Urban Flooding Risk Assessment Based on Numerical Simulation of Sub-catchment Area: A Case Study From Chengdu, China


 Urbanization and climate change usually result in frequent urban flooding. Since the floods cannot be avoided, the scenario simulation combined with risk analysis is an effective way to assess the disaster level and reduce direct damage loss when facing the emergency management problems. Different from the whole city dimension, the paper proposed a sub-catchment multi-index hesitant fuzzy evaluation model for the community planning level, and takes Jinjiang District of Chengdu city as the research object. Firstly, based on the PSR (Pressure-State-Response) model, the risk assessment system has been established in three aspects, including the current situation of urban drainage, the basic geographic information, and the social influence. Secondly, A total of 14 evaluation indexes were selected, among which the pressure index came from the calculation results of ArcGIS and EPASWMM5 model such as runoff coefficient, maximum water depth, etc. Thirdly, the expert hesitate fuzzy evaluation method was used to obtain the weight of 14 indexes of each sub-catchment. Finally, the 224 evaluation results were compared, and the urban flooding disaster risk map has been drawn. It is mainly concluded that 160 medium-higher risk areas were mainly concentrated in high built-up area in study area. Furthermore, the evaluation model is very useful as a decision-making tool for mitigation of the flood hazard and its associated risk.


Introduction
Flooding has been identified as the most common but serious natural hazard all over the world (Sahoo and Sreeja 2016).In contrast to other natural hazards, the main disaster site of urban flood is crowded city, the occurrence of flood will bring immeasurable loss to human beings (Walsh et al. 2012).The mechanism of waterlogging includes both natural and man-made factors (Leopold 1968;Pyke et al. 2011).The dramatic climate change caused by global warming during the last century (as shown in Fig. 1a) will lead to high water vapor in rain areas and result in the increased frequency of extreme rainfall events, as well as the probability of urban flooding (Trenberth 2011).Moreover, with the high rate of rural-urban migration in many countries, the construction of reinforced concrete buildings on coasts and along rivers has changed the original geomorphology (Jha et al. 2012).More and more land resources have been occupied by infrastructure, which gradually increases the impervious surface and decreases the urban drainage capacity (Goonetilleke et al. 2005;Su 2020).Fig. 1b indicates that in the context of global warming, the number of waterlogged cities is positively related to the urbanization rate in China for nearly twenty years.Because of its severe impact on human lives and economic losses, floods have been the top three frequent disasters in the previous decade.Compared with the annual average of 149 events (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018), 194 flood disasters continued to be the deadliest type in 2019, accounting for 43.5% of the total natural disaster deaths (Centre for Research on the Epidemiology of Disasters 2020).Specifically, 5,110 of the affected 31 million people died, and $36.8 billion in agricultural, traffic, housing, and ecological damages were caused.From the seasonal and geographical distribution view, over 2,200 lives were reported to have been lost in various flooding episodes in India, Nepal, Bangladesh, and Myanmar from July to October, 2020.Monsoonal flooding also affected parts of southern China in June 2020, with 83 deaths and over $2.5 billion economic losses (United Nations 2019).In addition, the flood effects in urban areas are mainly due to extreme rainfall.In late January and early February 2019, a tropical low brought extreme rainfall and associated flooding in northern Queensland.The total 10-day rainfall attained 1,259.8mm in Townsville and exceeded 2000 mm in some coastal areas around Townsville. (Aon 2020).
In recent decades, urban flooding has been given more attention by the public in China and beyond.According to the statistics reported in China Flood and Drought Disaster Bulletin (Ministry of Water Resources of the People's Republic 2019), the direct economic loss caused by flood disasters is on the rise, accounting for the largest proportion of the total economic loss caused by meteorological disasters.In September 2010, a strong typhoon called "Fanabi" struck 68 cities, such as Guangzhou, Fuzhou, Nanning, and affected over 1.948 million people with inadequate measures to deal with storm events.There were 128 deaths, rainfall inundated 73,167.9ac, and $877 million loss in property in this hazard (Ministry of Water Resources of the People's Republic 2010).On July 21 st, 2012, an extreme rainfall event happened in Beijing with an average of 215 mm rainfall in urban areas within 16 hours.
The Beijing flood caused 24 hours of gridlock, destroyed 8,200 homes, and led to 79 deaths and $1.6billion financial losses in damages (Huang 2012).Those casualties and property damages provided an awareness of the alarming situation of urban floods in Chinese cities.
Presently, urban flood management has changed from engineering defense measures to risk management (Johnson et al. 2007).Urban flood risk assessment aims to analyze current or future flood risk information, identify high-risk areas, and provide decision support for flood mitigation measures (Moritz et al. 2016).Traditionally, three primary approaches are used to assess flood risk (Yin et al. 2015), historical disaster mathematical statistics method (HDMS) (Nott 2006), multi-criteria analysis (MCA) (Guo et al. 2014), and scenario simulation analysis (SSA) (Zhu et al. 2016).Although these methods can independently assess flood risk, they all have their advantages and disadvantages.For example, HDMS has clear thinking and simple calculation but requires more accurate data, so it's hard to apply.MCA is the most widely used method, which can reflect the regional risk situation at a macro level.However, MCA is primarily used for large-scale risk assessment, and sub-catchment area studies are rarely considered due to their specific limitations in accuracy.Based on the physical mechanism, SSA reflects the actual rainfall situation of the city by simulating the flooding process, but the data amount is large and the calculation is complicated.In this paper, MCA and SSA are integrated to break through the limitations of MCA.EPASWMM5 simulation model and ArcGIS calculation method of model data are used to apply MCA to sub-catchment research, and a sub-catchment multi-index hesitant fuzzy evaluation model (SMHF) is proposed.EPASWMM5 is a software developed by the University of Florida to analyze the impact of rainfall processes on urban runoff.The simulation process requires a lot of accurate data as the foundation of modeling, so ArcGIS is used to provide accurate data calculation for EPASWMM5 simulation.
According to disaster systems theory, a hazard is caused by the joint action of disaster-inducing factors, hazard-pregnant environment, and disaster-bearing body (Shi 1991).In 1979, statisticians David J. Apport and Tony Friend proposed the PSR (Pressure-State-Response) model in Canada (Liu 2015).PSR believes that disasters are caused by the interaction between human beings and the environment, and the model is constructed according to the logical relationship of "cause-effectresponse".Therefore, the PSR was matched with the disaster-inducing factors, hazard-pregnant environment, and disaster-bearing body of urban flooding, and 14 indexes were selected to construct the SMHF model.In recent years, many scholars have established flood risk assessment systems (Xu et al. 2018;Geng et al. 2020;Wang et al. 2018;Hu et al. 2017).Most studies use city as risk assessment units (Zl et al. 2019;Ballesteros et al, 2018;Vojinovic et al. 2016;Ghosh 2018), resulting in poor model accuracy for small areas (such as communities).In the SMHF model, the simulation results of small areas are extracted, the human factors, social-economic factors, and environmental factors of each small area are comprehensively considered.In order to narrow the differences between decisionmaking groups, Torra and Narukawa put forward the concept of Hesitant fuzzy set (Torra 2010).As all kinds of indicators are clear indicators, hesitating fuzzy language is considered as the urban flooding evaluation tool in SMHF to determine the weight of indicators.
The purpose of this paper is to propose a risk assessment system that is based on EPASWMM5 simulation and to classify the risk area according to the risk index.Section 2 describes the problem of the study area and data.Flood risk assessment systems are presented in Section 3. The analysis results are in Section 4 and the conclusions are given in Section 5.

Studied area
Chengdu is located in the west of Sichuan Basin, in which the terrain is high in the northwest and low in the south.The city has a subtropical monsoon climate with abundant rainfall throughout the year.An increasing number of people immigrated from the rural area to cities, resulting in the continuous improvement of the urbanization level.Table 1 shows the major urban floods events that happened in Chengdu.The rainfall event selected for the experiment, which begins at 11 pm on August 11, 2020, and continued until 4 pm on August 12, 2020.

Table 1 Statistics of major urban floods events caused by rainstorm in Chengdu
Jinjiang District of Chengdu (Fig. 2) is characterized by low topography, many rivers, a complex water system, a narrow downstream outlet section, and poor urban drainage conditions.The annual rainfall is about 854 mm.There are more than 80 percent of the land is impervious.Hence the selection of Jinjiang District as the study area in this research.

Data used
The data used in the proposed method includes two parts: experimental data and socioeconomic data.(Li et al. 2019;Wu et al. 2018).
 The comprehensiveness principle: Disaster is a complex process, and each indicator needs to reflect the specific situation of urban areas comprehensively.
 The validity principle: Ensuring the effectiveness of index data is an important way to raise the accuracy of the risk assessment system.
 The convenience principle: In order to increase the applicability of the evaluation system, the indexes should be based on the objective situation of the city and easily obtained.
 The adaptation principle: The risk level can be evaluated objectively only when the indexes are adapted to the actual urban situation.

Pressure indicators (A1)
The pressure index is the effect of human economic and social activities on the environment.The level of urban flooding risk is closely related to the capacity of the drainage system (Wang and Zhou 2012).
Most of the study areas are old areas, and the lack of underground pipeline information leads to less optimistic research results.Therefore, EPASWMM5 should be used to calculate reasonable regional parameters.The rainfall event was applied to the established city model to simulate the process of the day's rainfall event, the simulation results (from EPASWMM5) can objectively reflect the actual situation of the region.
Runoff volume ratio (A13) is the percentage of runoff volume of sub-catchment ( ) of the mean runoff volume of the study area ( * ).
13, = * (2) Other pressure indicators are derived from the model, as shown in Table 2. Maximum submerged area in a sub-catchment

State indicators (A2)
Status indicators indicate the current state of the city over a period of time, such as building coverage, green rate, and so on.
Green rate (A21) is determined by the percentage of the area of green ( ) in sub-catchment (A) (such as lawn, green belt, etc., not including low impact development facilities).

21, = × 100%
(3) Slope (A22) determines the direction of water flow and is the direct cause of water accumulation in a particular area.It is calculated by Geographic Information System (ArcGIS), and the average slope of each sub-catchment is selected.
For the sub-catchment area, the area covered by buildings has almost no permeability.Therefore, the higher the building density, the higher the risk of urban flooding.Building density ( 23) is calculated as the area of building ( ) divided by sub-catchment area ().
Recently, the effectiveness of LID has also been highlighted as a potential option to reduce urban flood risks (Ellis 2016;Kim et al. 2016).Permeable pavement is one of the LID measures and is also the most widely used facility.The aim of permeable pavement is to improve the city's permeability by using more permeable pavement materials.In this article, Permeable pavement rate (A24) belong to the ratio of permeable pavement area ( ) to sub-catchment area ().

Response indicators (A3)
When a flood disaster occurs, society and individuals will take actions to mitigate, prevent and recover the negative impact of the disaster.These indicators are defined as response indicators.
The severity of the disaster in the region is closely related to population size.The larger the population, the more developed the economy, the denser the urban roads and traffic, and the more loss of people and property will be seriously caused by disasters (Huang 2020).Therefore, population (A31) is also an important indicator in the evaluation system.The demographic statistics in the statistical yearbook are based on the number of people per square kilometer ( * ), so the population in subcatchment is: Medical capacity (A32) and health management (A33) are important symbols of the ability to respond to disaster emergency measures.Medical capacity and health management are formulas 7 and 8 respectively.
where is the population of the study area; ℎ is the number of hospitals; is the number of fitness facilities.
A favorable economy can speed up recovery from hazards.In this paper, disposable income (A34) refers to the income obtained by residents in a year and can be used for free disposal.It's the per capita disposable income () multiplied by 31 .PCDI (RMB) can come from Jinjiang District Bureau of Statistics no calculation is required.

Model simulation and index quantification
The indexes of SMHF include three-type indicators (in section 3.

Model simulated by EPASWMM5 and ArcGIS
Step Step 2: Parameter calculation and setting.The model parameters are divided into measurement parameters and non-measured parameters, among which the measurement parameters are mainly calculated by ArcGIS, while the non-measured parameter parameters are selected according to the actual conditions.
The measurement parameters include the area, width impervious ratio, and slope of each subcatchment.The area of each sub-catchment can be calculated by using the zonal geometric statistics tool in ArcGIS, and the maximum area was 323 ha and the minimum area was 3 ha.As the length of the ground flow is difficult to determine accurately, the square root of the area of the sub-catchment is used to determine the width of the sub-catchment, assuming that each sub-catchment is a square region (Bisht et al. 2016).The impervious ratio and average slope of each sub-catchment can be obtained through the spatial analysis tool in ArcGIS, which is based on the land use map (Fig 4-c) and slope map Table 3 is the measurement parameters of some sub-catchments.
Table 3 The measurement parameters of some sub-catchments The non-measured parameter parameters include Manning's N for overland flow for pervious and impervious areas, depth of depression storage for impervious and pervious, and Percent of the impervious area with no depression.Since the simulated area belongs to a small watershed, Horton model was selected as the infiltration model, Dynamic Wave was selected as the confluence model, and relevant parameters were selected by referring to EPASWMM5 user manual (Rossman 2009;McCuen et al, 1996;ASCE 1992;ASCE 1982;Rawls et al. 1983).The other parameters were selected based on the local actual situation, as shown in Table 4. belongs to a concentrated area.In this study, the runoff coefficients of the rainfall event simulation are 0.67, which meet the national standards.The continuity error method represents the percent difference between initial storage total inflow and final storage total outflow for the entire drainage system.If they exceed some reasonable level (10%), then the validity of the analysis results must be questioned (Rossman 2009).Fig. 4e and Fig. 4f show the result of the run state and model of EPASWMM5.Surface runoff is -0.1%,Flow Routing is 0.04%, they're both within the margin of error.Therefore, the model is considered to have certain reliability.

Quantification of the indicators
According to section 3.1, 14 indicators can be calculated.As there are 224 sub-catchments, some subcatchments (the first 5) index data will be shown in Table 6, and all the index data are listed in Appendix A.

Calculation of the risk level
Flood risk assessment is a complex systematic process, and the risk is considered to be the result of a combination of three categories of Pressure, State, and Response, as follows.
where R is the risk of sub-catchment; is the indicator number; 1 , 2 , 3 is the weight of the Pressure, State and Response, respectively; is the score of the indicators; is the weight of the indicators.
The index score is divided in section 3.3.1, the expert weight is calculated in section 3.3.2,and the index weight is calculated in section 3.3.3.

Index classification
The index scores were evenly divided into 4 grades according to the calculation results in Appendix A.
For the 14 indicators, the score value is positively correlated with the risk value.The specific division is shown in Table 7.

Determination of expert weights
In 1965, professor Zadeh (1965) ) proposed the fuzzy set theory that the fuzzy membership function in the interval [0,1] could be used to quantitatively describe the system attributes.Such as Li et al (2015) used a 7 scaled score model to characterize the linguistic values {E-3, E-2, E-1, E0, E1, E2, E3} = {Extremely poor, Very poor, Poor, Medium, Good, Very good, Extremely good} as numbers {0, 0.17, 0.33, 0.5, 0.67, 0.83, 1}.This paper invited 5 risk assessment experts in Chengdu to evaluate the importance of risk assessment system indicators in Jinjiang District of Chengdu using hesitant fuzzy language.Using the theory of hesitant fuzzy set, the language of hesitant fuzzy is transformed into a real number in [0,1] by a harmonic function.There are 7 languages quantified: {Extremely unimportant, Very unimportant, Unimportant, Medium, Important, Very important, Extremely important} = {0, 0.17, 0.33, 0.5, 0.67, 0.83, 1}.Experts can give one or more evaluation languages, using the attitude-neutral extension, the expert criteria importance judgements were extended to the same length.
The traditional evaluation method usually gives random or the same weight to each expert, which leads to the evaluation results not being very practical (Gholamreza et al. 2020;Sandeep and Mohit 2021).The expert weight in this paper is determined by using the Euclidean distance model to maximize the group consensus and minimize the expert differences (Li et al. 2019).In this model, firstly, experts ( = 1,2, ⋯, ) use fuzzy language to judge the importance of each indicator, and then transform the fuzzy language into the corresponding value, and expand to the same length (E).The optimal evaluator weight is calculated by using the optimized Euclidean distance model.

Determination of indicators weights
The importance of indicators can be determined from the hesitant fuzzy evaluation given by experts.
The weighted average operator method (Li et al. 2017) is applied to determine the indicators weights.
To ensure comparability, the original hesitant fuzzy evaluation value , = { , | = 1, ⋯, , = 1, ⋯, , = 1,2, ⋯, } of each expert was extended to the same length, E. The expert weights obtained from model ( 10) are applied to the hesitant fuzzy evaluation value with the same length: The parameters of the weighted average algorithm ( , , ) are calculated as follows.
Before determining the index weights, the weights of pressure ( 1 ), state ( 2 ) and response ( 3 ) in the PSR model must be determined respectively.
The hesitating fuzzy evaluation value is transformed into the corresponding triangular intuitionistic fuzzy number, and the weight of the index is calculated.
where 1 represent the weight of each sub-index of the pressure; 2 represent the weight of each subindex of the state; 3 represent the weight of each sub-index of the response.

Original information processing
Five urban disaster risk management experts put forward different opinions on the importance of each indicator.The original evaluation data are shown in Table 8.Using the attitude-neutral extension, the expert criteria importance judgements were extended to the same length The processed information is shown in Table 9. (0.17,0.33) (0.5,0.67) (0.83) (0.5,0.67) (0.33,0.5) The expert weights are calculated by Formula (11) as follows, = {0.215806017983369,0.220849921926395, 0.183094470536286, 0.194173753605929, 0.18607583578458}.According to the expert weight ( ) and using formula (12-13), the weight of pressure ( 1 ), status ( 2 )and response ( 3 ) were calculated respectively.At the same time, the weight of the sub-index was calculated by equation ( 14).As shown in Table 10.

Calculation of the risk of sub-catchment (R)
The risk value of urban flooding in Jinjiang District of Chengdu was calculated according to the comprehensive evaluation method, that is, the risk value was obtained by multiplying the weight of the three indexes by the fraction (Formula 10).From the calculated results of risk value, the urban flooding risk area of Jinjiang District in Chengdu city is divided into 5, which are higher, high, medium, low, and lower respectively.Among them, the risk value less than 1.8 is a lower risk area, the risk value between 1.8 and 2 is the low-risk area, the risk value between 2 and 2.5 is medium risk area, the risk value between 2.5 and 3 is high-risk area, and the risk value above 3 is higher risk area.
The northwest of the study area is mainly residential and commercial, with poor permeability and relatively low terrain, so it is mainly high-risk areas.On the contrary, the southern region is mainly dominated by urban parks and farmland, with strong water absorption capacity and good drainage facilities, so it is not easy to occur urban floods.
For the risk calculation results, the study area has 87 medium-risk areas, the highest number of risk areas.This was followed by 52 high-risk areas and 40 low-risk areas.According to the risk map (Fig 5), low-risk areas are mainly distributed in areas with sparse buildings in the city, while the medium-risk area is distributed in each densely built area.Of course, the most dangerous areas are higher-risk areas ( 21) and high-risk areas ( 52), which are more concentrated in old urban areas (due north of the study area map).There are also 24 lower-risk areas located near the city park in Jinjiang District, where the greening rate is higher and the drainage facilities are advanced, so the risk is lower. The difference of sub-catchment is considered in risk assessment, which increases the accuracy of risk value.
 The expert hesitant fuzzy evaluation method is considered in determining the index weight, which increases the objectivity of the evaluation results.
According to the assessment results, the Jinjiang District was divided into the five levels of urban flooding disaster risk areas.
 The study areas are mainly medium-higher risk, and most of them are concentrated in dense urban areas.
 Since this experiment is based on the evaluation of an extreme rainfall event, it is considered that the existing drainage facilities in the study area are difficult to prevent such extreme rainfall.
 It is suggested to prepare for flooding disaster prevention and mitigation in urban dense areas to reduce the loss of disasters.
Although this research method has some advantages, it also has some limitations to provide a reference for future research.The selection of indicators is mainly based on expert opinions and cities, which can be considered more comprehensively. Min.Infiltration.Rate Rawls, W.J. et al., (1983)

Fig. 1
Fig.1 (a) the annual global average temperature growth trend and (b) The annual urbanization rate and the annual number of waterlogged cities in China 1).Pressure indicators and State indicators were calculated by establishing EPASWMM5 and ArcGIS model.Response indicators are calculated by statistics.

1 :
Division of sub-catchment.It is particularly important in EPASWMM5.Sub-catchment area refers to the catchment range of catchment points, and there are two main ways to divide the area: Tyson polygon and artificial partition.Due to the lack of data, the more scientific Tyson polygon method was chosen.Hydrological analysis (Fig 4-a) is the first step to constructing the basic model of ArcGIS.After the fill, flow direction calculation, and flow rate calculation, the Tyson polygon method is used to divide the catchment, and finally, 224 sub-catchments are obtained by adjustment (Fig 4-b).
Fig. 4 Development of model for the study

Fig. 5
Fig.5 Study area flooding disaster risk distribution map Finer resolution observation and monitoring (2018).http://data.ess.tsinghua.edu.cn/ Rainfall event Huiju Environmental (2020) Huiju Data.https://hz.hjhj-e.com/home N-Impervious and N-Pervious McCuen R, Johnson P, Ragan R (1996) Highway hydrology.Hydraulic design series No. 2. U.S. Department of Transportation, Federal Highway Adminstration, Washington, D.C. 326. Dstore-Impervious and Dstore-Pervious ASCE (1992) Design and construction of urban stormwater management systems, New York. % Zero-Impervious Select this parameter based on local conditions  Max.Infiltration.Rate and Decay Const Rossman L A (2009) Storm water management model user's manual version 5.0 [M] Washington: National risk management research laboratory, Office of research and development, U S Environmental Protection Agency.

3 Methodology 3.1 Establish Urban flood assessment indicator system
The experimental data include digital elevation model (DEM), rainfall, land use map.The DEM

Table 2
Concept of other pressure indicators

Table 4
Table of Non-measured parameterand lacks the measured pipeline flow data, the comprehensive runoff coefficient method and continuity error method is adopted to verify the model.The theory of the comprehensive runoff coefficient method is to compare the runoff coefficient obtained by model simulation with the comprehensive runoff coefficient of the study area to ensure it is in the value range.The code for comprehensive runoff coefficient in different areas is given in GB 50014 --2006 Code for Design of Outdoor Drainage (Table5)(Shanghai Municipal Construction and     Transportation Commission 2016).According to the impervious rate of each region calculated by ArcGIS, the average impervious rate of the research region is 56.2%.Jinjiang District in Chengdu **Dstore-PerviousDepth of depression storage on the pervious portion of the sub-catchmentStep 3: Model calibration.To ensure the reliability and accuracy of the model, it is necessary to verify the constructed model.As part of the pipeline network in the study area is still under planning

Table 5
Comprehensive runoff coefficient in different areas

Table 6
The first five sub-catchments quantitative indicators

Table 7
Index score grade division

Table 8
Original data for indicators