Urban Flood Resilience Assessment of Zhengzhou Considering Social Equity and Human Awareness

Abstract

Flooding is one of the world’s most devastating natural disasters, and the effects of global climate change further intensify its impact. In the context of flood management, urban resilience emerges as a promising perspective. While existing urban resilience assessment systems predominantly encompass economic, social, ecological, and infrastructural factors, they often neglect crucial dimensions like social equity and human awareness. We aimed to assess urban flood resilience considering social equity and human awareness. We have developed an indicator system called the 3-Dimentional Disaster Urban Flood Resilience Index System (3D-UFRIS) to address the issue. We also introduced social media data to explore the use of big data in urban flood resilience assessment. Scrapy was used to collect data and AHP-EWM was used to calculate the results. Our findings reveal a layered distribution of urban flood resilience of Zhengzhou, categorized into five levels: highest, higher, medium, lower, and lowest resilience. Notably, the highest resilience areas, covering a mere 3.06% of the total area, were primarily situated in the Jinshui district, characterized by strong economic activity, high public awareness, and a history of waterlogging incidents. Conversely, the lowest resilience areas, encompassing the largest portion at 36%, were identified in Zhongmou County, Xinzheng, and Shangjie District, marked by lower public awareness and limited medical accessibility. This study presents a pioneering approach to comprehending urban disaster resilience, offering valuable insights into mitigating flood-related risks and enhancing urban planning strategies.

1. Introduction

Over the past few years, the escalating effects of global climate change have amplified the severity of flooding disasters [1]. This has resulted in profound consequences for people’s livelihoods and the progress of society and the economy. Consequently, the enhancement of urban preparedness, resilience, and recovery has emerged as a prominent subject within disaster management discourse. Amidst the backdrop of the ever-evolving landscape of social technology [2], research into resilience has significantly shaped our understanding of the concept and introduced a fresh outlook on urban flood management, garnering considerable interest [3,4].

The term “resilience” originates from the Latin word ”resilio”, which originally meant to bounce back to the original state [5,6]. The introduction of resilience theory in urban studies was primarily aimed at urban disaster prevention. In 1990, the integration of resilience theory with urban systems marked its inception within the realm of urban planning and development [7,8,9]. This convergence broadened the horizons of urban disaster research [10]. At that time, resilience emphasized the city’s ability to withstand and recover from disasters, maintaining its original state. Initially, urban resilience referred to the city’s ability maintaining its original state when faced with shocks and disturbances, including the resistance level during changes and the efficiency of restoring the system’s original structure. The concept of urban resilience has gained increasing attention, providing a new perspective for addressing urban issues. Subsequently, scholars worldwide embarked on comprehensive investigations, delving into the intricacies of the concept, its associated themes, and the domains encompassing disaster resilience. They also explored various assessment methodologies to bolster disaster resilience and fortify the foundation of urban resilience construction.

As a complex and comprehensive system, urban resilience is also intricate, requiring a comprehensive concept that can measure the overall characteristics of a city. When confronting the intricate challenges of urban environments, the overarching concept of urban resilience often lacks specificity. Consequently, scholars have segmented resilience into distinct types, tailoring it to address particular issues and forming a conceptual framework encompassing various resilience concepts, such as urban disaster resilience. Urban disaster resilience [11,12,13,14] is characterized by a city’s comprehensive capability to avert disasters and mitigate their impacts. Within the realm of resilience, flood resilience emerges as a more precise category. Urban flood resilience [15], in particular, pertains to a city’s capacity to withstand flood-related disasters, encompassing its ability to preserve its existing state and adapt to impending calamities.

The most common method used for evaluating urban flood resilience is a multicriteria indicator evaluation method [16,17]. The first step of this method is to create a reasonable evaluation indicator system. After summarizing relevant research on urban flood resilience assessment systems, we found that evaluation systems can be divided into three categories based on the selection of evaluation indicators: evaluation based on the basic components of the city [18], evaluation based on the characteristics of urban flood resilience [19,20,21,22,23,24], and evaluation based on the stages of urban flood resilience [4,25,26].

After meticulously compiling and evaluating the available body of research on indicators for assessing urban flood resilience [25,26,27,28,29,30], a discernible trend becomes apparent. The prevailing indicator systems overwhelmingly prioritize economic, societal, ecological, and infrastructural aspects, frequently overlooking the essential roles of social equity and human awareness within the framework of a comprehensive urban resilience assessment. Consequently, there exists an urgent imperative for the refinement of evaluation systems tailored to measure the resilience of urban centers. Moreover, in previous studies on disaster resilience, the assessment data mainly relied on historical yearbook data. However, these data have limitations as they lack spatial information, and their temporal persistence and dynamics are low. With the emergence of big data, there is an urgent need to collect spatiotemporal data from multiple sources and establish a scientific indicator system for resilience research. This remains a pressing challenge that needs to be addressed.

To address these limitations, we have undertaken efforts to significantly enhance the existing indicator system. Our improvements are structured across three essential dimensions: disaster-causing factors, disaster-resistance factors, and disaster-bearing factors. To consider public awareness and social equality, we have incorporated the density of flood-prone areas as a crucial component within disaster-causing factors. In evaluating the disaster-resistance factor, we have introduced the consideration of public perception to gauge residents’ responses to disasters. Furthermore, in assessing the disaster-bearing factor, we have integrated the accessibility of medical treatment to underscore the critical importance of medical response time. Our endeavors to refine the indicator data encompass the utilization of diverse data sources, including social media data and point-of-interest (POI) data, to the fullest extent possible. These measures aim to fortify the comprehensiveness and effectiveness of our enhanced indicator system.

After determining the assessment indicators systems, the next step is to determine the index weight. Now, there are three main ways: the subjective empirical method [9,31,32], objective analysis method [33,34], and subjective–objective combined method [35,36].

Comparing the three methods, the subjective–objective combined method can lead to a more reasonable evaluation, so we try to combine the analytic hierarchy process (AHP) and entropy weight method (EWM) to calculate index weight. This helps alleviate the problems of incomplete expert evaluation information and subjective preferences in AHP.

This study introduces a novel urban flood resilience assessment method, built upon the development of an improved urban resilience indicator system. To address the limitations of existing flood resilience indicators, we established a new evaluation indicator system considering social equity and public awareness. We also used social media data to extract non-quantifiable indicators including public perception and flood-prone areas. We proceeded to integrate the AHP and EWM to compute the weight assigned to each index. This integration effectively addresses issues related to incomplete expert evaluation data and subjective preferences that often arise in AHP. To validate the efficacy of this innovative methodology, we conducted an evaluation using Zhengzhou as a case study, rigorously assessing its performance against other existing evaluation approaches.

The subsequent sections of this paper provide an overview of the research area and data, followed by a comprehensive explanation of the research methodology, which includes the construction of the index system and the calculation method employed. Finally, we present the study’s results and engage in a comprehensive discussion of our findings.

2. Study Area and Data

2.1. Study Area

The study area was Zhengzhou, China. Zhengzhou, which is situated in the northern and central portions of Henan Province downstream of the Yellow River. It is located between longitude 112°42′ E–114°14′ E and latitude 34°16′ N–34°58′ N. It covers a total area of 7567.18 square kilometers, which includes six districts and five county-level cities. The overall terrain trend of Zhengzhou is high in the southwest and low in the northeast. The geographic location of Zhengzhou is shown in Figure 1.

Following the flood that struck Zhengzhou in 2021, extensive research efforts have been dedicated to understanding flood-related disasters in the region. Nevertheless, the majority of these studies have primarily concentrated on flood monitoring [37] and flood risk assessment [35], with relatively limited emphasis placed on the crucial aspect of flood resilience.

An analysis of precipitation data for Zhengzhou spanning the last two decades reveals a noticeable uptick in the occurrence of heavy rain and subsequent floods in recent years. Furthermore, there is an observable increase in both the frequency and intensity of abnormal precipitation events. These developments have resulted in substantial losses to personal and public property, as well as disruptions to the normal functioning of the city. Delving deeper into the statistics of heavy rainfall over the past decade, it becomes evident that such events predominantly transpire in the months of July and August. Additionally, when examining floods beyond the city limits of Zhengzhou, it is noteworthy that the annual average incidence of severe floods within Zhengzhou itself is approximately three to four times higher. The bulk of these floods occurs during the months of July and August, constituting a staggering 80% of the total flood occurrences throughout the year.

2.2. Data

This study primarily used four types of data: (1) the POI, including hospitals, shops, and hotels, sourced from Amap (http://m.amap.com/ (accessed on 1 January 2023)); (2) remote sensing data; (3) social media data, including Weibo data obtained by keyword crawling and Weibo check-in data during the 2021 Zhengzhou flood; (4) statistical yearbook data, which mainly consisted of administrative unit-based statistical data sourced from the official website of the Zhengzhou Bureau of Statistics. The reason for data selection is shown in Appendix A (

Table A1. Indicator explanation.

Index Name	Explanation
Distance to flood-prone spots	When floods occur, floods are more likely to occur near many flood-prone points. For an area, the closer it is to flood-prone points, the greater the probability of floods.
DEM	Generally speaking, the greater the absolute elevation, the less likely this area will be to flood due to its high terrain and the stronger its resilience.
Impervious surface ratio	When flooding occurs, the higher the impervious surface ratio of an area, the worse the flood drainage capacity, the greater the probability of flooding, and the lower the area’s resilience level.
Short-term rainfall	The necessary condition for the occurrence of flood disasters is rainfall, which is an important manifestation of the risk of disaster factors.
Annual average rainfall	The necessary condition for the occurrence of flood disasters is rainfall, which is an important manifestation of the risk of disaster factors.
Runoff coefficient	The larger the runoff coefficient is, it means that regional rainfall is less likely to be absorbed by the soil, which will increase the load on drainage ditches. The regional flood drainage capacity is worse, and the regional resilience level is negatively correlated with the runoff coefficient.
Public perception index	The higher the public perception index, it means that the public in the administrative region is more sensitive to flood-related information and has stronger dissemination and reception capabilities, so the regional resilience is higher.
River density	The greater the density of the river network, the greater the possibility that the river will cause flood disasters, which will reduce the resilience of the region.
Road density	Road density is the length of roads per unit area of a region. The greater the road density, the more paths there are for disaster avoidance and the higher the resilience of the region.
Hotel density	After a flood disaster, hotels can provide temporary residence for the affected people, so the density of the hotel affects the level of resilience. The greater the density of the hotel, the higher the level of resilience.
Number of Internet users	Communication capability refers to a region’s ability to transmit and receive information when a flood disaster occurs. It is represented by the density of the number of mobile phones and the number of Internet users. The stronger the communication capability, the stronger the ability to receive and transmit information when a flood disaster occurs, and the stronger the city’s flood resilience.
Number of cell phone users	Communication capability refers to a region’s ability to transmit and receive information when a flood disaster occurs. It is represented by the density of the number of mobile phones and the number of Internet users. The stronger the communication capability, the stronger the ability to receive and transmit information when a flood disaster occurs, and the stronger the city’s flood resilience.
Flooding standard	The higher the drainage standards, the less likely floods will occur and the higher the resilience of the area.
Population density	Population density is the number of people living in a unit area. Population density reflects the degree of population exposure in a city, indicating the number of people living in a city per unit area who may be affected by floods. The greater the population density in a region, the more people will be affected by floods, and the lower the resilience level of the region.
GDP	GDP is often used as an indicator to evaluate the economic level of a region. GDP per unit area is the ratio of gross domestic product to total area achieved within a year. The higher the value, the greater the total value generated per unit area, and the economic response to disasters. The higher the capability, the higher the resilience level of the area.
Shop density	The indicator used for the number of large retail stores is the number of large retail stores in each administrative district. The larger the value, the more food and daily necessities the administrative area can provide when a disaster occurs, and the higher the regional resilience level.
Medical sources accessibility	The accessibility of medical resources must not only consider the number of hospital beds and medical staff, but also other factors that affect the ease of individuals obtaining necessary medical services, such as distribution, quantity, quality, cost, and transportation. It is a more comprehensive indicator of the availability and convenience of medical services in a region.
Percentage of population with education level below elementary school	Education level will affect the public’s ability to respond to flood disasters, self-rescue ability, and judgment ability to a certain extent. The proportion of non-illiterate population in each region is used to represent the education level. The higher the proportion, the lower the education level, and the public’s ability to respond to flood disasters. The worse the response ability, the lower the resilience level of the disaster-bearing factor dimension. On the contrary, the higher the proportion, the better the public response ability, and the higher the resilience level of the disaster-bearing factor dimension.
Percentage of population over 60 years old and children	The elderly and children are most vulnerable to threats and injuries from floods due to their weak mobility. The small proportion of elderly people and children means that the city’s social resilience is stronger.

).

Table 1. Statistical table of data.

Data Name	Spatial Resolution (m)	Data Sources
River density	1000	https://www.resdc.cn/ (accessed on 1 January 2023)
DEM	30	https://www.resdc.cn/ (accessed on 1 January 2023)
Runoff coefficient	1000	https://www.resdc.cn/ (accessed on 5 January 2023)
GDP	1000	https://www.resdc.cn/ (accessed on 5 January 2023)
Population density	1000	https://www.resdc.cn/ (accessed on 5 January 2023)
Short-time rainfall	1000	University of California, San Diego/CHG
Historical rainfall data	1000	University of California, San Diego/CHG
Impervious surface ratio	1000	Extract based on Lansat.
Density of flood-prone spots	1000	https://weibo.cn/ (accessed on 1 March 2023)
Public perception index	1000	https://weibo.cn/ (accessed on 1 March 2023)
Number of Internet users	1000	Zhengzhou Statistical Yearbook
Number of phone users.	1000	Zhengzhou Statistical Yearbook
Percentage of people with only primary-level education	1000	Zhengzhou Statistical Yearbook
The proportion of the population over 60 years old and children	1000	Zhengzhou Statistical Yearbook
Hotel density	1000	Amap Point of Interest (http://m.amap.com/ (accessed on 1 January 2023))
Shop density	1000	Amap Point of Interest (http://m.amap.com/ (accessed on 1 January 2023))
Medical treatment accessibility	1000	Extract based on street and hospital data from Amap (http://m.amap.com/ (accessed on 1 January 2023)) and https://www.zgylbx.com/ (accessed on 1 January 2023)
Road density	1000	OSM

provides a comprehensive overview of these data sources. The reason for indictor selection is shown in Appendix A (Table A1).

Crawlers play a crucial role in collecting both POI and social media data. The POI information was sourced directly from Amap, while the social media data were extracted from Weibo using two specialized crawlers—one designed for keyword-based data retrieval and the other for capturing check-in data. These crawlers were implemented within the Scrapy framework, serving as the backbone for the data collection process.

3. Methodology

3.1. Framework

The framework of our research mainly included two parts.

Firstly, regarding the important indicators that are currently being ignored in the evaluation indicators systems, like the density of flood-prone areas, public perception, and medical treatment time, we improved the indicators systems from three dimensions: causing factors, resistance factors, and bearing factors. This led to the development of a novel indicator system referred to as the 3D-UFRIS.

Secondly, building upon the foundation of the 3D-UFRIS, we employed a combination of the AHP and the EWM to assign weights to each of the indicators. This approach addresses challenges related to incomplete expert evaluations and subjective biases within AHP, resulting in a more robust and balanced assessment.

3.2. Construction of 3D-UFRIS

By compiling and evaluating the available research on urban flood resilience assessment indicators, we found current evaluation systems for urban resilience mostly consider economic, social, ecological, and infrastructure factors, but they overlook the importance of social equity and human awareness.

Several significant gaps and limitations are evident in the current research landscape. Firstly, there is a notable lack of attention to the impact of vulnerable flood-prone areas, despite the frequent occurrence of flood disasters in these regions. Secondly, the existing body of research often neglects the critical aspect of resident response due to the challenges associated with quantifying resident response indicators and accessing relevant data.

To address these aforementioned limitations, our study adopted a foundation rooted in disaster system theory, aiming to enhance the existing indicator system substantially. We undertook this enhancement through a comprehensive three-dimensional approach, encompassing causing factors, resistance factors, and bearing factors, resulting in the creation of a novel indicator system referred to as the 3D-UFRIS. Within the realm of causing factors, we introduced the density of flood-prone areas, acknowledging their critical impact. For the disaster-resistance factor, we incorporated the evaluation of residents’ responses to disasters, specifically through the lens of public perception. In the domain of disaster-bearing factors, we placed emphasis on the accessibility of medical treatment to emphasize the significance of timely medical assistance.

Furthermore, we ensured that our indicator data were enriched with real-time sources such as social media data and POI data, enhancing the timeliness and relevance of our assessments. Building upon these improvements and drawing insights from both domestic and international indicator systems, we have constructed a comprehensive flood resilience evaluation system for cities. This system encompasses three evaluation levels and comprises 19 evaluation indicators, enabling a more holistic assessment of urban flood resilience.

The development of the 3D-UFRIS is depicted in Figure 2.

3.3. Quantification of the Resilience Assessment Index

In this study, we extracted the density of flood-prone areas based on Weibo data and extracted medical accessibility based on Internet Map APIs. Other indexes were extracted by ArcGIS, these indexes were shown in Figure 3. All the data were processed into uniform 1 km × 1 km grid cells.

3.3.1. Extracting Distance to Flood-Prone Areas Based on Weibo Data

We extracted flood-prone areas based on Weibo data, and the flowchart outlining this process comprises six primary steps as follows:

Step 1. Collect data.

We utilized Scrapy for data collection, focusing on keywords associated with flood disasters, such as “stagnant water” and “flooding.” Our efforts resulted in the crawling of 32,111 Weibo data posts within the time frame spanning from 18 July to 21 July.

Step 2. Clean data.

To clean the data, we implemented a process to eliminate redundant tweets, special characters, and hyperlinks within the tweets. Additionally, our study categorized the relevant text data into two groups: one containing waterlogging point information and the other without such information. This classification was achieved through the application of the RNN binary classification algorithm. We used the RNN binary classification method to train on a selected set of 400 samples. Among these, 200 samples were flood-related microblogs, such as “It’s raining heavily, and I’m flooded,” which were considered positive samples. The remaining 200 samples were unrelated microblogs, such as “XXX’s house is flooded tonight,” which were considered negative samples. In the end, a total of 3161 check-in records were identified as microblogs related to heavy rain and floods.

Step 3. Chinese word segmentation.

We opted to segment the urban flood-related microblog text using the Jieba text segmentation tool.

Step 4. Stop word removal.

The stop word list that is frequently utilized by Baidu, HIT, and the machine intelligence laboratory of Sichuan University was integrated into this research together with a unique stop word in light of the dispersed text content and limited attributes relevant to urban flood situations.

Step 5. Extracting Flood-Related Weibo Posts Based on textRank.

From classified text data, entities concerning waterlogging points were extracted using the TextRank [38] entity extraction approach.

Step 6. Geocoding.

By retrieving the geographic coordinates from Amap API, the longitude and latitude coordinates for the extracted geographic entity may be obtained.

Between 18 July and 21 July, a total of 32,111 microblog posts were identified using keywords such as “flood” and “waterlogging.” Through a model analysis and text comparison of the original content, we were able to extract 201 geographic names and 165 locations affected by waterlogging in Zhengzhou. Interestingly, there were 69 points that overlapped between the two categories, as depicted in Figure 4 (some locations with similar distances are represented as the same point in the figure). This finding aligns with the information provided by the Zhengzhou Traffic Police Public Number, which listed 86 waterlogging sites within the Zhengzhou Area. Consequently, the overall coverage rate of microblog-reported waterlogging locations compared to official waterlogging sites stands at a remarkable 80.23%.

Figure 4a illustrates the spatial distribution of flood-prone areas in Zhengzhou. The red dots represent flood-prone areas identified based on the data extracted from Weibo, while the green dots represent officially announced flood-prone areas. The blue dots indicate the overlapping areas between the flood-prone areas identified by us and those officially announced. Flood-prone areas identified through social media sources complement the officially announced ones, offering valuable supplementary information. These areas are predominantly situated in the central urban areas and are concentrated in low-lying zones, including subway stations, bridges, and tunnels.

The impact of flood-prone areas extends beyond a single point, encompassing a broader geographical area. To comprehensively assess the influence of these areas, this study utilized the distance from the flood-prone locations as a crucial parameter. To achieve this, the ArcGIS Pro buffer tool (found under Analysis Tools->Multiple Ring Buffer) was employed to create buffer zones at intervals of 500 m, 1000 m, 1500 m, and 2000 m from the flood-prone areas. We think that the areas within a 500 m radius of flood-prone points are equally affected, as well as the areas within a range from 500 m to 1000 m, 1000 m to 1500 m, and 1500 m to 2000 m. Based on the distance of the buffer zone, we divide them into four levels. The outcome of this process is depicted in Figure 4b.

3.3.2. Extracting the Public Perception Index Based on Weibo Check-in Data

Flood events can significantly disrupt residents’ daily lives and mobility. In response to these challenges, a considerable number of individuals turn to social media platforms to share their emotions, challenges, and information related to disasters. Social media has emerged as a crucial source for monitoring shifts in public sentiment and facilitating disaster response efforts [39,40,41,42,43,44].

Refer to Figure A2 for a visual representation of the daily variations in Weibo check-ins. The heavy rainstorm event labeled as “7.20” refers to the severe rainfall experienced on July 20th and 21st. During this period, there was a substantial surge in Weibo check-ins compared to regular days. The fluctuations in social media activity correspond closely to the prevailing weather conditions. Notably, on July 20th and 21st, 2021, when the city witnessed unprecedented torrential rainfall and widespread flooding, Weibo posts addressing rainstorm-related issues and flooding reached levels significantly higher than usual. This indicates a direct correlation between the temporal patterns of Weibo data and the concurrent weather conditions. Specifically, residents’ perception of flood disasters appears to be closely tied to the duration and intensity of rainfall, with shifts in perception occurring in response to the severity of the calamity.

The “public perception index” reflects the collective perspective of the general population regarding disaster-related information. It quantifies this perception by calculating the ratio of microblogs discussing rainstorms and floods within a specific administrative region to the total number of registered microblogs. This index is defined by Equation (1).

$$\begin{array}{l}public perception =\frac{number of flood-related Weibo}{number of Weibo checked in}\times 100\%\\ \end{array}$$

(1)

where the “number Weibo checked in” is the total number of registered microblogs of the area, the “number of flood-related Weibo” is number of microblogs discussing rainstorms and floods of the specific area.

In this study, the public perception index for Zhengzhou was extracted from Weibo check-in data. The extraction process can be summarized in four main steps.

Step 1. Collect data.

This study utilized the Scrapy framework to acquire Weibo check-in data for Zhengzhou during the research period. The obtained data includes text information, published latitude and longitude coordinates, as well as publication timestamps. A total of 20,160 check-in data points were collected for the flood month of 2021, spanning from July 1st to July 31st, in Zhengzhou.

Step 2. Clean data.

To ensure the quality of the check-in data and address limitations such as sampling bias, location inconsistencies, and the presence of “noise,” several criteria were applied for data cleaning. Check-ins that were located outside of Zhengzhou were removed from the dataset. Additionally, users who checked in multiple times at the same location within a half-hour timeframe were considered as a single check-in record. These data-cleaning measures aimed to enhance the accuracy and reliability of the check-in data for subsequent analysis.

Step 3. Identification of rainstorms and flood.

To train and select the dataset for analysis, the RNN binary classification method was employed. A total of 400 corpora were chosen, comprising 200 positive corpora related to floods, such as “It rained so heavily that it flooded,” and 200 negative corpora unrelated to floods, like “There was a flood in xxx household tonight.” Following this classification, a final dataset of 2004 check-in records was identified as Weibo data related to rainstorms and floods for further analysis.

Step 4. Calculation of public perception index.

The calculation is based on the formula of the public perception index.

Figure 5 shows the distribution of the public perception index in Zhengzhou.

3.3.3. Calculating Medical Accessibility Based on Internet Map APIs

Efficient and timely medical treatment is of paramount importance for urban residents in the aftermath of a flood disaster. Previous analyses of urban resilience primarily relied on metrics like the number of hospital beds and the number of physicians to gauge the ability to access medical care. However, these metrics only addressed one facet of post-disaster relief: the availability of beds. In reality, the time it takes to receive medical attention is a critical factor, and for the injured, time is life. To address this crucial aspect, this study incorporated medical accessibility as an indicator to evaluate the level of medical resources for resilience analysis. Accessibility is a pivotal measure of equity and efficiency within a healthcare system, signifying how easily and swiftly individuals or community members can access appropriate healthcare services.

The steps of the analysis of medical accessibility in Zhengzhou are as follows:

Step 1. Collect data.

Collect data on hospital and street center points. The information on the hospital was collected from the China Medical Insurance Network, and the position of the street and hospital was collected from Amap.

Step 2. Time calculation.

Use Amap API to calculate the travel time between the origin and destination based on the mode of driving and public transportation.

Step 3. Accessibility calculation.

Choose the Gaussian-based 2SFCA method as the accessibility calculation method.

The formula of Gaussian-based 2SFCA methods is as follows:

Calculate the supply-demand ratio $R_j$ for each hospital j using Equations (2) and (3).

$$R_j=\frac{S_j}{{\displaystyle \sum _{k\in dkj<d0}G(d_{kj},d_0)D_k}}$$

(2)

$$G\left(d_{kj},d_0\right)=\left\{\begin{array}{c}\frac{e^{-\frac{1}{2}\times \frac{dkj}{d0}}-e^{-\frac{1}{2}}}{1-e^{-\frac{1}{2}}},dkj\le d_0\\ 0,dkj\ge d_0\end{array}\right.$$

(3)

where $S_j$ is the service capacity of hospital j, i.e., the number of beds; $D_k$ is the population within the service search range of the street; $dkj$ represents the travel time between each street k and hospital j; $G(d_{kj},d_0)$ represents the weight of the service capacity of hospital k.

Calculate the accessibility by Equation (4).

$$A_i={\displaystyle {\sum }_{jϵ\left(d_{kj}\le d_0\right)dkj\text{¡}\le d_0}^{n}G(d_{ij},d_0)\times R_j}$$

(4)

Finally, the results of public transportation and driving were weighted equally (50% each) to obtain the final evaluation index of medical accessibility in Zhengzhou.

Step 4. Accessibility grading.

We categorized accessibility into four levels: first-level accessibility for values below 3, second-level for 3–10, third-level for 10–70, and fourth-level for values above 70. The higher the level, the better the accessibility. The first level indicates untimely medical treatment, the second level indicates basic timeliness, the third level indicates good timeliness, and the fourth level indicates excellent timeliness. Figure 6 shows the distribution of accessibility in Zhengzhou.

3.4. Normalized Data

As this indicator system encompasses both positive and negative indicators, we opted for the extreme value method (differential standardization) to standardize the raw data.

Equation (5) represents the formula for positive indicators.

$$Y_i =\frac{x_i-Min{x}_i}{Max{x}_i-Min{x}_i}$$

(5)

Equation (6) represents the formula for negative indicators.

$$Y_i =\frac{Max{x}_i-x_i}{Max{x}_i-Min{x}_i}$$

(6)

where x_i is the raw data for indicators, and Y_i is the normalized processed data.

3.5. Weighting

The integration of AHP and EWM for calculating index weights alleviates issues related to incomplete expert evaluation information and subjective preferences within AHP.

3.5.1. Analytic Hierarchy Process

AHP, a multi-level decision-making model employed to assess and compare multiple options, is integral in arriving at a final decision. This method places significant emphasis on the insights and expertise of experts to gauge the relative value of the available options. The steps involved in AHP are as follows:

(1)

Define the problem and identify the overall goal:

The target level is the urban flood resilience index, and the criterion level includes disaster-causing factors, disaster-resistance factors, and disaster-bearing factors, while the indicator level is the specific indicator.

(2)

Pairwise comparisons:

The importance of each criterion Bi to objective A and the importance of each index $C_{ij}$ to criterion Bi were compared in pairs. The importance rating criteria were used in the 1–9 scale method to obtain the comparison matrix W.

(3)

Consistency check.

To check whether the weights are reasonably assigned, a consistency check is required for the comparison matrix. Equation (7) represents the formula for positive indicators.

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

(7)

where CR is the random consistency ratio of the comparison matrix, CI is the general consistency index of the comparison matrix, $CI=\frac{{\mathsf{\lambda }}_{\mathrm{m}\mathrm{a}\mathrm{x}}-n}{n-1}$, RI is the average stochastic consistency index of the comparison matrix.

If the CR value is less than 0.1, it can be considered that the consistency of the judgment matrix is acceptable. Otherwise, the judgment matrix needs to be revised.

(4)

Calculate the indicator weight vector.

Calculate the unit eigenvector w corresponding to the largest eigenvalue of the comparison matrix W, and each component of w is the importance order of each evaluation factor.

3.5.2. Entropy Weight Method

(1)

Calculate the entropy of each criterion by Equations (8)–(10).

$$E\left(j\right)=-{\displaystyle \sum _{i=1}^{n}p}\left(i,j\right)\ln \left(p\left(i,j\right)\right)$$

(8)

$$ES=-{\displaystyle \sum _{j=1}^{m}w}\left(j\right)\ln \left(w\left(j\right)\right)$$

(9)

$$DE=1-ES$$

(10)

where E(j) is the entropy of the jth criterion, $p\left(i,j\right)$ is the normalized value of the ith alternative for the jth criterion, ES is the entropy of the system, and ln is the natural logarithm, DE is the degree of entropy.

(2)

Calculate the weight of each criterion by Equation (11):

$$w2\left(j\right)=\frac{1-E\left(j\right)}{{\displaystyle \sum _{k=1}^m\left(1-E\left(k\right)\right)}}$$

(11)

where w2(j) is the weight of the jth criterion, and E(j) is the entropy of the jth criterion.

3.5.3. Comprehensive Weight for AHP and EWM

To facilitate calculations while considering both methods, based on existing studies, we consider them to be equally informative [9,36,45,46,47,48]. The weight is calculated by Equation (12).

$$W_j=\frac{w1\left(j\right)+w2\left(j\right)}{2}$$

(12)

where w1(j) is the weight of AHP and w2(j) is the weight of EWM.

The weight of the index is shown in

Table 2. Weight of the index.

Dimension	Index Name	AHP Weight	EWM Weight	Comprehensive Weight
Disaster-causing factor	Distance to flood-prone spots	0.0522	0.0691	0.06065
	DEM	0.0776	0.1013	0.08945
	Impervious surface ratio	0.0625	0.0463	0.0544
	Short-term rainfall	0.0669	0.0112	0.03905
	Annual average rainfall	0.0433	0.0236	0.03345
	Runoff coefficient	0.0705	0.0596	0.06505
Disaster-resistance factor	Public perception index	0.0465	0.0636	0.05505
	River density	0.0639	0.1163	0.0901
	Road density	0.0173	0.0728	0.04505
	Hotel density	0.0939	0.0776	0.08575
	Number of Internet users	0.0618	0.0533	0.05755
	Number of cell phone users	0.0416	0.018	0.0298
	Flooding standard	0.0677	0.0446	0.05615
Disaster-bearing factor	Population density	0.0685	0.0616	0.06505
	GDP	0.0326	0.0382	0.0354
	Shop density	0.0349	0.0489	0.0419
	Medical sources accessibility	0.0411	0.0718	0.05645
	Percentage of population with education level below elementary school	0.0205	0.0176	0.01905
	Percentage of population over 60 years old and children	0.0347	0.0236	0.02915

.

3.6. Urban Flood Resilience Index

The urban flood resilience of each dimension is calculated by

$$r={\displaystyle \sum _{j=1}^n{W}_j}{D}_{ij}$$

(13)

where r represents the resilience index of the primary indicator of flooding, $W_j$ represents the weighting of each indicator on urban flood resilience, and $D_{ij}$ represents the standardized data of each indicator.

$$R=r_{DCF}{w}_{DCF}+r_{DBF}{w}_{DBF}+r_{DRF}{w}_{DRF}$$

(14)

where R represents the urban flood resilience index. The higher the R value, the more resilient the city is to floods. The weighting coefficient for each dimension is represented by w, while $r_{DCF}$, $r_{DRF}$, and $r_{DBF}$ denote the resilience value of the disaster-causing factor, disaster-resistance, and disaster-bearing factor dimensions.

4. Results

4.1. Disaster-Causing Performance Assessment

The disaster-causing factor is considered a negative index, which means that lower values are preferable. Consequently, it exhibits a negative distribution. Figure 7 illustrates the distribution of disaster-causing performance. The central city area exhibits the highest disaster-causing performance, primarily due to its relatively flat terrain. The changes in the high range are minimal. From a terrain perspective, the central urban area consists mainly of construction and residential land with high building density and a high flow coefficient. Additionally, the impermeable nature of the soil surface makes it difficult for rainwater to infiltrate, resulting in a dense network of flood-prone points and an increased probability of flood disasters.

4.2. Disaster-Resistance Performance Assessment

Disaster resistance reflects a region’s capacity to withstand flood disasters, which encompasses both infrastructure development and residents’ ability to manage such disasters. Figure 8 demonstrates that the central areas of Zhongyuan District and Jinshui District exhibit relatively strong disaster resistance. In the urbanization process, these economically developed regions in Zhengzhou have experienced robust infrastructure development, fewer heavily polluting enterprises, and a reduced risk of secondary disasters. In terms of residents’ awareness and responsiveness to flood disasters, Zhongyuan District residents demonstrate the highest level of perceptiveness, indicating their heightened readiness to respond to disaster situations.

4.3. Disaster-Bearing Performance Assessment

Disaster recovery capability primarily reflects an area’s capacity to endure flood disasters and manage losses caused by such disasters, and stronger economic regions tend to exhibit greater resilience to both. As illustrated in Figure 9, most disaster-bearing performance indicators are based on administrative divisions, resulting in clear administrative boundaries in the figure. Administrative regions like Jinshui District and Zhongyuan District, characterized by robust economic strength, boast favorable medical conditions, predominantly secondary and tertiary economic sectors, abundant living security facilities, higher per capita disposable incomes, and elevated cultural levels, contributing to their higher resilience levels. In contrast, other administrative regions primarily consist of townships, marked by lower population densities, reduced cultural development, pronounced aging, weaker economic growth, and limited public awareness of floods, resulting in lower resilience levels.

4.4. Urban Flood Resilience Assessment

The outcomes of the urban flood resilience evaluation for Zhengzhou are visualized in Figure 10a. It is evident from the figure that the central region of Zhengzhou, constituting the main urban area, exhibits the highest resilience, followed by Dengfeng. Conversely, Xinzheng, and Zhongmou County displaying the lowest resilience levels, primarily attributable to their weak disaster-bearing and resistance performances.

Based on the grid-based statistics of urban flood resilience in different areas of Zhengzhou, the urban flood resilience classification for Zhengzhou is presented in Figure 10b. The classification of flood resilience encompasses five distinct levels: highest resilience, higher resilience, medium resilience, lower resilience, and lowest resilience. The highest resilience category accounts for a mere 3.06% of the total area and is primarily concentrated in Jinshui district. This area is characterized by high economic activity, strong public awareness, and a history of severe waterlogging incidents. The higher resilience category covers approximately 11.75% of the area and is primarily located in Zhongyuan District, Erqi District, and Guancheng District. These regions exhibit high population density, extensive urban development, strong public awareness, and good medical accessibility. In contrast, the lowest resilience areas, comprising the largest portion at 36%, are situated in Zhongmou County, Xinzheng, and Shangjie District. These areas are marked by low public awareness and limited medical accessibility.

The occurrence of the “7.20” torrential rain in Zhengzhou, it has garnered the attention of numerous scholars. However, most of the focus has been on the assessment and extraction of flood risks, with little research dedicated to the urban flood resilience. Evaluating the city’s flood resilience following major disasters contributes to the subsequent recovery and development of the urban area. This study assesses the resilience of Zhengzhou, with a particular emphasis on public response and social equity. The evaluation results encompass a human-centered resilience, which prioritizes the needs of individuals and recognizes their significance within the urban context. These findings provide a reference for the government to enhance and plan the flood resilience of the city, ultimately constructing a people-centric resilient urban environment.

5. Discussion

In the study, we have constructed an indictor system 3D-UFRIS considering social equity and human awareness and introduced social media data, based on 3D-UFRIS, we assessed the urban resilience of Zhengzhou. We found the highest resilience covered only 3.06% of the area and was mainly located in the Jinshui district, with high economic and public perception and serious waterlogging. The higher resilience covered 11.75% of the area and was mainly located in Zhong-yuan District, Erqi District, and Guancheng District, with dense population and buildings, with high public perception and medical accessibility. The lowest resilience areas covered the largest area (36%) and were found in Zhongmou County, Xinzheng City, and Shangjie District, with low public perception and medical accessibility.

5.1. Construction of Urban Flood Resilience Indictor System

Compared to other studies [33,49], this indictor system we constructed considered social equity and human awareness, and we used social media data to extract non-quantifiable indicators. We improved the indictor system from three dimensions.

We took the impact of waterlogging prone points into account in disaster-causing factor dimension. Multiple studies have shown that urban flood disasters have a high probability of recurring near flood-prone points [46]. When flood disasters occur, areas near flood-prone points are extremely vulnerable and will repeatedly cause heavy losses. The impact of flood-prone points needs to be included in resilience assessments. The disaster-causing factor dimension introduces the distance from flood-prone points; in order to analyze and judge the city’s potential flood-prone points in addition to the officially announced city’s potential flood-prone points, and more accurately assess the urban waterlogging situation, this study selected the flood-prone points and flood-prone points based on social media data extraction. The union of officially announced flood-prone points replaces the official flood-prone points as the indicator input.

We considered public response to flood disasters in the disaster-resistance factor dimension. To evaluate residents’ response to disasters, the public perception index that characterizes public response indicators was included. The flood public perception index describes residents’ ability to withstand, adapt to, and recover from flood disasters. Some studies have pointed out that residents’ understanding of floods is very important [39,40,41,42,43,44]. Good understanding and perception can allow residents to take effective measures to avoid adverse impacts when floods occur. This study used information on social media to examine public responses to flooding. Other studies have had difficulty measuring how people respond to flooding, and only a few have used surveys. But investigations are expensive and involve few people. Using social media is a more reliable way to find out what the public is thinking.

We incorporated the accessibility of medical resources into the disaster-bearing factor dimension. Most studies only focus on the number of doctors and hospital beds to measure medical capacity but ignore the time. After a flood disaster, it is very important to ensure timely medical treatment for the injured. In addition to the number of beds and the number of medical personnel that have been studied in previous studies, the time to obtain medical resources is also a variable that needs to be considered [50]. Thus, we considered time and incorporated the accessibility of medical resources into the disaster-bearing factor dimension.

The improvement of the indictor system has put forward more detailed requirements and suggestions for cities to strengthen resilience construction, especially for human awareness and social equity.

5.2. Strategies for Zhengzhou to Improve Urban Flood Resilience

Based on the outcomes, it is evident that Zhengzhou’s resilience requires reinforcement. We offer three recommendations to bolster the city’s ability to withstand and recover from flooding events.

Firstly, governmental authorities should implement strategies to decrease the concentration of flood-prone zones. In the main urban area of Zhengzhou, there is a high concentration of flood-prone areas. During the heavy rainfall event on 20 July, the number of drowning victims in the main urban area accounted for 30% of the total in the city. The government needs to enhance its management of flood-prone areas in the main urban area. This could involve transforming transportation zones adjacent to flood-prone regions into permeable surfaces, expanding green spaces in proximity to flood-prone areas, and enhancing drainage infrastructure. The investigation highlights a significant concentration of flood-prone areas within the primary urban region, where casualties are most prevalent, underscoring the urgency for prompt government intervention in these areas.

Secondly, there is a critical need to raise public awareness regarding flood risks. The public perception of flood disasters in Zhengzhou City is most favorable in the Central Plains region. However, as we move away from the Central Plains, the public perception of flood disasters is relatively poor, and the enthusiasm for disaster response is not high. The government needs to pay great attention to this issue and take measures to enhance the public perception of flood disasters in these areas. This can be achieved through educational initiatives and intensified promotional campaigns. Residents should be actively encouraged to engage in flood prevention and disaster mitigation efforts, and the development of a robust flood warning system is imperative. Notably, within the mountainous regions of Xingyang, Gongyi, Xinmi, and Dengfeng, Dengfeng demonstrated the highest resilience index, primarily attributed to its remarkably lower casualty count and missing individuals during the “7.20” rainstorm incident. This was further facilitated by proactive leadership that was physically present to coordinate disaster relief efforts in a timely manner.

Thirdly, there is a pressing need to enhance the accessibility of urban healthcare services, particularly during flood-related emergencies. The medical resources of Zhengzhou are unevenly distributed. Medical accessibility in areas outside the main urban area is poor, which cannot guarantee timely medical treatment for injured people after disasters. Medical accessibility construction in areas outside the main urban area should be strengthened to increase the availability of medical resources, such as medical facilities, medical professionals, and pharmaceuticals. Government authorities should also focus on acquiring and promptly updating reliable information pertaining to medical services. Additionally, measures should be implemented to enhance the affordability of medical care, ensuring that during flood disasters, healthcare remains accessible to all, irrespective of their income levels, with a particular emphasis on mitigating the impact on low-income groups.

5.3. Future Direction

This study presents a pioneering approach to comprehending urban disaster resilience, offering valuable insights into mitigating flood-related risks and enhancing urban planning strategies. However, this study also has a few shortcomings. Future research can also be conducted from the following aspects: (1) In terms of utilizing multiple sources of data, besides social media data, it is worth considering incorporating other data sources. The social media data chosen for this study are from Sina Weibo. However, since not all users use Weibo, the representation of the users represented by Weibo data is limited. Subsequent research can consider incorporating data from other platforms, such as Douyin and Twitter. (2) When extracting disaster information using social media, only the textual content in Weibo was taken into account, while other information such as images in social media data was overlooked. Subsequent research can use both textual and visual content from social media to extract flood disaster information.

6. Conclusions

In this study, we introduced a comprehensive multi-level urban flood resilience index system referred to as 3D-UFRIS considering social equity and human awareness. Through the application of 3D-UFRIS, we conducted an in-depth evaluation of Zhengzhou’s flood resilience, ultimately leading to the formulation of strategies aimed at bolstering the city’s resilience. The key contributions of this study can be summarized as follows:

(1)

An urban flood resilience indicator system, 3D-UFRIS, was proposed. On the basis of summarizing the deficiencies of existing research index systems, improvements were made to address these deficiencies. We introduced new ways to measure disaster-causing factors, like flood-prone areas. To assess how well residents respond to disasters, we created a public perception index. We also added a measurement for medical resource accessibility to evaluate whether residents could receive timely medical help during disasters. The improved index system followed the principles of independence, quantification, and spatial quantification and selected 19 indicators from multiple levels, including public perception, economy, infrastructure, society, and ecology, to construct the index system composed of a target layer, criteria layer, and indicator layer.

(2)

The distribution of urban flood resilience in Zhengzhou was evaluated. To evaluate the urban flood resilience of Zhengzhou, we obtained remote sensing image data, social media data, POI data, and historical yearbook data. Flood resilience was classified into five levels: highest resilience, higher resilience, medium resilience, lower resilience, and lowest resilience. The highest resilience covered only 3.06% of the area and was mainly located in the Jinshui district, with high economic and public perception and serious waterlogging. The higher resilience covered 11.75% of the area and was mainly located in Zhongyuan District, Erqi District, and Guancheng District, with dense population and buildings, with high public perception and medical accessibility. The lowest resilience areas covered the largest area (36%) and were found in Zhongmou County, Xinzheng City, and Shangjie District, with low public perception and medical accessibility.

As global climate change continues to intensify, the frequency of meteorologically-induced natural disasters, particularly floods, is on the rise. Learning from both our successes and failures in managing flood disasters can play a crucial role in strengthening a city’s ability to effectively respond to such calamities. Zhengzhou, a sprawling city located in China, is experiencing ongoing economic growth and shares similarities with many other urban centers worldwide. Our aim is for this study conducted in Zhengzhou to provide a valuable reference for similar municipalities, thereby contributing to a global approach in addressing urban flood disasters.