Simulation of dam-break flood and risk assessment: a case study of Chengbi River Dam in Baise, China

Of all natural disasters, floods are considered catastrophic events with effects that can last for a long time, and some of the most catastrophic flash floods are related to dam breaks. There have been many dam-break events with heavy losses and great effects worldwide. In 1959, the Malpasset Arch Dam in France collapsed, resulting in 421 deaths. In 1991, heavy rainfall in Romania resulted in the destruction of the Belch Dam by floods, killing 25 people. In 2019, a mining dam in Brazil's Minas Gerais State collapsed, killing at least 50 people and leaving more than 200 missing. In 2020, water flow from the Edenville Hydropower Station in the United States destroyed the Sanford Dam downstream, which forced the urgent evacuation of more than 10,000 people in central Michigan. In order to reduce the loss of life and property caused by dam-break floods, it is necessary to combine engineering and non-engineering measures to build a complete flood control and disaster reduction system. In constructing a flood control system, focusing only on engineering measures is not enough to avoid repeated flood disasters. Non-engineering measures such as studying dam-break flood evolution and conducting risk assessment are equally important (Farooq et al. 2019). Therefore, dam-break flood simulation and risk assessment have become hot scientific issues in current hydrological science research.

Many models can be applied to dam-break flood evolution, among which physical models are the most accepted and commonly used for assessment. Flood models can be one-dimensional (1D) or two-dimensional (2D). In 1D flood models, such as HEC-RAS 1D, SOBEK 1D, and MIKE 11, the given terrain is represented as a sequence of cross-sections of rivers and flood plains perpendicular to the direction of flow (Brunner et al. 2015). However, 1D models do not consider the transverse flow of floods and are not suitable for evaluating urban flood evolution. Fortunately, 2D hydrodynamic models consider the flow changes in the longitudinal and transverse directions of a river. Current 2D flood models mainly include SOBEK, LISFLOOD-FP, FLO-2D Pro, HEC-RAS 2D, and MIKE FLOOD (Church et al. 2015). Horritt & Bates (2002) used HEC-RAS and LISFLOOD-FP to simulate the flood inundation of the channels of seven rivers in the UK. They found that the HEC-RAS model could provide a more accurate prediction of the inundated area, while LISFLOOD-FP needed to be corrected according to data of the independent inundated area to produce acceptable results. HEC-RAS can simulate unsteady flow in complex open channel networks and provide more detailed and accurate simulation results (Kaya et al. 2019). As stated by Pilotti et al. (2020), the HEC-RAS 2D model is suitable for dam-break flood research in valleys. Khattak et al. (2016) compared a flood simulation for 2010 with a flood image taken by MODIS through HEC-RAS, which clearly showed the close consistency between the two.

Conducting flood simulations and preparing flood risk maps are essential requirements before implementing non-structural flood hazard mitigation methods (Giustarini et al. 2015). Simulation results only show the extent of flooding in the downstream area and the degree of inundation. In order to analyze the impact of floods more accurately, it is necessary to draw a flood risk map. Flood risk maps help managers and planners to identify areas that are vulnerable to flood hazards and the extent to which they may be affected. Most importantly, they help in prioritizing and making disaster response more efficient (Hazarika et al. 2018).

Evaluating flood risk is the foundation for drawing a risk map. Generally, flood risk is a function of hazard and vulnerability (Skakun et al. 2014). Hazard refers to the physical and statistical aspects of actual flooding (e.g., flood return period, magnitude, and depth). The usual procedure is to apply flood frequency analysis to the flow data and define the corresponding return period flow, for example, by translating a 50-year flood event to inundation levels and depths (Apel et al. 2009). Vulnerability in this context is generally defined as exposure of people and assets to flooding and susceptibility to flood damage. Although the intensity of flood hazard is the same in a specific area, the risk may vary depending on certain conditions, and these conditions are related to vulnerability (Crichton 1999). Therefore, establishing a reasonable method for measuring vulnerability is critical in risk assessment. In Tingsanchali & Karim's (2005) research, vulnerability was measured through an indicator that is only relevant to population density. With this strategy, vulnerability is incapable of reflecting economic losses. If more data were used to describe vulnerability, such as economic data, flood risk could be more plausible and reliable. Some risk assessment studies that assess potential damage from flooding combine social and economic data to generate flood risk maps. This economic assessment of losses involves many parameters, some of which are very subjective and interact in complex and non-linear ways (Costache et al. 2020). Therefore, further research is needed to quantify hazards and vulnerabilities in a more rational way.

The fluid dynamics method can improve the accuracy of flood hazard maps as well as vulnerability and risk maps. As a result, more scholars are applying hydraulic models to study flood hazards. At the same time, flood vulnerability research based on the land use and socioeconomic information of affected areas is also increasing. Aggarwal et al. (2016) used satellite data and a digital elevation model to generate different dam breach parameters using the Teesta river basin in India as an example and used the hydrodynamics module of MIKE-11 software to model and simulate possible glacial lake outburst floods. Flood values and process lines were obtained and a flood risk assessment was performed. Zhang et al. (2022) summarized the complex environmental impacts of dam-break floods in four aspects, including geomorphological changes and water pollution, and proposed a method to quantify the overall environmental impacts of dam-break floods, which provides a scientific basis for assessing flood consequences and managing flood risks. Masood & Takeuchi (2012) used a 1D hydraulic model for hazard-assessed vulnerability using land use data and successfully mapped flood risk. Al Baky et al. (2020) used a 2D flood simulation using the Delft3D model and calculated floodplain inundation depths for hazard assessment. Using inundation depth as a flood loss parameter, depth loss curves for crops and settlements were constructed and reasonable flood risk maps were generated. Afifi et al. (2019) conducted flood simulations based on the 3Di hydraulic model and determined the inundation extent and inundation depth to obtain flood hazard maps. Hierarchical analysis was used as a multicriteria decision-making method to determine the weights of each vulnerability factor and mapped the flood vulnerability areas. Finally, flood risk maps were derived from the vulnerability and flood hazard maps to show the risk level of flood hazards in residential areas. These studies and innovations have enhanced our understanding of the dam-break flood risk. However, many existing studies: (i) used a single parameter to illustrate the characteristics of the downstream evolution of dam-break floods (partly attributed to the use of a 1D flood model), potentially leading to a biased evaluation or (ii) adopted a complex function to characterize the riskiness, limiting the transportability of the innovations.

In this context, the main objectives of this study are: (a) to simulate the flow process of dam break under multiple-scenario, based on the HEC-RAS model; (b) to analyze the evolutionary characteristics of dam-break floods, taking the discharge process of dam break as the input condition of HEC-RAS 2D; and (c) to develop an easy-to-use risk quantification method to assess downstream flood risk based on the evolutionary characteristics of dam-break floods.

The HEC-RAS model is available as free hydraulic calculation software, developed by the US Army Corps of Engineers, which performs unique functions (e.g., hydrology, hydraulics, reservoir operations, and flood impact analysis) and is widely used in many studies and applications (Garcia et al. 2020). The basic parameters of the model are channel geometry data, initial conditions, and boundary conditions. The parameters of the HEC-RAS model also contain dam breach duration development curves, the final morphology of the breach, etc. The main differences between the 2D and 1D models are the generalization form of the dam and the expression of the downstream influence area. The 1D model uses a linear structure to describe the dam, and the downstream influence area is represented by the centerline and cross-section of the river. In the 2D model, the SA/2D flow area is used to connect the dam and reservoir, and an irregular grid is used to represent the downstream influence area.

In this study, the 2D model was simulated based on unsteady flow. During the simulation process, there are two upper boundary conditions, both of which use flow processes. One is located at the dam site section of Chengbi River Reservoir and the other is located near the Bailin Bridge on the Boai River. The dam section of Chengbi River Reservoir adopts three dam-break flood flow processes, and the section of Bailin Bridge adopts the 10-year discharge flood process of the Baise Water Conservancy Project. The downstream boundary condition is normal water depth and is set at the section near the downstream Jiangbayoujiang Bridge.

The unsteady flow calculation of the model is based on the continuity and momentum equations. The four-point implicit finite difference scheme is used to discretize the non-linear equation, and the Newton Raphson iterative method is used to calculate.

The weight of each indicator was calculated using the entropy weight method. This is an objective assignment method that uses the idea of entropy to determine the impact of an indicator on the composite evaluation based on its dispersion. If the information entropy of an indicator is smaller, it indicates that the greater the variability of the indicator, the more information it provides, the greater the role it can play in the comprehensive evaluation, and the greater its weight. On the contrary, the greater the information entropy of an indicator, the smaller the variability of the indicator, the less information it provides, the less it plays a role in the comprehensive evaluation, and the smaller its weight. The determination of weight relies only on the dispersion of the data itself. The detailed calculation process of the entropy weight method is as follows (Gu 2019):

• (1)

  Assuming there are n samples, each with m indicators. Number the data, and set indicator j of sample i as ⁠.

• (2)

  Dedimensionalize each indicator. For positive indicators, a higher value means that the indicator is better. Conversely, for negative indicators, a smaller value means a better indicator. The range method is used to standardize the original data, and the calculation formula is as follows:

Through the above formula, the original data is converted to ⁠.

• (3)
  is the proportion of the indicator j in the sample i, in order to calculate the variation of the index. is the information entropy of a set of data. Calculate the proportion of sample value under indicator i and entropy value of indicator ⁠:

  

  (12)

  

  (13)
• (4)
  Calculate information entropy redundancy (difference) and the weight of each indicator ⁠:

  

  (14)

  

  (15)

In this study, the dam breach mechanism was set as a diffuse roof breach, and the flood risk under different breach cases was explored. The mechanism of a dam break is complex, and there is no recognized and unified research method to deal with its evolution. Therefore, three breach sizes were set in the simulation: full breach, 1/2 breach, and 1/3 breach. Based on the engineering information of the dam and the proposed breach sizes, the final bottom width and bottom elevation of the breach can be determined. According to the operation mode of Chengbi River Hydropower Station, dam height, material, construction quality and other factors, the break duration is set to 2 h. When the dam breaks completely, the elevation of the breach bottom is set to 120 m. Due to the high risk of dam break when the flood is overtopping, it is assumed that the dam will begin to break when the reservoir encounters a 10,000-year check flood and the water level in front of the dam reaches 190.4 m (dam crest elevation). The height of the parapet wall is neglected in the simulation because the parapet wall is more easily damaged by flood and hydrostatic pressure than the dam. Based on some dam-break events, the breach shape of the Chengbi River Dam was set as an inverted trapezoid. In view of the above considerations, three schemes and corresponding evaluation criteria are formulated for different dam-break situations, as shown in Table 1.

As shown in Figure 7, with the increased dam break degree, the composite hazard of dam-break flood increases. The inundation area is mainly concentrated on the banks of the Boai and Right Rivers, while the inundation area on both banks of the Chengbi River is smaller. According to the administrative division and flood characteristics, the study area was divided into Yongle Town, Baicheng Street, and Longjing Street. Finally, using ArcGIS geostatistical tools, the percentage of hazardous areas at all levels was counted, as shown in Table 3. The Right River is downstream of the Chengbi River Dam, while Longjing Street is adjacent to the Right River and is relatively low-lying. Therefore, it can be seen that under each scheme, the inundation area of Longjing Street is the largest. Due to the flat terrain in urban areas and the blocking effect of buildings, the downstream area is less affected by flooding.

The hazard and vulnerability levels were determined according to Table 1. As can be seen from Figure 8(d) and Table 4, most regions generally have medium or higher vulnerability. The areas with high and extremely high vulnerability are mainly distributed in Longjing Street and Baicheng Street. These two areas comprise relatively high-value urban residential land and have a high population density.

Desalegn & Mulu (2021) used GIS and HEC-RAS to map the flooded areas along the Fertam River and statistically found that the flooded areas were higher in the upper and middle reaches of the river than in the lower reaches. Munna et al. (2021) used HEC-RAS to map inundation areas for return periods of 2, 5, 10, 25, 50, and 100 years and concluded that these maps could help in understanding local flood risk. Similarly, in this study, we generated a flood inundation map by HEC-RAS 2D. Through the flood inundation map, taking option 1 as an example, it can be seen that the maximum inundation depth is 5–40 m, the maximum flow velocity is 35 m/s, and the average inundation duration is 2–5 h, which can also clearly see the inundation area and help to understand the flood risk of Baise City. These successful simulation examples show that although 2D simulation requires high primary data and computer performance, it can describe the evolution characteristics of dam-break floods in detail. Rangari et al. (2019) also drew flood risk maps based on HEC-RAS 2D. According to those simulation results, 17% of the total area was prone to flooding, of which the high-risk area accounted for 9%, the medium-risk area for 52%, and the low-risk area for 35%. However, it should be noted that in his research, the classification standard of flood risk level was the inundation depth (m) generated over the simulation run period.

In this study, in the generation of flood risk maps, we considered not only the inundation depth of the flood but also its flow velocity and inundation duration and the GDP, population density, and land use data of downstream areas. Therefore, the flood risk classification standards we used are also different. Masood & Takeuchi (2012) used HEC-RAS to carry out flood simulation and ArcGIS to create a flood hazard map based on simulation results. They also prepared risk maps to assess flood risk. Similar to this study, in their study risk was defined as a product of hazard (i.e., inundation depth) and vulnerability (i.e., exposure of people or assets to flooding). However, their method of calculating risk and vulnerability indices was not consistent with this study, and the classification level was also different. Al Baky et al. (2020) used inundation depth as the flood damage parameter, constructed depth-damage curves for crops and settlements, and created risk maps. Although they did not use HEC-RAS as the 2D hydrodynamic model, they correctly considered a range of conditions (flood vulnerability) that may contribute to differences in flood risk to produce reasonable flood risk maps.

It has been reported that three factors will affect the accuracy of flood simulation: (1) the error of the structural model itself, (2) input data error, and (3) numerical error (Kim et al. 2014). Through flood simulation, Shustikova et al. (2019) suggested that the mesh size plays an important role in the accuracy of the simulation results. One of the main reasons is that a finer resolution model can capture more topographic details and consider depressions and terrain to guide the water flow in the right direction. In order to reduce the data error, in this paper we combine 30 and 12.5 m resolution DEM to modify the terrain data through the measured control points. The flood routing process can be simulated more accurately by inputting the modified DEM data into the model.

In addition, to test their reasonableness, the simulation results were analyzed by the water volume balance of the statistical model. The water volume in 72 h simulation time was measured, and the relative error between upstream inflow and downstream outflow and regional water storage was calculated. It can be seen from Table 6 that the water volume error of the three schemes is 3.19, 4.62, and 4.85%, respectively, which basically satisfy the water quantity balance requirement, indicating that the simulation results of the model are stable.

Different indicators are selected, and the results may have some uncertainty due to the non-independence of some indicators, which will produce different calculation results. Taking vulnerability as an example, the impact of potential non-independence of indicators on the results is further analyzed. Referring to the study of Wangyang et al. (2019), three indicators are selected to characterize vulnerability, namely, population density, average GDP, and land use type, respectively, considering the impact of loss of life, economic loss, and production stagnation. It was found that removing the land unit price indicator, while considering the indicators non-independence (GDP and land price), produced a significant difference in combined vulnerability: all moderately vulnerable areas disappeared, 99.62% of very high vulnerability areas were distributed throughout the study area, and only 0.38% of high vulnerability areas were distributed in the less populated northern edge of the study area. Without considering the land unit price indicator, most of the sparsely populated mountainous areas are judged to be extremely high vulnerability areas, as are the densely populated urban areas, but this does not correspond to the actual situation on the ground. The main reason for this difference is that the normalized (Equation (9)) vulnerability values for both population density and GDP are high (the lowest value of vulnerability for population density is 0.78 and the lowest value of vulnerability for GDP is 0.58), which leads to a combined vulnerability above 0.58 regardless of the entropy weighting method, i.e., the final distribution of vulnerable areas is mainly high and very high vulnerability areas are dominated. In order to characterize the impact of production stagnation in the floodplain, we additionally included land unit price as one of the indicators for vulnerability calculation. The unused land (e.g., bare soil, grassland, and sparse woodland), which is widely distributed in the study area, has a very low price and hardly causes production stagnation even when subjected to flooding. It can be seen that the inclusion of the land unit price indicator in the vulnerability calculation will reduce the vulnerability of most of the study area, which can also reflect the actual situation on the ground more realistically. Based on this, the three indicators of vulnerability such as GDP, land use type (land price), and population density are still used simultaneously in this study to characterize vulnerability. However, the non-independence that exists in them needs to be noted in the study.

The simulations show that the greater the degree of the dam break, the greater the risk of flooding downstream. According to this rule, local authorities should regularly inspect, repair, and strengthen the dam, which will minimize the extent of any breach and reduce the flooding hazard downstream. The extent and duration of inundation can be observed in the results of the flood evolution. In particular, the area of Longjing Street near the Right River is larger than the other areas. In order to minimize the loss of life and property in the downstream area, the local government should issue emergency warnings to evacuate the residents before any flood. Based on the flood risk maps, Longjing Street is the administrative district most affected by dam-break floods. According to this result, the local government should focus on strengthening the construction of flood control projects, such as renovation and reinforcement of embankments. The local government should dredge the rivers and construct flood diversion and storage projects for the nearby rivers. Such flood risk maps can be used for long-term land use planning in downstream areas, making appropriate development plans, and assessing the cost of flood control measures taken by local authorities.

Compared with the dam breach simulation study under a single scheme, in this study we considered more scenarios. Based on HEC-RAS 2D, we simulated the flooding process at different levels of dam breach and derived flood hazard parameters including maximum inundation depth, maximum flow velocity, and inundation duration. These parameters can illustrate the characteristics of the downstream evolution of dam-break floods. An important innovation in this study is the use of a novel risk assessment method to analyze flood risk in downstream areas. The method is simple and can be used for other areas with different characteristics. It offers researchers the flexibility to apply more evaluation indicators in order to obtain more reasonable flood risk maps. The Chengbi River Dam plays an important role in protecting the lives and properties of downstream residents, but there are very few dam-break flood simulation results. Therefore, the flood risk maps generated in this study indicate the need to construct flood control projects in the downstream area. In addition, Baise City belongs to karst areas. Karst landscapes are among the most vulnerable landform types in the world and are highly sensitive to environmental changes. With climate change and continuous economic development in Baise City, it is of particular research significance to simulate dam-break floods and analyze the risks associated with the Chengbi River Dam.

This study has the following limitations: There is no unified research method to deal with the evolution of dam breaks. For the breach flow process, only three schemes (full breach, 1/2 breach, and 1/3 breach) were used in this study, which cannot dynamically reflect the evolution process of a dam opening. In particular, it should be noted that the number of indicators selected in this study was small due to the limited amount of relevant information and data, which may have affected the results of the risk assessment to a certain extent. If possible, further consideration will be given to population age structure or ecological indicators. However, the acquisition cost of some information is high, such as the distribution and value of on-site buildings, which requires a lot of time and effort. In addition, the influence of river-related facilities such as bridges and wharves on flood evolution was not considered in the simulation process. Although the influence of these structures on dam-break flood evolution simulations may be relatively small, from a scientific point of view, information on such structures can be collected and added to simulations in the future to match the actual situation much as possible. At the same time, considering the importance of Baise City, flood risk is calculated according to the most unfavorable combination during risk assessment. However, the relationship between water depth and flow velocity can be further considered when conducting hazard analysis for areas of different importance.

Performing evolution simulations and risk studies of dam-break floods is important in order to reduce the loss of life and property caused by floods, or even prevent flood disasters. This study used a 2D hydrodynamic model to simulate dam-break floods at the Chengbi River Dam in China and drew corresponding flood risk maps. First, we found that the degree of dam break has a specific influence on the discharge and water level characteristics of the breach. The peak flow of S1 was 335,693 m³/s, which is 5.5 times the peak flow of S2 and 10.8 times the peak flow of S3. Second, in different dam-break schemes, the downstream inundation process was similar, but the inundation degree differed. The greater the degree of the breach, the greater the inundation depth, maximum velocity, and inundation duration of the flood. Third, the risk involved in dam break is mainly distributed on both sides of the Boai and Right Rivers. Under the three schemes, the flood risk area of Longjing Street accounts for 65.37, 71.41, and 66.22% of the total risk area, respectively, and is the area most extensively impacted by flooding. Correct use of a flood risk map would enable residents to understand the extent of flood damage and prepare for it in advance, thereby reducing the scale of the disaster. For government departments, this study may provide a basis for carrying out different levels of flood emergency preparedness. In addition, the flood risk map clearly shows the distribution of high-risk areas in the region, which can help government departments make more cautious development plans for these areas. Furthermore, as far as flood risk analysis is concerned, the analytical approach to hazard, vulnerability, and risk indicators is innovative and effective. The new quantification risk analysis method not only visualizes and quantifies the three evolution characteristics (inundation depth, maximum flow velocity, and inundation duration) but also considers some downstream socioeconomic aspects (GDP, population, and land use data), making the risk map reasonable and reliable. The method is also simple and can be applied to other areas with different characteristics, such as plains or mountains. In addition, it offers researchers the flexibility to use more evaluation indicators to obtain a more reasonable flood risk map. Some of the indicators identified in this study can be replicated in risk assessments of other natural hazards. Specifically, many of the indicators used in this study can be used in risk assessments related to other types of flooding (fluvial and pluvial), including population density, land use, and economic data. A similar approach can be taken for other hazard phenomena but requires the selection of appropriate hazard and vulnerability indicators according to the specific framework of the risk analysis.

W. C. and X. L. conceptualized the study; W. C. and Y. R. did data curation; W. C. and H. B. did formal analysis; C. M. and Z. X. acquired funds; W. C. and Y. R. investigated the study; C. M. and W. C. performed methodology; C. M. and S. L. did project administration; W. C. and Y. R. collected resources; W. C. and H. B. did software analysis; Z. X. supervised the study; v W. C. and X. L. validated the study; W. C. and H. B. visualized the study; W. C. and H. B. wrote the original draft; Y. S wrote, reviewed, and edited the article. All authors have read and agreed to the published version of the manuscript.

This research was funded by the National Natural Science Foundation of China (51969004, 52269002, and 51979038) and the Interdisciplinary Scientific Research Foundation of Guangxi University (2022JCC028).

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

The authors declare there is no conflict.