Dammed lakes are an important secondary hazard caused by earthquakes. They can induce further damage to nearby humans. Current hydrology calculation research on dammed lakes usually lacks spatial expressive ability and cannot accurately conduct impact assessment without the support of remote sensing, which obtains important characteristic information of dammed lakes. The current study aims to address the issues of the potential impact area estimate of earthquake-induced dammed lakes by combining remote sensing (RS), a geographic information system (GIS), and hydrological modeling. The Tangjiashan dammed lake induced by the Wenchuan earthquake was selected as the case for study. The elevation-versus-reservoir capacity curve was first calculated using the seed-growing algorithm based on digital elevation model (DEM) data. The simulated annealing algorithm was applied to train the hydrological modeling parameters according to the historical hydrologic data. Then, the downstream water elevation variational process under different collapse capacity conditions was performed based on the obtained parameters. Finally, the downstream potential impact area was estimated by the highest water elevation values at different hydrologic sections. Results show that a flood with a collapse elevation of at least 680 m will impact the entire downstream region of Beichuan town. We conclude that spatial information technology combined with hydrological modeling can accurately predict and demonstrate the potential impact area with limited data resources. This paper provides a better guide for future immediate responses to dammed lake hazard mitigation.

Dammed lakes formed by earthquake landslides are an important secondary hazard induced by earthquakes. Damming objects, sliding from mountain slopes, comprise rocks and weathered soil, which have strong penetrability, low stickiness, and high instability. Damming objects easily collapse and threaten human beings who live downstream [

Remote sensing techniques are widely applied for extracting water bodies and monitoring floods [

Remote sensing (RS) was also used to survey the hydrological variations caused by earthquakes. Hsu et al. [

Meanwhile, several studies on dam-break have been promoted recently. They combined a hydraulic empirical model and a physical model to improve the forecasting accuracy of dam-break floods [

Summarizing the above studies, we conclude that many studies on dammed lakes only use the RS technology that focuses on the current situation, such as the location of the damming object and the current inundated area; these methods have failed to consider future trends. Several GIS studies about dammed lakes have an excellent spatial expressive ability to obtain the reservoir storage, the submerging depth, and the submerging area. However, most of them simply deal with water elevation problems wherein the water surface is assumed to be flat. The hypothesis is approximately appropriate to the upstream of the dammed lake, whereas the hypothesis is not obviously reasonable for area downstream. Hydraulic calculations can exactly forecast the future flood routing, but they are usually not combined with RS and GIS and lack a spatial expressive ability. Moreover, hydraulic calculation is time-consuming and requires massive boundary conditions, whereas the dammed lake formed by a sudden landslide usually lacks complete data to support the complex hydraulic calculations. Dammed lake disasters must take full advantage of limited data resources to conduct emergency response in a short time. The response must effectively and rapidly estimate the corresponding potential impact area based on different given dam-break conditions. Therefore, this study combines RS, GIS technologies, and hydrologic data to establish an effective method to estimate the impact risk of dammed lakes in the corresponding downstream regions under different water elevation conditions. The study in this paper can be widely applied to different earthquake-induced dammed lake cases because it only requires DEM, historical hydrological data, and current RS images that are easy to obtain.

On 12 May 2008, a magnitude 8.0 earthquake occurred in Wenchuan country, Sichuan province, China. The earthquake induced a massive mountain landslide that destroyed the natural river discharge function and formed 33 large dammed lakes. The largest dammed lake, Tanjiashan dammed lake, was selected as the study area. Tanjiashan is located at the upstream of the Xun River and is only six kilometers from Beichuan town, which is the largest town downstream from the dammed lake and was also severely afflicted by the Wenchuan earthquake disaster. Thus, Tanjiashan dammed lake has the highest risk level among all of the dammed lakes.

The experimental data mainly include DEM, various satellite RS images, aerial photographs, and historical hydrologic data (observed at Tongkou and Xiangshui hydrologic stations) (

The research process is demonstrated in

The dammed lake’s characteristic parameters, including its location, current submerged area, storage volume, and other geometric characteristics, can be extracted based on the DEM data obtained before the earthquake and the RS images acquired after the earthquake. The dammed lake’s geometric parameters, including the location, size, and area of the damming object, were acquired by visual interpretation from Formosat-2 images. The reservoir storage capacity curve was calculated based on the DEM when the location of the dammed lake was known. The elevation versus reservoir capacity curve is essential in forecasting the tendency of the downstream potential submerged area that affects the estimation of the submerging loss and risk avoidance decision. We applied the eight neighborhoods seed-growing algorithm [

The principle of the algorithm is to locate damming objects from RS images acquired after the earthquake, where a line segment across the river channel is set as a pretend dam body. A seed origin at the center of the upper dammed lake is selected to spread on the DEM below a given elevation. In designing the algorithm, the elevation value of the line segment across the river channel can be first set to an invalid value (Null) or a maximum (such as elevation = 99,999). When a seed grows nearby the dam site, it cannot cross the dam and proceed downstream because of the existence of the invalid values. This method is similar to the clustering method in the image process. It is an iterative or recursive judgment process for computer programming. It repeatedly judges whether the given elevation _{0} is higher than seed’s eight neighborhoods. If the elevation of any neighborhood is lower than _{0}, then the neighborhood (pixel) would be viewed as a submerged grid and pushed to the top of the stack. Then, the pixel on the top of stack will be selected as the current seed. If no neighborhood satisfies the condition (lower than

Finally, the number of grids, once pushed into the stack, represents the submerged area. The accumulated volume is the reservoir capacity corresponding with the

After a group of given water level values were assumed to calculate the corresponding water surface area and reservoir capacity with DEM data, the elevation versus area and reservoir capacity curve was established by the above calculation results.

The weir flow empirical formula (Equation (2)) was adopted to calculate dam-break flow:

The Muskingum flood routing method was adopted in this paper to calculate the potential impact area downstream because of its accuracy and simplification.

The principle of the Muskingum method can be described by the difference form of continuous equation and the channel storage equation [

Through simultaneous of Equations (4) and (5), the Muskingum flood routing equation is defined as follows:

The key of the Muskingum flood routing method is determining the equation parameters. The traditional method of solving equation parameters is a tentative calculation algorithm with low efficiency and low precision. Simulated annealing (SA) [

The algorithm structure is shown in

The SA method is independent of an initial value and asymptotic convergence. It improves on the traditional Muskingum tentative calculation and avoids solving middle parameters

When the runoff peak spreads through downstream hydrologic stations, the submerged area achieves a maximum (potential impact area). The Muskingum flood routing result is the indirect runoff information, not the elevation information itself. Thus, the flow must be translated into a corresponding elevation by the runoff versus elevation curve acquired from survey data polynomial fitting. The polynomial fitting on every hydrologic station section represents the relationship between the runoff with the water level based on long-term observation data. The quadratic or cubic equations usually fit the runoff versus elevation curve very well for mountainous rivers, most of them having simple regular fluctuations.

Through the Muskingum flood routing method and the runoff versus elevation curve, we obtained all of the maximum water levels at every hydrologic station section. However, the hydrologic stations were limited, and the downstream water elevation was descending along the river channel. Therefore, this study adopted the distance interpolation method to deal with the maximum water level of the control sections (including dam, hydrologic stations) and generated a series of dense elevations in all of the segments along the river. The river channel was divided into many segments. All of the segments were manually divided by approximately equal distances and were vertical to the river channel. The maximum water level at each section was interpolated by distance and the known water level.

The potential impact area at every segment was generated by the dense elevation values and the DEM data. We used the seed growing algorithm identical to the one used when calculating the elevation versus reservoir capacity curve. The calculation formed a series of segment documents. The entire impact area downstream was generated by overlaying all of the segment documents. The submerged elevation at every segment was viewed as flat. Thus, the junction about the adjacent segments was not smooth. Sometimes we could use enveloping line or smooth algorithm to obtain the smooth submerging boundary. Once the impact area was known, we could generally analyze the submerging depth and area and evaluate the disaster population, economy loss, and other essential damage information by overlaying the impact area, DEM, and RS image.

According to the results interpreted by the multisource remote sensing images (Formosat-2, Aerial photograph.), the location of Tangjiashan dammed lake is approximately 104°24′27″ E and 31°51′30″ N. The shape of dammed objects is rectangular. The length along the river is approximately 803 m, and the width is 611 m. Combining RS images with the DEM shows that the top of the damming objects is 775.0 m on the right bank of the river and 793.9 m on the left.

When basic spatial information about Tangjiashan dammed lake is known, we assume a group of characteristic water level values and apply the eight neighborhoods seed growing algorithm to calculate the reservoir storage capacity curve based on the DEM data. The elevation-area-reservoir capacity curve and submerged range are shown in

The weir flow empirical formula (Equation (2)) and continuity equation (Equation (3)) were adopted to calculate dam-break flow. In this calculation example, the water elevation

The result demonstrates that the entire flood process finished in a short time (approximately 60 min), and that the flow suddenly decreased to the upstream natural inflow whose value adopted the observation average (76 m^{3}/s). The dam-break period (approximately 17 min) can be neglected when compared with the time when the flood spreads in the channel. Meanwhile, the potential impact area concerns the flood peak; thus, we can select the dam-break flow maximum (_{0} = 83,620 m^{3}/s) as the initial inflow.

The Muskingum Parameters were trained by abundant historical data acquired from hydrological observation stations (Tongkou and Xiangshui stations). When the cost function convergence error in the SA method was stable, the Muskingum parameters in Tongkou and Xiangshui achieved optimization solutions (

The measuring runoff data after the earthquake was selected to validate the Muskingum parameters. The flood that occurred on 10 June 2008 was used as a test sample to check the accuracy of the Muskingum parameters. The Muskingum flood routing equation (Tangjiashan to Tongkou) is expressed as follows:

The field survey data _{2}, _{1}). The 23 m^{3}/s value is viewed as the initial outflow condition (_{1}) at Tongkou station. The outflow sequences could be calculated according to the Equation (8). Comparing outflow sequences with field survey data

Similarly, the flood that occurred on 12 June 2008 was used to check the Muskingum parameters of Xiangshui section. The Muskingum flood routing equation (Tongkou to Xiangshui) is expressed as follows:

The process of calculation is the same as the above. The data table is abbreviated because of the length of paper. The coefficient test result at Xiangshui station is shown in

Once a dammed lake is generated, it blocks the surface runoff, thus leaving only a small base flow in the river. The base flow primarily generated by groundwater is a small, stable value. It can be acquired by observation after the formation of a dammed lake (dammed lake 3.2 m^{3}/s, Tongkou 4 m^{3}/s, and Xiangshui 2 m^{3}/s). They are the initial inflow and outflow in Muskingum flood routing. The flood routing result (

The runoff is plotted in

Finally, the study adopted the distance interpolation method to deal with 700 m in dammed lake, 565.89 m in Tongkou, and 537.47 m in Xiangshui and to generate a series of dense elevations in all the segments along the river (

When the water elevation

From the submerging result combining the DEM and RS images, we find that a given dam-break flood forming after an earthquake can destroy the entire Beichuan Town, which was already severely afflicted by the Wenchuan earthquake disaster. Based on downstream impact research, quickly estimating the loss of dam-break flood disasters is possible by using a GIS overlay analysis function that can generally analyze the population, economy, submerging depth, and other essential information.

Usually, flood routing using hydraulics methods requires complicated grid calculation and various boundary conditions. Consequently, the hydraulics methods are time consuming and cannot provide a series of submerged areas corresponding to different water levels in a short time. The study in this paper obtained potential impact areas with scarce data resources (DEM, historical hydrological data, and current RS images). Furthermore, the improved Muskingum method could simplify the calculation process and significantly improve efficiency. It could rapidly forecast the downstream submerged area corresponding with different dam-break elevations.

According to the similar dam-break mode and calculation method, we assumed a group of dam-break elevations (

After an earthquake, the communication and transportation in a disaster region is cut off. Traditional land measurement cannot immediately provide first-hand estimation under such conditions. Accompanying the development of spatial techniques, rapidly estimating the impact of dammed lakes, one of the secondary disasters induced by earthquake, is possible. This paper proposed a method to respond immediately to dammed lakes and obtain their corresponding potential impact area by using a DEM, historical hydrological data, and current RS images. The proposed method can be used to survey the location, quantity, and scale of dammed lakes, and to estimate the potential submerging loss. The proposed method can forecast the upstream potential impact area according to different water levels and downstream potential impact areas according to different dam-break styles without requiring a field survey. Its time-efficiency is important in selecting transfer routes and generally assessing disaster loss. It is also helpful in providing rapid decision support to choose reasonable schemes and to effectively decrease the harm of dammed lakes. Research on rapid estimation for potential impact areas of earthquake-induced dammed lakes can satisfy the requirements of dammed lake management.

This paper is financially supported by the National Natural Science Foundation of China (Nos. 51409021, 51509007 and 51379023), and Basic Scientific Research Operating Expenses of Central-Level Public Academies and Institutes (CKSF2016033/GC). The authors also greatly appreciate the anonymous reviewers and academic editor for their careful comments and valuable suggestions to improve the manuscript.

Shengmei Yang and Bo Cao conceived the study; Bo Cao performed the experiments; Bo Cao, Shengmei Yang, Song Ye wrote the paper.

The authors declare no conflict of interest.

Study area and data.

Research flow chart.

Simulated annealing algorithm flow chart.

Dammed lake area in different water level.

Tongkou station coefficient test result.

Xiangshui station coefficient test result.

Muskingum flood routing results and elevation interpolation method. (

Potential impact area of the dammed lake downstream (

Potential impact area of dammed lake downstream (

Specifications of the satellite remote sensing image, aerial photograph, and satellite radar data used in this study.

Data | Acquisition Date | Resolution | Format | Region |
---|---|---|---|---|

Formosat-2 | 20 May 2008 | 2 m | TIFF | Tanjiashan Dammed Lake |

Aerial photograph | 18 May 2008 | 0.15 m | TIFF | Beichuan Town |

Envisat | 20 May 2008 | 30 m | TIFF | Tanjiashan Dammed Lake |

DEM | 2003 | 1:50,000 | TIFF | Xun River Basin |

Elevation-area-reservoir capacity curve.

Water Level (m) | Reservoir Capacity (10^{8} m^{3}) |
Area (m^{2}) |
---|---|---|

690 | 0.248 | 1.591 |

695 | 0.338 | 1.958 |

700 | 0.445 | 2.430 |

705 | 0.572 | 2.680 |

710 | 0.714 | 2.975 |

715 | 0.872 | 3.380 |

720 | 1.076 | 4.538 |

725 | 1.318 | 5.077 |

730 | 1.587 | 5.671 |

735 | 1.888 | 6.276 |

740 | 2.229 | 7.223 |

745 | 2.607 | 7.883 |

750 | 3.018 | 8.564 |

Dam-break flow calculation result (

Dam-Break Time | Flood Peak | Peak Appearance Time | Ending Flow |
---|---|---|---|

1 h | 83,620 m^{3}/s |
17 min | 76 ms^{3}/s |

Muskingum parameters.

Reach | C_{0} |
C_{1} |
C_{2} |
---|---|---|---|

Tangjiashan to Tongkou | 0.1722 | 0.1615 | 0.6663 |

Tongkou to Xiangshui | 0.5197 | 0.2819 | 0.1984 |

Coefficient test result between Tangjiashan and Tongkou section.

Time (h) Tangjiashan | Water Level (m) | Runoff ^{3}/s) |
Time (h) Tongkou | Water Level (m) | Runoff ^{3}/s) |
Flood Routing Result | Relative Error % |
---|---|---|---|---|---|---|---|

0:00 | 620.90 | 33.1 | 0:00 | 535.31 | 23 | ||

1:00 | 621.05 | 64.3 | 1:00 | 535.36 | 28 | 31.743 | +13.37% |

2:00 | 622.75 | 858 | 2:00 | 535.52 | 44 | 179.282 | +307.46% |

3:00 | 622.25 | 576 | 3:00 | 537.19 | 240 | 357.210 | +48.84% |

4:00 | 621.93 | 415 | 4:00 | 537.1 | 210 * | 402.496 | +91.66% |

5:00 | 622.03 | 465 | 5:00 | 538.27 | 594 | 415.279 | −30.09% |

6:00 | 622.62 | 780 | 6:00 | 537.33 | 333 | 486.114 | +45.98% |

7:00 | 625.90 | 3240 | 7:00 | 538.57 | 690 | 1007.800 | +46.06% |

8:00 | 629.54 | 6870 | 8:00 | 545.05 | 3300 | 2377.770 | −27.95% |

9:00 | 629.54 | 6870 | 9:00 | 545.65 | 3590 | 3876.827 | +7.99% |

10:00 | 629.54 | 6870 | 10:00 | 545.65 | 3590 | 3876.827 | +7.99% |

11:00 | 629.59 | 6930 | 11:00 | 547.16 | 4400 | 4885.981 | +11.04% |

12:00 | 629.23 | 6530 | 12:00 | 548.25 | 5060 | 5499.190 | +8.68% |

13:00 | 627.34 | 4640 | 13:00 | 549.20 | 5720 | 5517.713 | −3.54% |

14:00 | 626.94 | 4320 | 14:00 | 549.73 | 6210 | 5169.716 | −16.75% |

15:00 | 626.47 | 3800 | 15:00 | 548.80 | 5420 | 4796.622 | −11.5% |

Note: * Interpolation.

Muskingum flood routing between Tangjiashan and Tongkou.

Reach Parameters | _{2} (m^{3}/s) |
_{1} (m^{3}/s) |
_{1} (m^{3}/s) |
_{2} (m^{3}/s) |
---|---|---|---|---|

Tangjiashan to Tongkou | 83,620 | 3.2 | 4 | 14,402.6 |

C_{0} = 0.1722 |
76 | 83,620 | 14,402.6 | 23,114.1 |

C_{1} = 0.1615 |
76 | 76 | 23,114.1 | 15,426.3 |

C_{2} = 0.6663 |
76 | 76 | 15,426.3 | 10,303.9 |

76 | 76 | 10,303.9 | 6890.9 | |

76 | 76 | 6890.9 | 4616.7 | |

76 | 76 | 4616.7 | 3101.5 | |

76 | 76 | 3101.5 | 2091.9 | |

76 | 76 | 2091.9 | 1419.2 |

Muskingum flood routing between Tongkou and Xiangshui.

Reach Parameters | _{2} (m^{3}/s) |
_{1} (m^{3}/s) |
_{1} (m^{3}/s) |
_{2} (m^{3}/s) |
---|---|---|---|---|

Tongkou to Xiangshui | 14,402.6 | 4 | 2 | 7486.6 |

C_{0} = 0.5197 |
23,114.1 | 14,402.6 | 7486.6 | 17,557.8 |

C_{1} = 0.2819 |
15,426.3 | 23,114.1 | 17,557.8 | 18,016.4 |

C_{2} = 0.1984 |
10,303.9 | 15,426.3 | 18,016.4 | 13,278.1 |

6890.9 | 10,303.9 | 13,278.1 | 9120.2 | |

4616.7 | 6890.9 | 9120.2 | 6151.3 | |

3101.5 | 4616.7 | 6151.3 | 4133.7 | |

2091.9 | 3101.5 | 4133.7 | 2781.6 | |

1419.2 | 2091.9 | 2781.6 | 1879.1 |

Runoff versus elevation curve in Tongkou and Xiangshui.

Hydrologic Stations | The Runoff (y) vs. Elevation (x) Curve | Flood Peak | Elevation Maximum |
---|---|---|---|

Tongkou | y = 20.9186473861676x^{2} − 22,283.2134587244x + 5,934,150.94321418 |
23,114.1 m^{3}/s |
565.89 m |

R^{2} = 0.986247730187266 |
|||

Xiangshui | y = 7.26568228353629x^{3} − 11,429.4504515713x^{2} + 5,993,229.82457408x − 1,047,570,044.59771 |
18,016.4 m^{3}/s |
537.47 m |

R^{2} = 0.997800010971559 |