Comparative analysis and risk assessment of dam-break floods: Taking Pingshuijiang Reservoir as an example

Construction of reservoir dams and other water conservancy projects can deliver many social and economic benefits, such as flood control, power generation, irrigation, water supply, fishery, tourism, etc. (Safarzadeh 2017; Guo et al. 2019; Khoshkonesh et al. 2022). However, the huge potential energy generated by flood storage also poses a great threat to the safety of life and property of people living downstream (Xu et al. 2021). In recent years, the increased frequency of extreme climate phenomena (Alcayna et al. 2022) has put the reservoir flood control standards calculated and formulated using existing hydrological data under scrutiny (Chen & Sun 2021). Climate extremes also lead to the operation of reservoir dams under adverse conditions for a longer time, which directly affects the stability of the dam structure and accelerates ageing of hydraulic materials (Cai et al. 2011), enhancing the dam-break potential.

Evidence from historical dam-break events in various countries highlights the potential for catastrophic damage to downstream people and properties (Qiu et al. 2021; Seiya et al. 2021). One of the most significant dam-break events in China was on 8 August 1975, the collapse of the Banqiao Reservoir dam in western Henan province due to extreme rainfall generated by a tropical cyclone. The resulting floods affected 11 million people across 29 counties, with more than 26,000 deaths, 5.96 million houses destroyed, 11,300 km² of farmland inundated, 302,300 livestock lost and 102 km of the Beijing-Guangzhou railway destroyed; direct economic losses of nearly 10 billion yuan were recorded (Bai & Zhen 2014). The Banqiao event involved a complete dam collapse, but even partial collapses can have devastating impacts due to chain effects. On 19 May 2020, the flood wave generated by the partial failure of the Edenville dam in Michigan in the United States led to overtopping of the Sanford dam and Poseyville embankment downstream, resulting in more than 10,000 people having to evacuate their homes and damage to more than 2,000 homes, as well as multiple roads, bridges and businesses (Pradel & Lobbestael 2021).

Dam failure is mostly caused by floodwater infiltration, overtopping, material instability or external damage. The dam failure process may be gradual or instantaneous; the nature of dam failure (partial or complete, gradual or instantaneous) influences the flood process and downstream impacts. Instantaneous failures are often associated with catastrophic downstream impacts, such as the 1959 dam-break dam-break at Malpasset in France (Duffaut & Larouzée 2019) and the 2011 Fujinuma dam-break in Japan (Pradel et al. 2017). Systematic analysis of different dam-break modes requires the determination of flood parameters such as peak flow, time to peak, velocity and area of inundation.

Numerous approaches to modeling dam-break flood parameters and risk have been published in recent years. Delenne et al. (2012) assessed the potential of local sensitivity analysis for the analysis of uncertainty with respect to flash flood and dam failure risks in river hydrodynamics. Seyedashraf et al. (2018) used a method based on computational intelligence systems to simulate the classical one-dimensional dam-break flow problem that addresses the shortcomings of existing analytical and numerical models. Tedla et al. (2021) used hourly hydrological and meteorological data as well as high-resolution land surface datasets to simulate design floods for piping dam failure with empirical dam breach methods. With the requirements of decision makers for high-precision flood prediction, identification of flood-prone areas, flood inundation mapping and inundation assessment, GIS, high-definition remote sensing images, high-precision digital elevation model (DEM), MIKE software and The Hydrologic Engineering Center – River Analysis System (HEC-RAS) have gradually been added to dam-break numerical modeling (Alvarez et al. 2017; Mao et al. 2017; Enea et al. 2018; Tamm & Tamm 2020; Mudashiru et al. 2021). A recent example is the work of Li et al. (2021) who used high-precision topographic mapping and DEM of the area around Qianping Reservoir in Ruyang County, Henan Province, China as the basis for one- and two-dimensional coupled dam-break numerical models using Geographic Information System (GIS) technology and MIKE software. Li et al. (2021) simulated the dam-break and flood discharge process under the 5000-year check flood level, calculated the flow process of the reservoir dam-break and the downstream evolution of the flood, obtaining flood risk information such as submerged area and flow patterns. Weilian et al. (2021) revealed the deficiencies of the existing systems for the floods simulation: the input data parameters are complex and the simulation process is cumbersome and time consuming. Then they developed a platform for the rapid 3D reproduction of dam-break floods which supports rapid reproduction at different times and positions. Based on Monte Carlo Simulation (MCS), HEC-RAS 2D, and HEC-LifeSim models, EL Bilali Ali et al. (2022) proposed an integrated approach on dam-break simulation, which can deal with the flood risk with probability rather than analyzing a few dam-break scenarios.

Most existing research focuses on a single dam-break scenario, and there is a lack of comparative analysis of flood process and downstream impact under different dam-break scenarios. Given the numerous potential causes of dam failure, there is potential for great differences in flood flow and downstream evolution with different dam-break scenarios. Single-scene dam-break flood analysis cannot fully meet the needs of practical applications; it is important to study the flood propagation process and downstream effects under various dam-break scenarios. This study explores the flood discharge and evolution process under three typical dam-break scenarios. Characteristic parameters such as breach flow, outflow time, time to peak, submerged range, flood velocity and flood arrival time are compared and analyzed for each scenario, and the disaster implications for downstream peoples and the structures are evaluated. The results are of significance for reservoir management departments to formulate flood control emergency plans, analyze flood risks and provide decision support for emergency agencies. The work also has reference value for government to improve regional flood control and disaster reduction capabilities and to enhance people's awareness of risk avoidance.

Pingshuijiang is a medium-sized reservoir with comprehensive functions of flood control, irrigation, water supply and power generation. The dam site is about 20 km from Shaoxing City, along the main stem of the Pingshui River. The river is a mountain system, with short and scattered tributaries. Floods are mainly generated by typhoon rainstorms and plum rain, which rise and fall rapidly and generally last for about 3 days. For dam-break floods, the potential area influenced downstream is about 357 km², including Pingshui Town and Yuecheng District. The northern part of the reservoir is adjacent to Pingshui Town, which is in a basin and is surrounded by mountains. If a dam-break occurs, flood waters can only be discharged through the Kuaijishan gorge in the northern part of the town, and are directed to the Jishan Street area of Yuecheng District. This kind of reservoir is common in the economically developed areas of the Yangtze River Delta. It is of great significance to study the dam-break flood of this kind of reservoir for the safety risk analysis of the whole Yangtze River Delta region.

A range of data sources were used in the study to provide input for the simulation, including information on terrain and reservoir parameter characteristics. For terrain data, a 2 × 2 m DEM data of Shaoxing City was utilized. The layer structure of the basic risk map geographic information database was established using the spatial database of ArcGIS, and the processed DEM and related data were then constructed. The database coordinate system is CGCS2000, on the Gauss–Kruger projection, and the elevation reference is the China national elevation datum 1985. Baseline information on reservoir and dam characteristics was obtained from the Pingshuijiang Reservoir Control Operation Plan, as shown in Table 1.

A dam-break flood model was constructed to determine the breach flood process and downstream evolution of the flood wave. The model has two elements: the breach flow process and the flood routing process.

Considering the most unfavorable situation where the breach develops vertically to the bottom of the dam and develops at a uniform speed, the longitudinal development process of the breach can be obtained by analogy using Equation (1).

For flood routing, the MIKE21 two-dimensional hydrodynamic calculation model was used to simulate the flood evolution and inundation process downstream of the reservoir. The model generalizes the flood-affected area, using a triangular mesh as the basis for simulation. As Pingshui Town as the key research object, the mesh was encrypted. The 357 km² area was divided into 69,253 grids cells, with terrain data use to assign the grid elevation. The upper boundary of the model was set at Pingshuijiang dam site or the auxiliary dam site, according to the scheme chosen for simulation. The lower boundary was set using closed features such as Shaozhu Expressway, Cao ‘e River and Sanjiang River. The calculation time step was set as 1 min, and the flood routing process was determined over 36 h.

Failure of earth-rock dams may be gradual, due to flood infiltration and overtopping, or instantaneous, for example, due to strong external damage. Based on the regional background, engineering status and external factors, three dam-break scenarios for the design flood level (48.25 m) were formulated and compared for Pingshuijiang Reservoir. Details of each scenario are given in Table 2 and outlined below:

• (1)

  Gradual collapse of the main dam. As the water gradually rises to the design flood level during the event, water penetrates and erodes the main dam, resulting in its gradual collapse. This generates a large discharge flood in a short time. The upper boundary of the model is set as the breach flow process at the main dam site.

• (2)

  Instantaneous complete collapse of the main dam. This scenario considers the case where the reservoir is at the design flood level and is affected by strong destructive factors, such as earthquake activity or war, so that the main dam fails instantaneously. This results in a huge discharge flood in a very short time. The upper boundary of the model is set as the breach flow process at the main dam site. Due to the rapidity, there is no need to consider the breach development process in this scenario. The average width of the dam is 230 m, and the flow process after dam-break can be calculated directly using Equations (4) and (5).

• (3)

  Instantaneous complete collapse of auxiliary dam. This scenario considers the case where the water level is at the design flood level, but there is continued rainfall that may lead to flooding, so the auxiliary dam is artificially blown up to speed up flood discharge and ensure the safety of the main reservoir. This results in a small discharge flood in a very short time. The upper boundary of the model is set as the breach flow process at the auxiliary dam site. The average width of the auxiliary dam breach is 74 m.

In Scenario 2, flood discharge peaks in the initial stage at 29,632 m³/s, with a spillway discharge of 485 m³/s (Figure 3(b)). The huge flow magnitude causes the reservoir water level to drop rapidly so that 10 min after the dam-break, water level is lower than the elevation of the spillway, and flood discharge from the spillway ceases. After about 30 min, discharge is around 9,000 m³/s and then gradually decreases to 111 min when the event is essentially completed.

Under Scenario 3, all floodwater is dumped at the auxiliary dam and discharge peaks at 2,589 m³/s at the initial stage (Figure 3(c)). Due to the relatively small discharge, flood duration is significantly longer (780 min), and the final reservoir water level is maintained at 38.7 m, which is the bottom elevation of the auxiliary dam.

Flood duration in Scenarios 1, 2 and 3 is 249, 111 and 780 min, respectively. The longest flood duration is associated with instantaneous failure of the auxiliary dam (Scenario 3), which is 3.1 times that of Scenario 1 and 7 times that of Scenario 2. The main reason for this is the relatively small breach of the auxiliary dam, which allows only small volumes of outflow and the flood takes longer to be fully discharged.

Based on the breach flow of each Scenario, combined with MIKE21 model analysis, the inundation range and downstream flood process were derived. The main characteristics of flood inundation under each Scenario are shown in Table 4. The results show that all three dam-break flood Scenarios would have a direct impact on most towns and streets in Pingshui Town and Yuecheng District. The area of inundation and water depth, and the process of inundation are compared and analyzed below.

Figure 7(b) shows changes in submergence depth over time at Jishan Street, Langgang New Village. Here, flooding commenced at 150, 90 and 200 min after dam-break in Scenarios 1–3, respectively. The maximum submerged range was reached after about 10, 8 and 11 h, respectively.

This study compares and analyzes the downstream impacts of different dam-break scenarios for Pingshuijiang Reservoir in southeastern China. The downstream flood routing process is simulated for three scenarios, with flood characteristics such as submergence depth, inundated area and arrival time calculated. Flood development dynamics and regional flood impact are evaluated, and emergency rescue conditions are analyzed. The results show significant differences in flood process and submergence under the three scenarios due to dam-break size and dam-break mode.

• (1)

  Scenario 1 involves the gradual failure of the main dam. Though peak discharge reaches 11,440 m³/s, which almost submerges the whole of Pingshui Town, and has a direct impact on most township streets in Yuecheng District. However, flooding occurs gradually, and submerged water depth is largely below 1 m within 30 min of dam-break, and it takes another 30 min to submerge the main area of the town. Hence, there is sufficient time to warn and evacuate people living in the affected areas.

• (2)

  Scenario 2 involves the instantaneous complete collapse of the main dam. Peak discharge is significantly larger than in the other scenarios (30,117 m³/s), and occurs at the beginning of the dam-break. The flood reaches Pingshui Town after 7 min and the town is submerged within 20 min. The rapidity and extent of flooding would have a devastating effect on the whole town, with no time for early warning and immediate rescue/evacuation.

• (3)

  In Scenario 3, instantaneous failure of the auxiliary dam, peak discharge is much lower at 2,589 m³/s due to the relatively small breach and low rate of flow. Although the flood reaches the town in 12 min and submerges the town after 45 min, the water depth is generally less than 2 m. Hence, there is ample opportunity for rescue and evacuation.

There are two shortcomings to the study that could be addressed in future work. Due to the uncertainty of upstream rainfall conditions, inflow to the reservoir during dam-break is not considered. In addition, to fully consider the impact of the most unfavorable dam-break condition, the main dam instantaneous collapse scenario was adopted; however, this scenario is extremely rare for large and medium-sized earth-rock dams.

X.J. conceptualized the whole article; X.J. and C.Z. developed the methodology; X.J. and J.M. wrote the original draft; C.Z. and Q.X. conducted funding acquisition; J.M. and X.J. performed the investigation; X.J. and B.F. conducted data curation; C.Z. and D.M. visualized the article; Y.Z., Q.X. and C.J.Z. wrote the original draft and the review, and edited the article. All authors have read and agreed to the published version of the manuscript.

This study was sponsored by the Science and Technology Development Foundation of Yellow River Institute of Hydraulic Research (Grant No. HKF202202), the National Natural Science Foundation of China (Grant Nos. U2243220, 51909100 and 52009047) and the Fundamental Research Funds for the Central Scientific Research Institute (Grant No. HKY-JBYW-2020-03).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.