Dynamic vulnerability assessment framework for urban flood disasters considering real-time population changes

Under the influence of global climate change and urbanization, flood disasters caused by extreme weather events have become more frequent (Mukherjee et al. 2022). Flooding disasters have caused loss of life, damage to houses and businesses, and disruption of critical infrastructure systems, which have had a serious impact on urban socio-economic activities. In China alone, over a hundred cities have experienced urban flooding every year since 2006, resulting in losses exceeding 100 billion yuan due to flooding disasters (Jiang et al. 2018; Robinson et al. 2023). On July 21, 2021, severe urban waterlogging occurred in Zhengzhou, China, resulting in 392 deaths and $5.71 billion in economic losses (Zhou et al. 2022, 2023). Therefore, how to achieve refined management of flood disasters and alleviate their adverse effects has become an important subject in urban disaster prevention and reduction work.

Flood vulnerability assessment is an important tool for mitigating floods (Salah & Soufiane 2023), which can provide an important basis for flood management personnel to understand the spatial distribution characteristics of flood disaster vulnerability and strive to reduce expected extreme event losses (Nasiri et al. 2016). Following the IPCC's (Intergovernmental Panel on Climate Change) vulnerability framework (Chang et al. 2021; Thornes 2002), which separately considers exposure, sensitivity, and adaptive capacity, many studies have conducted flood vulnerability assessments at national, regional, and urban scales. Erena & Worku (2019) evaluated 24 indicators that affect flood vulnerability levels from social, economic, and physical perspectives. Based on sampling survey data, they used the flood vulnerability index (FVI) method to evaluate the vulnerability levels of 110 villages in Dire Dawa City. Nasiri et al. (2019) proposed a district-level FVI from economic, environmental, and physical components, and selected 10 indicators to analyze which district of Kuala Lumpur is more susceptible to flood disasters. Han et al. (2020) found a strong correlation between the edge-expanding types of development and flood. Salazar-Briones et al. (2020) developed a vulnerability assessment framework that considered social, economic, and physical indicators. This framework integrated government statistical data on social and economic components and assessed the spatial distribution of vulnerability in 329 urban constituent units of México.

Although these studies considered multiple indicators to evaluate the vulnerability of urban flood disasters, they did not fully examine the impact of different building types on urban flood vulnerability. Buildings are the main body carrying out urban flood disasters, and there are significant differences in the functions, values, and population density of different types of buildings (Lv et al. 2021). For example, in the context of the same level of flood disasters, residents may suffer more severe flood disasters due to the higher exposure of a large population to flood disasters. In fact, there is little evaluation of different types of buildings (residence, industry, agriculture, commerce, public service). In addition, most of these studies calculate the vulnerability of urban flood disasters on a single scale (Afsari et al. 2023; Choe et al. 2023; Machairas & van de Ven 2023). However, due to the different focuses of urban flood vulnerability assessment at different scales, there may be certain differences in vulnerability assessment results. The vulnerability of flood disasters at the administrative district scale (regional scale) reflects the overall characteristics of flood disasters in each region, which can provide a basis for overall disaster prevention and reduction in administrative districts (Nasiri et al. 2019). However, it ignores the disaster vulnerability characteristics within different constituent units. The flood vulnerability at the constituent unit scale (grid scale) can clarify the spatial distribution characteristics of each bearing body and provide an important basis for refined urban flood prevention (Erena & Worku 2019). However, the flood vulnerability analysis at the grid scale ignored the impact of regional hydrological and hydraulic connections, and it is difficult to directly reflect the overall flood disaster vulnerability of each region, which may pose challenges to urban flood joint prevention and control. Therefore, this study attempts to explore a comprehensive calculation method for urban flood vulnerability at different scales and building types, which is one of the innovations of this study.

In addition, previous studies have shown that the impact of real-time population changes on urban flood disasters may be significant. Zhu et al. (2023) demonstrated that the physical vulnerability of pedestrians is closely related to flood events through experiments. Dong et al. (2022) analyzed the changes in hazard degrees of people at nine measurement locations by simulating the process of urban flooding. The results showed that the hazard degrees of people increased sharply after 16:30 pm (the maximum hazard degree occurred at 16:56 pm), which demonstrated that there are differences in the hazard degrees of people in flood disasters at different times. Ritter et al. (2021) found that flooding increased as population density increased. The safety of affected populations should be a top priority to be protected. Therefore, the vulnerability assessment of flood disasters should consider real-time changes in the population of different constituent units. However, most research has not fully examined the impact of real-time population changes on urban flood vulnerability. Most previous flood vulnerability studies have examined the impact of annual population density (Ritter et al. 2021) on flood vulnerability. The urban function and socio-economic attributes determined the significant mobility of the population within different constituent units (Robinson et al. 2023), and real-time change in population may lead to significant changes in flood vulnerability.

Given the above considerations, this study attempts to propose a multi-scale and multi-categorized urban flood dynamic vulnerability assessment method that considers real-time population changes. Firstly, the real-time population thermal data were used to analyze the dynamic changes in the population of various carriers at different times. Based on this, the flood vulnerability assessment framework was established for grid and regional scales. The characteristics of vulnerability changes in residence, industry, agriculture, commerce, public service, and road were quantified.

Twenty-six indicators were selected to evaluate the evolution characteristics of urban flood disaster vulnerability under real-time population changes. Among them, population density, distance from the river network, land-use type (residence, road, public service, commerce, industry, agriculture, green land, body of water, bare land), distance from the flood control center, distance from medical rescue points, distance from fire rescue points, and distance from the shelter were used for grid-scale flood vulnerability assessment. Population density, river network density, the proportion of different land-use types, drainage network density, and total distance to the emergency centers were used for regional scale flood vulnerability assessment. The sources and detailed descriptions of each indicator data were as follows.

Land-use data including the distribution and regional proportion of land-use types. The proportion of each land-use type refers to the proportion of one land-use type's area to the regional area in each administrative region (Supplementary Material 1). Land-use data were obtained by extracting the 0.5 m high spatial resolution map of the Pleiades Satellite in May 2018.

River network data including river network density and distance from the river network. River network density was calculated using the density analysis tools of geographic information systems (GIS). Distance from the river network referred to the distance between the constituent unit and the river network, which was calculated using distance analysis tools of GIS.

Drainage network density: The density of the rainwater drainage network reflected the drainage capacity of the administrative area, which was calculated using the density analysis tool of GIS.

The population density data were obtained by crawling Baidu's real-time thermal map, which collected hourly population thermal data from April 10th to 14th, 2023. Six typical time periods were selected to analyze population dynamics and migration, namely 0:00, 8:00, 10:00, 12:00, 15:00, and 18:00.

Distance from emergency center including distance from flood control centers, distance from medical rescue points, distance from fire rescues, and distance from shelters. The distribution of flood control centers, hospitals, and fire stations was derived from the Point of Interest (POI) data of Baidu Maps in 2019, and the data of shelters were derived from the location of emergency shelters released by the Zhengzhou Emergency Management Center. These indicators were calculated using distance analysis tools of GIS.

In the process of vulnerability assessment for urban flood disasters, there may be significant differences in the characteristics of exposure, sensitivity, and response capacity of urban flood disasters at different assessment scales. The vulnerability assessment at the grid scale can quantify the differences in vulnerability among different bearing bodies, but it is difficult to reflect the impact of hydrological and hydraulic connections within the region on vulnerability assessment. On the contrary, the vulnerability assessment at the regional scale needs to consider the comprehensive state of all bearing bodies within the administrative region, and the hydrological and hydraulic conditions of the administrative region, but often ignores the disaster vulnerability characteristics within different bearing bodies. Therefore, this study proposed a multi-scale vulnerability assessment model for urban flood disasters considering real-time population changes.

The determination of weights was the key step in the calculation of urban flood vulnerability models. Due to significant differences in vulnerability models for urban flood disasters at different scales, this study calculated the weights of vulnerability models for different scales separately.

The weight calculation method mainly included subjective and objective methods. The subjective method judged the weight of each indicator based on the degree of importance that decision-makers attach to each indicator. The AHP was a typical subjective method that can fully use the wisdom of experts to reflect the importance of various indicators to the vulnerability of urban floods. However, this method that only relies on expert scoring has strong subjectivity. The objective method determined the weight of each indicator directly according to the characteristics of the original data. The CRITIC method was an objective weighting method, which allocated weights by the variability and the conflict of indicators. The conflict between indicators was usually measured by the correlation coefficient. The larger the correlation coefficient, the lower the conflict between indicators, and the smaller the weight of indicators. The variability of indicators was usually measured by standard deviation. The larger the standard deviation, the greater the weight of the indicator. However, since the dimensions of indicators were often different, there was a deficiency in using standard deviation to measure the variability of indicators. To this end, the coefficient of variation was introduced to represent the variability of indicators instead of standard deviation. The coefficient of variation can overcome the errors caused by the different dimensions of indicators and express the importance of indicators by the degree of differences. However, the objective method often overlooks the actual significance of indicators, resulting in weight calculation results that may not be in line with the actual situation. Therefore, this study proposed a CRITIC-AHP method that combines the advantages of the CRITIC method and the AHP to make the weight results more realistic and reliable. The process of using the improved CRITIC-AHP to calculate the weights of indicators at each level was as follows:

• (1) Calculate the coefficient of variation for each indicator

  

  (13)
• (2) Calculate the correlation coefficient matrix based on the standardized results of each indicator

  

  (14)

  where refers to the correlation coefficient between the s-th and l-th indicators, refer to the average value between the s-th and l-th indicator after standardization processing.
• (3) Calculating the quantitative coefficient of the degree of independence

  

  (15)
• (4) Calculate the improved CRITIC weight for the m-th indicator

  

  (16)

• (5) Using the AHP method to calculate subjective weights ⁠, the specific steps and complete mathematical description of the AHP method can be found in Chhetri & Kayastha (2015) and Stefanidis & Stathis (2013).

• (6) Calculate the comprehensive weight of the m-th indicator based on the calculated improved CRITIC weight and AHP weight

  

  (17)

The dynamic vulnerability classification can intuitively reflect the distribution characteristics and changing trends of flood vulnerability in various bearing bodies and regions (Li et al. 2023). The natural breakpoint classification method is a commonly used method for the classification of drought and flood disasters (Priest 2023). It was the univariate classification method of cluster analysis, which is classified based on the distribution law of numerical statistics (Yan et al. 2021). It has strong applicability and can be used for the classification processing of multiple types of data. Therefore, based on the vulnerability index calculated by the vulnerability functions in Sections 2.2 and 2.3, this study used the natural breakpoint classification method to divide the vulnerability assessment results into five levels: high-vulnerability (5), medium-high vulnerability (4), medium-vulnerability (3), low-medium vulnerability (2), and low-vulnerability (1).

From the directionality of the standard deviation ellipse, it can be seen that population density has significant directionality at different time periods. From 0:00 a.m. to 8:00 a.m., the long axis direction of the standard deviation ellipse changed from northwest–southeast to northeast–southwest, because the population mainly migrated to Jinshui District and Zhengdong New Area in the northeast during the morning peak. From 10:00 a.m. to 15:00 p.m., due to the small amount of population migration, there is almost no change in the long axis direction of the standard deviation ellipse. During the evening peak, the long axis direction of the standard deviation ellipse returned to the northeast–southwest, because the population migrated from Jinshui District and Zhengdong New Area in the northeast to the central, western, and southern residential areas in this period. It should be noted that although the long axis direction of the standard deviation ellipse of populations was the same in the morning and evening peak, their potential population migration direction was opposite. These results show the characteristics of population migration at different times of the day. In general, the cross-regional migration of the population at the morning peak and evening peak was the highest. In the daytime, Jinshui District and Zhengdong New Area gathered more population, while in the evening, the population migrated to the central, western, and southern residential areas. Therefore, in the process of urban flood response, attention should be paid to the changes in flood disaster vulnerability of personnel in corresponding regions at different time periods.

Buildings (residence, commerce, industry, public service) and roads were the main places that carried the population in cities. Therefore, this study used the overlay analysis tool of GIS to analyze the population changes in residence, commerce, industry, public service, and road at different time periods.

To further understand the changing characteristics of flood vulnerability in different land-use types, this study used GIS to analyze the flood vulnerability changes in residence, industry, commerce, public service, and road (Supplementary Materials 2–6). The vulnerability of flood disasters in residences at night was significantly higher than in the daytime (Supplementary Material 2). The medium- and high-vulnerability areas in the residence were mainly concentrated in the central and western parts of Zhengzhou City. These areas have high residential density and low urban drainage capacity, resulting in a higher vulnerability of flood disasters in these areas. Therefore, urban flood prevention and control should pay more attention to the vulnerability of nighttime flood disasters in the residents, especially in these high-vulnerability areas in the central and western parts of Zhengzhou City.

The higher vulnerability areas of flood disaster in roads were mainly concentrated on the main roads in Jinshui District, the South Third Ring Road, the West Third Ring Road, and Zhongzhou Avenue (Supplementary Material 3). These areas were the main traffic channels in Zhengzhou, bearing most of the traffic capacity of Zhengzhou, especially Zhongzhou Avenue and Jinshui Road, which are prone to congestion during the morning and evening peak periods. Moreover, there was a significant difference in the vulnerability of flood disasters in roads over time. The vulnerability during the morning and evening peak periods was higher than in other time periods. The main reason was that the road carried a large number of people during the morning and evening peak periods, leading to a significant increase in vulnerability during these periods. Therefore, the government and flood control personnel should pay more attention to the potential flood disaster losses caused by the high-vulnerability areas in roads.

Supplementary Material 4 showed the spatial distribution and spatiotemporal evolution of vulnerability in commerce of Zhengzhou City. The commerce was mainly distributed in the central, southern, and northern regions of Zhengzhou City, with almost no commerce in the southwest. The main reason was that the terrain of Zhengzhou showed a trend of high in the southwest and low in the northeast, and there were many mountains in the southwest direction. The unfavorable terrain leads to relatively poor development in the southwest direction of Zhengzhou. From 0:00a.m. to 18:00 p.m., the vulnerability of commerce showed a gradually increasing trend in Zhengzhou City, which was related to the characteristics of commerce itself. Commerce was almost always closed at night, and the commercial population began to increase during the daytime. By 18:00 p.m., some people who came off work began to migrate to commerce, leading to a gradual increase in the flood vulnerability of commerce. The high-vulnerability commercial areas were mainly concentrated in the central and eastern regions of Zhengzhou City. Therefore, urban flood management personnel should pay more attention to the increased vulnerability of flood disasters in commerce during the daytime in the central and eastern parts of Zhengzhou City, especially in commercial places with sunken commercial squares and underground spaces.

Supplementary Materials 5 and 6 showed the spatial distribution and vulnerability changes of industry and public service. The industry in Zhengzhou was relatively scattered, and the population carried by the industry was small. Although Zhengzhou was a transportation hub in central China, the industrial development was relatively slow. The high-vulnerability areas of flood disasters in the industry were mainly concentrated in the western and southern regions of Zhengzhou City. On the contrary, the high-vulnerability areas of flood disasters in public service were mainly concentrated in the eastern part of Zhengzhou City. From a temporal perspective, the vulnerability of industry and public service during the daytime was higher than at night, as a large population was concentrated in industry and public service areas during the daytime, increasing the vulnerability of flood disasters. Therefore, urban flood prevention and control should pay attention to the increased vulnerability of flood disasters in industry and public service areas during the daytime.

The changes in high-vulnerability areas often have a significant impact on urban flood disasters. Supplementary Material 7 revealed the changes in high-vulnerability areas of different land-use types. The high-vulnerability areas of residence, road, and public services are relatively large, while the high-vulnerability areas of industry and commerce were relatively small. It should be noted that there are significant differences in the high-vulnerability areas at different time periods. At night, the high-vulnerability area of residence was the highest. During the morning and evening peak periods, the high-vulnerability areas in roads were the highest. And from 10:00 a.m. to 15:00 p.m., the high-vulnerability areas in the public service area were relatively large. These results can provide important references for the prevention, control, and response of urban flood disasters. In the process of urban flood prevention and control, it is not only necessary to refine the flood vulnerability of different land-use types, but also to clarify the changing characteristics of flood vulnerability of each land-use type at different time periods, and to formulate flood control plans based on the spatiotemporal changes in urban flood vulnerability.

Overall, the results of flood vulnerability at the grid scale and the regional scale were relatively similar in the spatiotemporal distribution. During different time periods, the flood vulnerability at night was significantly lower than in the daytime. In terms of spatial distribution, the vulnerability of flood disasters in the central part of Zhengzhou City was significantly higher. However, due to the different indicator systems for vulnerability assessment at grid and regional scales, there were certain differences in the focus and results of vulnerability assessment. The grid scale showed the spatial distribution and change characteristics of vulnerability of each bearing body. The results showed that the vulnerability of the morning and evening peak periods on roads increased significantly. The areas with high vulnerability were mainly concentrated in Zhongzhou Avenue, South 3rd Ring Road, West 3rd Ring Road and main roads in Jinshui District. Residences have the highest vulnerability at night, mainly concentrated in the central and western parts of Zhengzhou City. These results can provide an important basis for refined urban flood prevention and response. However, the results of vulnerability assessment at the grid scale cannot intuitively reflect the changes in vulnerability in townships and districts, which may pose challenges for regional managers to coordinate urban flood prevention and control. The vulnerability assessment at the regional scale effectively solved this problem. The results demonstrated that the vulnerability of flood disasters in Jinshui District, Erqi District, and Guancheng District increased significantly during the daytime, and the high-vulnerability areas were mainly concentrated in some townships in the middle of Zhengzhou City. Urban flood prevention and control should focus on the potential flood disaster vulnerability in these regions.

This study constructed a dynamic vulnerability assessment model for urban flood disasters at grid and regional scales. Considering the real-time population changes in a day, the spatiotemporal evolution characteristics of flood disaster vulnerability in various bearing bodies and different regions are analyzed. The research results provide the theoretical basis for refined prevention and control of urban flood disasters and overall management. The main conclusions were as follows:

• (1) There are significant differences in the spatial distribution and migration of populations at different time periods. During the morning and evening peak periods, the cross-regional migration of the population was the highest, and the road carried the most of the population. In the daytime, the population gathered in the northeast of Zhengzhou City, while in the evening, the population migrated to the central and western regions. Therefore, in the process of urban flood control and response, attention should be paid to the potential flood disaster vulnerability of personnel in corresponding regions at different time periods.

• (2) The impact of real-time population changes on the vulnerability of urban flood disasters was significant. The high-vulnerability areas of nighttime flood disasters were mainly concentrated in the central part of Zhengzhou City. During the daytime, affected by population mobility, the areas of vulnerability significantly increased and migrated towards the east and south of Zhengzhou City. The high-vulnerability areas during the morning and evening peak periods were significantly higher than in other periods, with the main areas concentrated in the central region of Zhengzhou City. Therefore, urban flood prevention and control should focus on the urban flood disasters in the central area of the city during morning and evening peak periods.

• (3) The vulnerability of flood disasters in different land-use types indicated that residence, road, and public service had a higher vulnerability, while the vulnerability of commerce and industry was lower. At night, the high-vulnerability areas were mainly concentrated in residences in the central part of Zhengzhou City. During the morning and evening peak periods, the area of high-vulnerability areas on the road was the highest. From 10 to 15 o'clock, high-vulnerability areas were relocated to public service areas in the eastern part of Zhengzhou City. Therefore, urban flood management develops targeted flood control plans based on the spatiotemporal changes in flood vulnerability.

This study considered the spatiotemporal changes of urban flood disaster vulnerability under real-time population changes. In the urban flood prevention and control process, urban managers can take targeted flood prevention and control measures according to the current and future vulnerability assessment results, including timely release of high-risk area hedging guidelines, timely adoption of traffic control measures, and advance allocation of flood control sandbags and drainage equipment. These more targeted prevention and control measures based on dynamic vulnerability assessment results can provide important reference and basis for reducing flood disaster losses.

We appreciate the underlying surface and population data provided by the big-data management department of Zhengzhou City. We also thank the anonymous reviewers for their valuable comments.

S.L. participated in data curation, developed the methodology, validated the process, wrote the original draft, wrote the review, and edited the article. D.Y. developed the methodology, validated the process, wrote the review, edited the article, supervised this work, and was responsible for funding acquisition. H.L. contributed to methodology. J.L. contributed to methodology, supervised the work, and was responsible for funding acquisition. Z.Y. validated the process. L.C. supervised the work.

The research was funded by the Youth Fund of the National Natural Science Foundation of China (No. 52209038), the National Key R&D Program of China (No. 2021YFC3200203).

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

The authors declare there is no conflict.