data of Jiangsu Province”, which were corrected using Mercator projection techniques. Point of Interest (POI) data were collected from the Gaode map for 2018 and included four categories: ponding points, supermarkets, companies and enterprises, as well as Park Plaza locations. Rainfall data were obtained from the National Meteorological Information Center (http://data.cma.cn/ accessed on 22 May 2022) and consisted of daily rainfall records spanning from 10 June to 21 July 2020.

The data from social media were processed using a Python web crawler program. The Requests tool was employed to obtain the source code for parsing microblog pages, allowing for the filtering of relevant data between 10 June 2020 and 21 July 2020. We obtained 19,599 check-in records as background tweets, and then we used keywords such as “rainstorm rescue”, “rainstorm assistance”, “waterlogged road sections”, and “flooded”. The collected information primarily included user IDs, content of microblogs and release timestamps, among other details. In total, 966 tweets (4.9% of 19,599 background tweets) were identified as flooding-related tweets. To address the low effective data volume, we combined flood simulation data with social media data for cross-verification. Integrating small sample social media data with other sources can effectively mitigate data scarcity compared to existing international studies. Ref. [28] studied Twitter data on Hurricane Irma in Pinellas County, FL, USA, filtering it to count the number of posts per census tract, which ranged from 0 to 40. Finally, the geographic information gathered was converted into coordinate data through the Gaud API to facilitate further visualization.

Indicators of Socio-Spatial Differentiation Assessment: The data utilized in this study were obtained from the sixth and seventh national population censuses of Nanjing from 2010 and 2020. A total of 27 fundamental variables pertaining to individual, family, and regional characteristics were selected. The average annual growth rate of the population across each neighborhood over a decade was calculated to illustrate the differentiation and evolution of social spatial structures.

2.2.3. Methods

(1)

Assessing the Index of Flood Risk Awareness (FRA)

Existing studies [29,45] have demonstrated that Twitter data can capture public risk perceptions and disaster dynamics in real time, thereby complementing traditional monitoring data (

Table 2. Evaluation of flood risk awareness in related references.

Index	Formula	Description	Reference
Public ConcernIndex (PCI)	$PCI={\displaystyle \frac{Rainstorm related weibo}{Background Weibo}}$	The Public Concern Index (PCI) is calculated by dividing the number of tweets related to flooding by the total number of background tweets within a specific grid or time frame (such as a day or a two-hour interval). This index reflects the level of public awareness and concern regarding urban flooding.	[24]
Normalized Ratio Index (NRI)	$Ratio={\displaystyle \frac{Sandy related Tweets}{Background Tweets}}$ $NRI={\displaystyle \frac{Ratio}{Treat level}}$	The Ratio Index is defined as the number of disaster-related tweets within a region divided by the baseline number of tweets in the same region over a specified period, reflecting residents’ awareness of the event or its impact. The Normalized Ratio Index (NRI) further divides the Ratio Index by the disaster’s threat level, enabling comparisons of Twitter activity across study areas under equivalent threat conditions.	[20]
Hazard Risk Awareness (HRA)	$HRA={\displaystyle \frac{{{(RN}_{tweets})}^m}{{\left(hazard\right)}^n}}$ ${RN}_{tweets}={\displaystyle \frac{N_{tweets}}{Population\times {Pr}_{internet}}}$ ${hazard}_t=P_t\times W_t$	${\mathrm{R}\mathrm{N}}_{\mathrm{t}\mathrm{w}\mathrm{e}\mathrm{e}\mathrm{t}\mathrm{s}}$ represents the standardized tweet count, ${\mathrm{N}}_{\mathrm{t}\mathrm{w}\mathrm{e}\mathrm{e}\mathrm{t}\mathrm{s}}$ denotes the total number of tweets, ${\mathrm{P}\mathrm{r}}_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{n}\mathrm{e}\mathrm{t}}$ indicates the percentage of households with internet access, P_t and W_t represent the rainfall and wind speed on day t, respectively. By substituting precipitation and wind speed with other drivers reflecting the severity of disaster events, the Hazard Risk Awareness (HRA) Index can be used to assess public awareness of hazard risks associated with various natural disasters.	[29]

). To analyze the intensity of social media responses to floods, this study assesses flood risk awareness (FRA) based on Weibo data, which reflects residents’ awareness of potential flood disasters. Weibo is the largest microblogging platform in China.

Considering the positive correlation between the number of Weibo posts and population size, where central urban areas exhibit high population densities and correspondingly high volumes of Weibo postings, while suburban areas demonstrate low population densities and low volumes of postings, the Weibo ratio is utilized to assess the impact of population density on the number of Weibo posts.

$$\mathrm{W}\mathrm{e}\mathrm{i}\mathrm{b}\mathrm{o} \mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}={\displaystyle \frac{\mathrm{F}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d} \mathrm{r}\mathrm{e}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d} \mathrm{w}\mathrm{e}\mathrm{i}\mathrm{b}\mathrm{o}}{\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n} \mathrm{s}\mathrm{i}\mathrm{z}\mathrm{e}}}$$

(1)

The number of flood-related Weibo posts refers to posts containing information about flood disasters; the population size is obtained from the permanent resident population data of the Seventh National Population Census of China. This study further combines the Weibo data with rainfall data and spatial data to analyze the differences in social media activity at identical flood susceptibility levels within the study area. The “Flood Risk Awareness (FRA) Index” is calculated by dividing the “Weibo ratio” by the “flood susceptibility level”.

$$\begin{array}{c}\mathrm{FRA}={\displaystyle \frac{\mathrm{W}\mathrm{e}\mathrm{i}\mathrm{b}\mathrm{o} \mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{s}}{\mathrm{f}\mathrm{l}\mathrm{o}\mathrm{o}\mathrm{d} \mathrm{s}\mathrm{u}\mathrm{s}\mathrm{c}\mathrm{e}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y} \mathrm{l}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}}}\end{array}$$

(2)

$\mathrm{F}\mathrm{R}\mathrm{A}$ is the index of flood risk awareness. The flood susceptibility level is defined as the ratio of the area inundated beyond 0.4 m for each community. The data for this study are derived from our research group’s previously published paper [46]. Flood simulations in Nanjing were conducted using the ArcSWAT 2012 model implemented within the ArcGIS 10.3 platform. The analysis incorporated four primary datasets: a 30 m resolution digital elevation model (DEM), land use maps, soil-type classifications, and rainfall records. First, the DEM was preprocessed in ArcGIS 10.3 to generate hydrological inputs. This involved filling depressions (Z-limit: 30 m) and extracting a river network using a flow accumulation threshold of 1000 cells. The processed DEM and derived river vector file were then imported into ArcSWAT to delineate the study area into 259 sub-basins. Hydrological response units (HRUs) were subsequently defined by integrating land use, soil type, and slope data, resulting in 4754 HRUs. Rainfall data were input to simulate runoff depth, which was spatially overlaid with neighborhood boundaries in ArcGIS. A runoff depth threshold of 0.4 m was established to indicate severe urban flooding. Neighborhoods were classified into five flood susceptibility categories (low, medium–low, medium, medium–high, high) based on the proportional area exceeding this threshold, with 99 neighborhoods categorized accordingly.

(2)

Assessing Index of Coping Capacity (CC).

To examine the disparities of coping capacity under the same threat level for the study area, we developed the second index, the coping capacity (CC). Coping capacity (CC) reflects the ability of individuals, organizations, and systems to utilize available skills and resources in order to effectively manage risks and disasters [47]. The socio-spatial differentiation index was used to identify the adaptability and resilience of social groups in different regions when facing disasters. Based on the Seventh National Population Census of China, a total of 27 population and economic-related variables were selected to assess the coping capacity of each community by factor analysis.

Factor analysis is conducted by using the PCA method in factor extraction to simplify the factor structure of a group of original variables. Generally, the linear combination used to determine factors is given as follows:

$${CC}_{ik}=\sum _j{W}_{jk}{X}_{ij}$$

(3)

CC_ik = score of coping capacity of community i on factor k;

X_ij = value of original variable j for community i, which is standardized in PCA;

W_jk = factor loading of variable j on factor k representing the proportion of variance of variable j explained by factor k.

The comprehensive score of community i is then calculated by using a weighted sum method.

$$S_i={\sum }_k\sqrt{{\lambda }_k}g{F}_{ik}/{\sum }_k\sqrt{{\lambda }_k}$$

(4)

S_i = comprehensive factor score of coping capacity of community i representing the coping capacity of the community;

λ_k = the eigenvalue of factor k.

(3)

Combining FRA and CC to FRC (Flood Response Capability Index)

The third index, FRC, represents the average comprehensive score from FRA and CC. We use min–max rescaling to normalize all of the indicators. The FRC scores range from 0 to 1. The positive index was used to standardize Formula (5), whereas the negative index was used to standardize Formula (6).

Positive indicators:

$$Y_{ij}=(X_{ij}-X_{j\min })/(X_{j\max }-X_{j\min })$$

(5)

Negative indicators:

$$Y_{ij}=(X_{j\max }-X_{ij})/(X_{j\max }-X_{j\min })$$

(6)

where ${\mathrm{X}}_{ij}$ is the raw data in row i, column j; $X_{jmax}$ is the maximum value of the raw data in column j; ${\mathrm{X}}_{jmin}$ is the minimum value of the raw data in column j; and ${\mathrm{Y}}_{ij}$ represents the data in row i, column j after normalization.

$$\begin{array}{c}FRC=FRA\times P_1+CC\times P_2\end{array}$$

(7)

FRC is the Flood Response Capability Index; $P_1$ and $P_2$ represent the weight of FRA and CC, respectively. The entropy method determines weights based on two indicators.

For the normalized value $Y_{ij}$, we calculated the proportion $P_{ij}$ of $i$ community under the index of item $j$.

$$P_{ij}={\displaystyle \frac{Y_{ij}}{{\sum }_{i=1}^m{Y}_{ij}}}\left(i=\mathrm{1,2},3,\dots ,n;j=\mathrm{1,2},3,\dots ,m\right)$$

(8)

where $n$ is the number of communities and $m$ is the number of measurement indicators.

We calculated the entropy ($e_j$) of the index $j$.

$$e_j=-\left({\displaystyle \frac{1}{\mathit{ln}\left(n\right)}}\right)\sum _{i=1}^m{p}_{ij}\mathit{ln}\left(p_{ij}\right),e_j\ge 0$$

(9)

We also normalized the difference coefficient, and calculate the weight ($W_j$) of the index $j$

$$W_j=(1-e_j)/\sum _{j=1}^m(1-e_j)\left(j=\mathrm{1,2},3,\dots ,m\right)$$

(10)

(4)

Analyzing influencing factors of FRC

A multiple linear regression was constructed with the Flood Response Capability Index as the dependent variable and influencing factors as the independent variable.

$$Y_i={\beta }_0+{\beta }_1{X}_{1i}+{\beta }_2{X}_{2i}+\cdots +{\beta }_k{X}_{ki}+{\mu }_i$$

(11)

$Y_i$ is the dependent variable (flood response capability); $\beta$ is the regression coefficients; and $\mu$ is a random error; $k$ is the number of explanatory variables; $i$ is the subscript of observation value; $n$ is the sample size.

3. Results

3.1. Flood Risk Awareness (FRA)

In this study, we utilized the previous research findings of our research group. The spatial distribution of urban flood susceptibility indicates the following percentages: low (17.17%), medium–low (27.27%), medium (26.26%), medium–high (14.14%), and high (15.15%). Most high and medium–high flood susceptibility areas in cities are located in urban centre and exurban areas, while most suburban and exurban areas are at low and medium–low flood susceptibility. The main factors contributing to flood susceptibility in Nanjing include topography and green coverage rate. Areas near the Yangtze River and low-lying areas are prone to flooding. Rapid urbanization has led to the conversion of a large amount of ecological land into construction land, exacerbating surface runoff [46] (Figure 3).

By analyzing the ratio of Weibo posts, a similar pattern emerges: high values are concentrated in central urban areas, while lower values are observed in suburban regions. Most areas exhibiting high values are situated near water bodies, including the Yangtze River and its tributaries, as well as the vicinity of Shijiu Lake, and this observation aligns with empirical realities. The areas located near water bodies and low-lying areas are mostly located in the central urban areas and Hexi New Town. These areas have a large number of Weibo posts and high population density. During flood events, the infrastructure, such as roads and public service facilities in these areas is severely impacted. The content of Weibo posts from these regions primarily focuses on rising water levels, damage to living conditions, and disruptions to daily life (Figure 4).

The flood susceptibility level and Weibo rate were combined to derive the Flood Risk Awareness Index using the natural breakpoint method, categorizing it into five levels: low, medium–low, medium, medium–high, and high. High-value areas are located near the Yangtze River, Xuanwu Lake, and Shijiu Lake. Higher values are found in the suburbs, particularly south and north of the Yangtze River, predominantly in densely populated regions surrounding Jiangning New Town and Jiangbei New Town. Residents in these areas exhibit a medium level of awareness regarding flood risk. The lower value area is located in the suburban region, where residents demonstrate reduced awareness of flood risks due to fewer Weibo posts and predominantly forested land.

In 2020, Nanjing experienced its longest plum rain season of the century, which lasted for 50 days. From 10 June to 20 July, daily rainfall exceeded 300 mm on multiple occasions. During this period, water levels in the Nanjing section of the Yangtze River reached unprecedented heights. Furthermore, analysis of the correlation between the number of Weibo posts and hourly rainfall variations indicates that peaks in rainfall correspond to increases in Weibo activity (Figure 5). Three peaks in rainfall occurred during the study period, specifically around 15 June, 28 June, and 17 July. Prior to the flood, the number of Weibo posts related to flooding remained at a relatively low and stable level. During the flood, the number of related Weibo posts reflected fluctuations in urban rainfall. After the flood, as the rain ceased, the number of related Weibo posts gradually decreased and eventually returned to pre-rainfall levels. This conclusion suggests that the number of flood-related Weibo posts is synchronized with rainfall.

3.2. Coping Capacity (CC)

To assess coping capacity, twenty-seven population-related evaluation indicators were selected based on existing literature and empirical studies [12,48,49]. To ensure the robustness of the analysis, we conducted a principal component analysis (PCA) on these indicators. The Kaiser–Meyer–Olkin (KMO) value was 0.705, and Bartlett’s test of sphericity yielded an observed statistic of 5141.194, with a corresponding p-value of 0.000. This indicates that the correlation coefficient matrix significantly differs from an identity matrix, confirming that factor analysis is both necessary and suitable for this evaluation. The results of the factor analysis identified five principal components, which collectively account for 85.37% of the variance, effectively reflecting the overall coping capacity (

Table 3. The principal factor loading matrix of coping capacity in Nanjing in 2020.

Principal Components	Variance Contribution (%)	Variables	Load
Factor 1. Rural population	24.41	Proportion of rural population (%)	0.741
		Proportion of illiterate people (%)	0.603
		Proportion of primary school (%)	0.712
		Proportion of high schools (%)	−0.917
		Proportion of colleges (%)	−0.752
		Average number of rooms per household (rooms/household)	0.786
		Per capita housing floor space (square meters/person)	0.826
		Population density (persons/km2)	−0.696
Factor 2. Elderly and female	22.52	Proportion of women (%)	0.539
		Proportion of aged 15–59 years (%)	−0.928
		Proportion of aged over 60 (%)	0.864
		Proportion of local households (%)	0.799
		Proportion of internal migrants in the province (%)	−0.721
		Proportion of migrants from other provinces (%)	−0.886
		Proportion of households with an elderly population (%)	0.766
		Population growth rate over the past 10 years (%)	−0.594
Factor 3. Highly educated population	19.76	Proportion of middle schools (%)	−0.879
		Proportion of undergraduates (%)	0.809
		Proportion of master students (%)	0.919
		Proportion of doctoral students (%)	0.876
		Proportion of one-generation households (%)	−0.622
		Proportion of second-generation households (%)	0.481
Factor 4. Children and migrants	10.64	Proportion of aged 0–14 (%)	0.935
		Preschool education (%)	0.904
		proportion of urban population with separate household registration (%)	0.431
Factor 5. Family size	8.04	Proportion of households with three generations or more (%)	0.804
		Average household size (persons/household)	0.897
Total	85.37

).

The first factor, representing 24.41% of the variance, was designated as the ‘rural population factor’. The first principal component is strongly correlated with the proportion of the rural population, the illiteracy rate, the proportion of primary school enrollment, the proportion of high school enrollment, the average number of rooms per household, and per capita floor space of housing. This indicates that household education and economic status are critical factors influencing flood coping capacity. Rural areas are typically characterized by larger family sizes, lower educational attainment, and relatively fewer information resources and social networks, particularly regarding social media use, compared to urban areas, resulting in a comparatively lower flood coping capacity (Figure 6).

The second factor, representing 22.52% of the variance, was designated as the ‘elderly and female factor.’ The second principal component is strongly correlated with the proportion of females, the proportion of individuals aged over 60, the proportion of local households, and the proportion of households with elderly members. This indicates that age and gender significantly influence flood coping capacity. Communities with a higher proportion of elderly and female populations require greater attention and assistance during flood events, which contributes to their relatively low flood coping capacity.

The third factor, representing 19.76% of the variance, was designated as the ‘high-education population factor’. The third principal component is highly correlated with the proportion of undergraduates, the proportion of master’s students, the proportion of doctoral students, and the proportion of first-generation and second-generation households. This suggests that educational level is a key factor affecting flood coping capacity; communities with highly educated individuals exhibit better preparedness and response to disasters. Consequently, these communities have higher coping capacities than those with lower educational attainment.

The fourth factor, representing 10.64% of the variance, was designated as the ‘children and migrant factor’. The fourth principal component is highly correlated with the proportion of individuals aged 0–14, the proportion of preschool education, and the proportion of the urban population with separate household registration. This indicates that children aged 0–14 are important factors affecting flood coping capacity. Communities with high proportions of children aged 0–14 are more vulnerable to floods and exhibit relatively low flood coping capacities.

The fifth factor, representing 8.04% of the variance, was designated as the ‘family size’ factor. The fifth principal component is highly correlated with the proportion of households comprising three generations or more. This indicates that household size is an important factor affecting flood coping capacity. During flood events, larger households offer greater opportunities for mutual assistance. However, larger family sizes entail more members requiring care, which complicates emergency evacuation and diminishes flood disaster coping capacity.

The spatial pattern of coping capacity generally exhibits a central–peripheral distribution, indicating that coping capacity is high in the central city and gradually decreases toward the suburbs. High levels of coping capacity are concentrated in urban centers, where a significant number of highly educated individuals reside. These individuals are better equipped to access disaster response information during crises, which contributes to their relatively high coping capacity. Although many older communities exist in urban centers, most of these communities were established before 2000 and are characterized by close neighborhood ties and a high level of community cohesion.

Medium levels of coping capacity are observed in suburban areas, primarily influenced by factors such as children, immigrants, and family size. Over the past two decades, Nanjing’s urban area has expanded into the suburbs, with university towns established in Jiangning, Jiangbei New Area, and Xianlin. This expansion has resulted in a higher concentration of young populations in these suburban regions, thereby amplifying the influence of child and migrant factors. However, since many suburban communities consist of migrants, community cohesion tends to be weaker, resulting in moderate flood coping capacity in these areas.

Low levels of coping capacity are found in exurban areas, primarily influenced by factors such as the rural population, the elderly, and women. These areas are primarily characterized as rural, with relatively lower levels of education and larger family sizes. The elderly and women represent a significant portion of the rural population. Residents in these areas often lack essential knowledge to effectively respond to disasters, coupled with a scarcity of resources. Consequently, this results in comparatively lower coping capacities.

3.3. Flood Response Capability (FRC)

By analyzing flood risk awareness and coping capacity, the two variables are plotted in a 2D coordinate system. Most neighborhoods are concentrated in the low risk awareness and high coping capacity (L-H) range (55.56%), followed by the low risk awareness and low coping capacity (L-L) range, which accounts for 36.36%. Only a few neighborhoods appear in the high risk awareness and high coping capacity (H-H) range, accounting for 5.05%, and in the high risk awareness and low coping capacity (H-L) range, accounting for 3.04% (Figure 7).

The first category is the L-H range, characterized by low risk awareness and high coping capacity. This category comprises 55.56% of the communities and is primarily distributed in urban centers and suburban areas, such as Jiangpu, Taishan, and Dingshan. These communities seldom experienced serious flood events in the past; the residents’ flood disaster risk awareness is generally low. These communities seldom experienced flooding in the past, and residents have low awareness of flood disaster risk. These areas have sufficient infrastructure, which effectively reduces the risk. Enhancing flood disaster risk awareness will be a priority in the future (Figure 8).

The second category is the L-L range, characterized by low risk awareness and low coping capacity. This category comprises 36.36% of communities, primarily located in exurban areas such as Chunhua, Hushu, and Moling. These rural areas lack adequate public services, have limited access to information, and possess few communication channels for residents. Consequently, this results in relatively low risk awareness and coping capacity.

The third category is the H-H range, characterized by high risk awareness and high coping capacity. This category comprises 5.05% of communities, primarily located in urban centers such as Yuejianglou, Shuangzha, Qixia, Xuanwumen, and Suojincun. These communities near the Yangtze River and Xuanwu Lake have a strong memory of flooding, resulting in a high awareness of flood risk. The urban center has dense infrastructure and strong community cohesion, which contributes to a high coping capacity.

The fourth category is the H-L range, characterized by high risk awareness and low coping capacity. This category comprises 3.04% of communities, primarily located in suburban and exurban areas such as Baguazhou, Shiqiu, and Yongning. These areas are situated near the Yangtze River or other water bodies, have a history of flooding, and residents possess a high level of risk awareness. However, their exurban location, sparse infrastructure, and lower income and education levels result in a low coping capacity.

3.4. Influencing Factors of Flood Response Capability (FRC)

The multiple regression model is used for calculation by SPSS24.0, taking the Flood Risk Awareness Index, Flood Coping Capacity Index and Flood Response Capability Index as Y and natural geographical conditions, population and economic indicators as X. The results show the following: the Flood Risk Awareness Index R² = 0.228, F = 3.893 (Sig. = 0.000), the Flood Coping Capacity Index R² = 0.837, F = 51.296 (Sig. = 0.000); the Flood Response Capability Index R² = 0.440, F = 8.705 (Sig. = 0.000). The regression equation has a high degree of fitting and effectiveness (

Table 4. Multiple regression model of socio-spatial differentiation of flooding in Nanjing, China.

Explanatory Variables	Model 1	Model 2	Model 3
	Flood Risk Awareness Index	Coping Capacity Index	Flood Response Capability Index
Density of flooding points X1	0.303 ** (2.429)		0.244 ** (2.296)
Water area ratio X2
Vegetation coverage rate X3		0.195 * (1.878)
Average slope X4	0.530 ** (2.249)	0.473 ** (3.022)	0.571 ** (2.840)
Average elevation X5	−0.454 * (1.802)	−0.588 *** (−3.516)	−0.535 ** (−2.486)
Construction land ratio X6	−0.222 ** (2.270)	0.153 ** (2.353)	−0.152 * (−1.821)
Highly educated population ratio X7		0.864 *** (8.158)	0.413 ** (3.031)
Density of supermarket X8	0.017 ** (2.480)		0.015 ** (2.545)
Density of business X9		0.008 ** (2.118)
Density of park and square X10	0.135 ** (2.598)		0.113 ** (2.555)
Constants	0.005 * (0.073)	0.082 * (1.924)	0.024 (0.445)
Adjust R2	0.228	0.837	0.440
F value	3.893	51.296	8.705

Note: t values in brackets; *, ** and *** are significant at 10%, 5% and 1%, respectively.

). According to the regression results, multiple linear regression equations can be obtained.

$$\begin{gathered} Y_{FRA}=0.303X_1+0.530X_4-0.454X_5-0.222X_6+0.017X_8+0.135X_{10}+0.005 \\ {Y}_{CC}=0.195X_3+0.473X_4-0.588X_5+0.153X_6+0.864X_7+0.008X_9+0.082 \\ {Y}_{FRC}=0.244X_1+0.571X_4-0.535X_5-0.152X_6+0.413X_7+0.015X_8+0.113X_{10}+0.024 \end{gathered}$$

The explanatory power of risk awareness, coping capacity, and response capability to flooding is 0.228, 0.837, and 0.440, respectively. Coping capacity is most influenced by natural geographical conditions, population data, and economic indicators. Response capability is primarily influenced by population data and economic indicators. Risk awareness tends to be less influenced by natural conditions and is more closely associated with the characteristics of population and the economy.

The main influencing factors of flood risk awareness include the density of flooding points, average slope, average elevation, construction land ratio, density of supermarkets, and density of parks and squares. Geographical conditions and public spaces are closely linked to social media activity. The construction land ratio reflects the extent of urban areas where social media activity is primarily concentrated. First, population density is higher in urban centers than in exurban areas, leading to a greater volume of social media data being collected from these areas. Second, the population characteristics of urban centers are dominated by young and middle-aged non-agricultural populations, who are more likely to use social media. The density of flood points reflects the intensity of rainstorms; such areas usually generate more social media data. During flooding events, supermarkets and public spaces, such as parks, reflect residents’ situations; groups in parks are more likely to be affected by floods, thereby increasing flood risk awareness.

The main influencing factors of flood coping capacity include vegetation coverage, average slope, average elevation, construction land ratio, the ratio of highly educated populations, and business density. The significance level of average elevation reaches 0.001, with a regression coefficient of −0.588, indicating that areas with lower elevations are more prone to water accumulation; however, residents in these areas demonstrate relatively stronger flood coping capacity. The significance level of the ratio of highly educated populations reaches 0.000, with a regression coefficient of 0.864. Regions with a high ratio of educated populations tend to have sufficient disaster information and emergency response capabilities, resulting in stronger flood coping capacity. Business density reflects socio-economic conditions, with higher densities indicating better economic conditions. Residents in these areas often possess better emergency response capabilities, demonstrating stronger flood coping capacity.

The main influencing factors of flood response capability include the density of flooding points, average slope, average elevation, construction land ratio, the ratio of highly educated populations, business density, and the density of parks and squares. The significance level of flooding points is 0.024, with a regression coefficient of 0.244, indicating that areas with flooding points tend to exhibit relatively higher response capability. The significance level of the ratio of highly educated populations is 0.003, with a regression coefficient of 0.413, indicating that higher education levels are associated with enhanced response capability. The significance level of supermarket density is 0.013, with a regression coefficient of 0.015, indicating that regions with a higher density of supermarkets are often situated in urban centers, suggesting that well-equipped areas tend to exhibit a higher response capacity. The significance level of park square density is 0.012, with a regression coefficient of 0.113, suggesting that areas with a higher density of park squares tend to have greater green space coverage, which can effectively mitigate rainwater runoff and enhance the region’s emergency response capabilities.

4. Conclusions and Discussion

The impact of flood disasters demonstrates spatial inequality, as communities lacking coping capacity are often situated in high-risk areas, thereby intensifying the consequences of flood disasters [2,41,44,50,51,52,53]. This study explores the spatial patterns, types, and influencing factors of flood disasters and socio-spatial differentiation through the lens of environmental justice. Previous studies have primarily focused on urban flood disaster events, underestimated the environmental justice aspects of flood disaster risk, and neglected the correlation between flood disaster risk and regional population characteristics. Current research tends to focus on both macro and fine-scale spatial patterns of urban flooding.

A clear positive correlation exists between social media activity during floods and changes in rainfall. By integrating social media data with flood susceptibility simulations, a Flood Risk Awareness (FRA) Index is constructed, revealing that the spatial pattern of flood risk awareness is primarily concentrated along the Yangtze River, Shijiu Lake, and other water bodies. Social media activity in urban centers is significantly higher than in exurban areas. Social media activity during flood events is influenced by a combination of natural geographical features, socio-economic attributes, and demographic factors.

The social space differentiation index is constructed, and principal component analysis is employed to analyze flood disaster response capacity. The five principal components include the rural population, the elderly and women, the highly educated population, children and immigrants, and family size. In general, Nanjing’s flood disaster coping capacity exhibits a spatial pattern of being high in urban centers and low in exurban areas, indicating that flood coping capacity is closely related to socio-economic factors such as population age structure, education level, and household size. This study calculates the Flood Response Capability Index by summing flood risk awareness and coping capacity using weighted values. Flood response capability is classified into four types: high risk awareness–high coping capacity (H-H), low risk awareness–low coping capacity (L-L), high risk awareness–low coping capacity (H-L), and low risk awareness–high coping capacity (L-H).

We employ multiple regression analysis to explore the influencing factors of socio-spatial differentiation in flood disasters. The dependent variables include the Flood Risk Awareness Index, Flood Coping Capacity Index, and Flood Response Capability Index, while natural geographical conditions, population, and economic indicators serve as independent variables. The main influencing factors of flood risk awareness include the density of flooding points, average slope, average elevation, construction land ratio, density of supermarkets, and density of parks and squares. The main factors affecting flood coping capacity include vegetation coverage, average slope, average elevation, construction land ratio, the ratio of highly educated populations, and business density. The main factors influencing flood response capability include the density of flooding points, average slope, average elevation, construction land ratio, the ratio of highly educated populations, density of supermarkets, and density of parks and squares. The density of flooding points significantly affects flood risk awareness, while areas with a higher ratio of educated populations tend to engage with social media more frequently.

The study found that areas with low risk awareness and low coping capacity are predominantly located in exurban areas. These regions often lack both risk awareness and public service facilities, particularly near large lakes (such as Shijiu Lake) in exurban areas, where several flood disasters have occurred, resulting in significant losses for residents. The prevention of flood susceptibility in exurban areas not only improves the accessibility of public services but also enhances residents’ risk awareness and information acquisition. There exists a spatial overlap between flood disasters and social differentiation, rendering regions with concentrated vulnerable groups more susceptible to disaster risks. By combining real-time social media data with census data, we can enhance flood response capabilities and strengthen community resilience.