Comparison of the Damaged Area Caused by an Agricultural Dam-Break Flood Wave Using HEC-RAS and UAV Surveying

This study examines the usability of unmanned aerial vehicle (UAV) data surveyed just after an agricultural reservoir collapse by comparing the survey results with the simulation results of the HEC-RAS (Hydrologic Engineering Centers River Analysis System) flood wave propagation to the downstream areas. A 61,400 m storage dam broken by 89.0 mm (over 30.0 mm/hr rainfall intensity) of rainfall on August 21, 2014 was considered. The reservoir water capacity curve and downstream damaged areas were estimated by drone surveying 3 days after the dam break. The flood wave by the overtopped dam break was propagated using the HEC-HMS (Hydrologic Engineering Centers Hydrological Modeling System) reservoir inflow from the watershed. The model results showed flood inundation depths of 0.1 to 2.2 m, mainly in rice paddy areas along the stream, and the overtopped dam-break scenario exhibited 59% correspondence with the drone-surveyed areas.


Introduction
Dam break is a serious devastating disaster that occurs around the world and causes large amounts of damage to people worldwide [1].In South Korea, there are 17,531 agricultural reservoirs, with 88.4%, 9.1%, and 2.5% having storage capacities of less than 0.1 million m 3 , between 0.1 and 1.0 million m 3 , and over 1.0 million m 3 , respectively.These reservoirs now hold 55% of the irrigation water among the total supply to 447,000 ha, with a total capacity of 2.78 billion m 3 .damaged areas estimated using many approaches.Carling et al. [2] conducted dam-breach modeling with 1-and 2-dimensional hydraulic models that were specially developed for ice dam failures via piping and overtopping.Sarhadi et al. [3] presented a methodology to assess floodplain mapping at ungauged rivers via 1-dimensional hydraulic inundation modeling.They used high-resolution digital terrain models (DTMs), land use characteristics, and channel properties to identify the critical and vulnerable areas affected by floods in different return periods.You et al. [1] reviewed earth-rock dam safety using high-resolution digital surface model (DSM) data.Evangelista et al. [4] presented the application of a multi-stage first-order centered scheme GMUSTA (Generalized MUlti STAge) to solve a two-phase flow model with four equations.Prakash et al. [5] attempted to simulate the impact of dam-wall failure scenarios using the Smoothed Particle Hydrodynamics (SPH) method on flood inundation, including modeling of dam-wall fragments.
The prediction of flood magnitude at ungauged reaches is an important task in designing river engineering and hydraulic structures and remains a fundamental challenge for hydrologists [6].Although considerable research has been conducted based on available information for dam-break problems, the lack of data from ungauged regions still remains an issue in providing accurate predictions of flood magnitude, which is a key factor for reliable flood inundation mapping [3].The most important data required to perform hydrologic and hydraulic dam-break simulations relate to the topographic relationship between the reservoir water level and storage and the cross sections of the stream, including protected lowlands.
A high-resolution DTM or DSM is critical to extract the details of the channel topography, obtain the water surface elevations at the cross sections, and simulate flood inundation and depth [3] [7].Many studies have considered the HEC-RAS (Hydrologic Engineering Centers River Analysis System) performance by high-resolution DSM extracted from new data sources, such as Light Detection and Ranging (LiDAR) and Synthetic Aperture Radar (SAR) data [8]- [13].In addition to these various studies, the unmanned aerial vehicle (UAV) remote sensing is an easy and increasingly popular approach to obtain high-quality terrain data.The UAV, also known as a drone, is an aircraft combined with a ground control station and communication support equipment that is flown without a pilot on board and can be remote controlled automatically or semi-automatically by using preprocessed programs or artificial intelligence.
Recently, many studies on dam break have been performed, with the resulting damaged areas estimated using many approaches.

Study Area and Input Data
The broken dam is located in the southeastern part of South Korea, with a latitude and longitude of 35.895˚N and 128.96˚E, respectively.The Goeyeon reservoir dam was built in 1945 for rice paddy irrigation water supply to the downstream area.The dam was broken by the 92.6 mm rainfall event with over 30 mm/hr rainfall intensity in August 21 st , 2014 after reaching the maximum water level to the bank top.The cause of reservoir collapse was predicted as overtopping by the rain storm, but was not certain with piping possibility.Figure 1 shows the location of the reservoir, the part of the broken dam, and the cut section of the bank.Figure 2 shows the GIS input data of the upper watershed of the reservoir for the HEC-HMS modeling.

UAV Specification and Imaging Process
A rotary-wing drone of 3.5 kg weight carried with the Sony NEX-5T sensors was used to acquire the image data (Figure 3).It has 4 cm by 4 cm spatial resolution with JPEG image format.Table 1 shows the specification of the drone.The autopilot flight was applied by the inertial measurement unit (IMU) and onboard GPS.The integration of sensors with IMU and GPS enabled to obtain the geo-referenced images.For post-processing, the DSM was produced by the terra    In order to orient and relate the drone images to the ground, twenty GCPs (Ground Control Points) were investigated and four checking points were arranged for uniform horizontal and vertical accuracy.With 60% end laps as well as 60% side laps, the drone routed two times for 0.43 km 2 areas with total 338 images of first 153 images and second 185 images respectively.In this study, calibrated and validated UVA geospatial data obtained from the Korea Rural Development Administration (KRDA).Details regarding the estimation process of the UVA geospatial data can be found in Park and Park [18].

HEC-HMS and HEC-RAS Model Theory
HEC-HMS and HEC-RAS were used to develop the dam break and inundation modeling.Numerous previous studies have shown these models to provide accurate and useful results in flood-related studies [19] [20].
The watershed runoff to reservoir was modeled using HEC-HMS.
( ) where Q is the runoff, P is the rainfall, S is the maximum potential reten- The flood wave is calculated by the following kinematic wave equation and watershed parameters (K, T c ) are calculated by Kraven and Sabol equations.
where Q is the flow discharge (m 3 /s), i V is the average velocity at cross-section i (m/s), i A is the area at cross-section i (m 2 ), R is the hydraulic radius (m), n is Manning's roughness coefficient, C is the conveyance (m 5/3 ), f S is the av- erage friction slope between adjacent cross-sections, A is the watershed area (km 2 ), L is the river length (km), S is the watershed slope, K is the storage constant (hr) from Kraven equation, and c T is the concentration time (hr) from Sabol equation.

Dam-Break Simulation Using HEC-HMS
The dam break was established using HEC-HMS, developed by the US Army Corps of Engineers.HEC-HMS allows the modeler to choose two different methods for computing outflow through two dam-breach options: overtopping and piping.
Two cases were applied in this study because the break cause was unknown.
The 'overtopping dam break' is designed to represent failures caused by overtopping of the dam.This failure is most common in earthen dams but may also If a sufficient amount of material erodes, then a direct piping connection may be established from the reservoir water to the dam face.Once such a piping connection is formed, it is almost impossible to stop the dam from failing.The method begins the failure at a point in the dam face and expands it as a circular opening.When the opening reaches the top of the dam, it continues expanding as a trapezoidal shape.The flow through the circular opening is modeled as orifice flow, whereas it is modeled as weir flow in the second stage [22].
To apply dam break model, we need information on the collapse width, slope, and shape for collapse duration, cause of collapse, inflow hydrograph, and downstream cross section.The Goeyeon reservoir is a small reservoir and has no specifications related to the reservoir.downstream topographic data were collected from a national [24].Figure 5 shows the flowchart of the study.

Digital Surface Model and Elevation-Storage Curve: UAV Observations
Using UAV images taken by Park and Park [18] to create accurate topographic and flood trace maps.In addition, we created a precise topographic map by using Google images before the collapse of the reservoir.These two topographical maps (Figure 6(b)) were used to compare and analyze the situation before and after the collapse of the reservoir.In particular, the elevation-storage curve was calculated via the DSM using Equations (( 8) and ( 9)) (Figure 7 and Figure 8).
The UAV reservoir terrain detection was possible because the land surface was exposed by the dam break.In this study, the relationship between the reservoir surface area and the reservoir storage was derived by using the DSM from UAV image, Equation (10), and Equation (11) (Figure 8).
( )  width × 1 m depth), the depth value of cell is equal to its volume.
The reported irrigation area of the reservoir is 125 ha and the effective storage capacity is 61,000 m 3 .Because of the lack of basic data such as dam minimum and maximum water levels in reservoir, we estimated water level in reservoir using accurate topographic map.The estimates of the topographic maps are overestimated.In order to calibrate this, we have derived the water level-water surface relation by setting the reported effective storage capacity of 61,000 m 3 to the maximum water level.As a result, the minimum water level is EL.176.77 m.
The area and the storage capacity of each water level are shown in Table 3 and     were adjusted using actual flood map from UAV images.Table 4 shows the final watershed parameters and watershed characteristics.

HEC-HMS Dam Breach and HEC-RAS Inundation Modeling
Figure 9 shows the discharge of water with 2 dam-break scenarios; (a) overtopping and (b) piping.According to the national hazard investigation report [24], the initial reservoir water level was set to 98% of full water level.The duration time of dam failure was set to 30 min.Under the condition of maximum 61,400 m 3 reservoir storage amount pouring to the downstream area, the peak discharge was up to 107.9 m 3 /sec by overtopping and 140.2 m 3 /sec by piping failure, respectively.To calculate the downstream inundation area and depth, HEC-RAS used the HEC-HMS flood wave discharges of Figure 8 as the boundary conditions, and the results were represented by HEC-GeoRAS. Figure 9 shows the estimated flood mapping results with two dam-break scenarios.To evaluate the performance of the model, the F statistic used Equation ( 12) to compare the drone and HEC-RAS inundation areas [3] [20] [25].
where o A is the UAV observed area of  F statistic varies from "100" when observed and predicted areas coincide perfectly to "0" when there is no overlap between the predicted and observed areas.
The inundation map indicated using UAV is the actual observation data, so it can be seen as flood trace.Two methods were used to evaluate flood inundation according to the collapse runoff of Goeyeon reservoir for the most suitable floodplain.The first is a simple comparison by area, and the second is the comparison of the flood trace using the Lee Sallee Shape Index (LSSI) method.The LSSI method is a method of measuring the degree of agreement between two data by overlapping the reference data with the measurement data [26].In other words, the intersection area between the two data is divided by the union area, and the value is calculated as the generalized index.The index means indicator that measures the degree of spatial alignment between two data.The reference data in this study is flood trace from UAV and the measured data is each scenario result (Table 5).Figure 10 shows the UAV observed damaged areas with

Figure 1 .
Figure 1.Location of the study reservoir: (a) Digital Elevation Model, (b) Front view of the broken dam, (c) Side view of the dam cut end.

Figure 2 .
Figure 2. Reservoir watershed data for HEC-HMS modeling: (a) Land use and (b) Soil texture.

Figure 3 .
Figure 3. Location of the study watershed and locations of the weather, streamflow, and water quality gauging stations.
-HMS was developed by the US Army Corps of Engineers and designed to simulate the rainfall-runoff processes of dendritic watershed systems.During the runoff process, the infiltration capacity (I a ) is quantified by the Soil Conservation Service-Curve Number (SCS-CN) based on land use and hydrologic soil group.The runoff is calculated from the following set of empirical equations: C.-G. Jung, S.-J.Kim DOI: 10.4236/as.2017.8100791094 Agricultural Sciences tion, a I is the initial abstraction, and CN is the runoff curve number.The flood wave propagation by dam breakage was modeled using HEC-RAS.HEC-RAS calculates one-dimensional steady and unsteady flow, and the model equations are described by Horritt and Bates [21].The system requires the output hydrograph from the HMS as input, and the parameters are representative cross-sections, including left and right bank locations, Manning's roughness coefficients, and contraction and expansion coefficients.To develop the floodplain maps, a digital elevation model (DEM) was formed by channel geometry.
occur in concrete arch, concrete gravity, or roller-compacted dams as well.The failure begins when the appreciable amount of water begins flowing over or around the dam face.Another method is designed to represent failures caused by piping inside of the dam, called "piping dam-break".This failure typically occurs only in earthen dams.The failure begins when water naturally seeping through the dam core increases in velocity and a sufficient quantity of fine sediment C.-G. Jung, S.-J.Kim DOI: 10.4236/as.2017.8100791095 Agricultural Sciences moves out of the soil matrix.

Figure 4 . 2 . 5 .
Figure 4. Location of the study watershed and locations of the weather, streamflow, and water quality gauging stations.

Figure 7 .
Figure 7. Reservoir water surface areas to estimate reservoir volume using drone derive DSM for reservoir elevation-area-volume graph (a) to 2.0 m (elevation: m), (b) to 2.4 m, (c) to 2.8 m, and to 4.0 m from reservoir bottom as zero.

Figure 8 .
Figure 8.The estimated relationship between reservoir elevation-area-volume a graph using drone derived DSM.

Figure 9 .
Figure 9. Flood wave discharge graph by dam-break using HEC-HMS: Case of (a) Overtopping and (b) Piping.

9 A:Figure 10 .
Figure 10.The HEC-RAS simulated inundation areas and depth distributions in case of (a) Overtopping and (b) Piping dam break conditions.
The technology of UAV remote
3D program.The DSM points converted to CAD file using the Virtual Surveyor Tools and then extracted contour line by using the Civil Pro program.

Table 2 .
Table 2 summarizes the data of the collapse width and slope data collected through various media.According to the Yeongcheon station from the Meteorological Agency, cumulative rainfall from Reservoir specification and collapse observation information.
7:00 am on August 20, 2014 to 9:00 am on August 21, 2014 is 92.6 mm, and it is confirmed that rainfall has ended before 9:00 am on August 21.Therefore, the reservoir level at the time of the collapse was a high water level, and we applied to the dam break model assuming no additional inflow.Figure4shows the rainfall hyetograph at Yeongcheon station during the collapse period.In this study, two collapse scenarios such as piping and overtopping were applied by fixing the collapse width (20 m), and slope (1:1) and considering construction year.

Table 3 .
Estimated water level-surface area and storage capacity.

Table 4 .
Reservoir specification and collapse observation information.
inundation, p A is the HEC-RAS-predicted area of inundation, and op A is the overlapped area between o A and p A .The

Table 5 .
Results of dam-break scenarios.