Dam breachfloods- downstream inundation analyses M. Hartnett,' T. Hayes,* B. Mangan,* o Department of Civil, Structural & Environmental Engineering, Trinity College, Dublin 2, Ireland, UK * ESB International, St Stephens Green, Dublin 2, ABSTRACT Many existing dams were constructed with little regard to the effects of a breach on downstream water levels. The failure of large and medium size dams, such as Teton Dam, however, has highlighted the need for downstream inundation studies. In this paper the authors present frequently used inundation prediction techniques and detail the model developed at the National Weather Service, US, DAMBRK. The application of this model and dimensionless graphs to a major hydroelectric scheme in Ireland is outlined and results discussed. INTRODUCTION In 1984, the Electricity Supply Board of Ireland, ESB, initiated a major safety review of all its hydro-electric dams. As part of this work it was decided to carry out inundation studies downstream of dams on the Rivers Lee, Liffey, Erne and Shannon. Before these studies were carried out ESB reviewed different flood routing techniques. Storage routing methods, such as the Muskingum-Cunge method, were considered to be of limited use in predicting time varying dam breach inundation levels at many cross sections in complex downstream valleys. One-dimensional flood routing techniques can be one of three types: kinematic routing, diffusion routing and dynamic routing. Kinematic techniques are used when inertia and pressure forces are negligible. When pressure forces are significant diffusion flood routing is employed to ascertain resistance forces in terms of the pressure forces. When both pressure and inertia forces are significent, then dynamic routing is necessary to quantify the resistance forces in the momentum equation in terms of the pressure and inertia forces. It was decided to employ dynamic routing, using the program DAMBRK, for the analyses proposed in these studies. Dynamic routing involves solving the Saint-Venant equations, which are described below. Sakkas [1] developed a method of predicting flood levels using dimensionless graphs for the US Army Corps of Engineers. This method produces reliable
126 Hydraulic Engineering Software results only when applied to prismatic channels and hence was not considered sufficiently accurate for this work. However, an application of the graphs is presented and results compared with DAMBRK output. GOVERNING EQUATIONS By considering the law of mass conservation for the control volume in Figure 1 the following expression can be derived Datum Figure 1: Control Volume An expression for conservation of momentum within the control volume can be obtained by applying Newton's Second law. The forces acting on the above control volume are described here. Firstly, vertical forces act on the water volume due to gravity. Friction forces are caused by shear stresses resisting the fluid movement along the bottom and sides of the river reach. Due to nonuniform flow conditions, we must include hydrostatic pressure forces. Wind shear forces are caused by energy transfer from wind blowing across the surface of the control volume. The rate of momentum inflow to the control volume is Imam = p[(3vq + (W, qdx] and the rate of momentum outflow from the control volume is Ignoring lateral inflows, and wind stress effects and assuming (3=1, we can write the conservative form of the momentum equation as 5x 5t 5x 5x Equations (1) and (2) are known as the Saint-Venant equations and have found wide application in dynamic flood routing problems. The form of the Saint-Venant equations as derived here are applicable when a flood wave (2)
Hydraulic Engineering Software 127 propagates through theriverchannel only and must be modified when the flood wave causes inundation of the adjacent floodplains. When this condition occurs the nature of the flow is quite complex because of geometric and hydraulic differences between the river channel and its floodplain. In order to account for the interactive flow between river and floodplain, Fread [2] modified the basic Saint-Venant equations. Fread then approximated the modified Saint-Venant equations using a weighted four-point implicit finite-difference scheme which are solved using the Newton-Raphson iterative method in the computer program entitled DAMBRK. LEE VALLEY STUDY Between 1953 and 1957 the ESB developed the River Lee Hydroelectric Scheme. The River Lee rises in steep mountains and then flows almost due east along a narrow valley for approximately 65km until it discharges into the Atlantic Ocean at Cork City on the southern coast of Ireland. Just west of Cork City there is a large weir, the Waterworks Works Weir, below which the river is tidal. Two dams were constructed to assist with hydro-electric generating, one at Carraigadrohid and the other at Inniscarra, see Figure 2. Carraigadrohid, located approximately 27km west of Cork City, is a concrete gravity dam, 107m long and about 22m high. The dam impounds a reservoir area of 9knf and has a geometric capacity of 50 x 10& rrp. Inniscarra dam, which is 14km downstream of Carraigadrohid, is a concrete buttress dam 45m high, impounding a reservoir area of 5km^ with a geometric capacity of 70 x 10* m\ Figure 2: River Lee Hydroelectric Scheme Model Development. The study methodology may be summarised as follows: (i) Initial site visits were made to the Lee Valley to assess its general characteristics. The valley between Carraigadrohid and Inniscarra dams is well
128 Hydraulic Engineering Software confined on both sides. Carraigadrohid bridge, located a narrow point on the valley about 900m downstream of Carraigadrohid dam, consists of four arches and would form a major obstruction to large discharges of water. Downstream of the bridge the channel widens and enters the Inniscarra Reservoir which extends for about 12.5km downstream to Inniscarra Dam. There is one bridge crossing over Inniscarra reservoir, Roove's Bridge, located approximately 4.5km downstream of Carraigadrohid Bridge. One main tributary enters the reservoir, the River Dripsey, which joins the reservoir 3km upstream of Inniscarra Dam. The valley downstream of Inniscarra Dam to the Waterworks weir is 13km long. Two tributaries flow into the River Lee along this reach; the Rivers Bride and Shournagh. Generally speaking, the valley downstream of Inniscarra dam may be divided into three reaches: the upper valley forms a steep, tree lined gorge, in the middle reach crops are grown on the floodplains, while the floodplains of the lower reaches are relatively large and grass covered, characteristic of a mature river. (ii) A contoured survey of the valley upstream of Inniscarra dam which was drawn up prior to the construction of the scheme was available. An aerial survey of the valley downstream of Inniscarra dam was carried out and contour maps produced, cross-sections of this reach were also surveyed during this study. Thirteen cross-sections were used to define the channel and flood-plain between Inniscarra and Carraigadrohid dams. Seventeen cross-sections were used to define the valley reach between Inniscarra dam and the Waterworks Weir. The other main inputs to the DAMBRK model were values for Manning's n, bridge coefficients and the downstream boundary condition. The Manning's n values for the main channel and active flood plains were obtained by two methods: a method developed by Cowan [3] and a photographic technique described by Chow [4]. Both methods gave similar values, of which average values were used for initial runs. The coefficient of discharge through the bridges were estimated using the U.S. highway technique, Chow [4J. At the downstream end, the Waterworks Weir is the control on the flow. A known rating curve existed for this weir up to an elevation of 7.07m OD. This rating curve was extended for this study using two methods: firstly by using Manning's formula and secondly applying a weir formula to compute discharge at various elevation, incorporating submerged weir effects. The results from both methods were similar and mean values were used in the model. (iii) Hydrologic studies of the tributaries flowing into the lower Lee Valley were performed. The hydrology of the Lee itself and the tributaries of the upper Lee were the subject of previous detailed studies carried out by ESB. (iv) The model was calibrated using a large historic flood, as detailed below. (v) Floods resulting from hypothetical breaches in the dams were routed through the valley and inundation maps were drawn up. Model Calibration Before the DAMBRK model of the system could be used to predict breach inundation levels the model was first calibrated using data from a historic flood. On the 6th August 1986 an exceptionally large flood occured on the River Lee catchment. The twenty-two hour rainstorm, causing unparalleled flooding, provided valuable data which facilitated calibrating the model in the difficult
Hydraulic Engineering Software 129 region of 'just' overbank flow. The maximum inflow from this storm event into Carraigadrohid Reservoir was 574mVsec, and was estimated by the authors to have had a return period of 250 years using an Annual Maximum Series Model. The maximum inflow into Inniscarra Reservoir during the flood was 528mVsec. The model of the River Lee system was run to simulate the inundation which occurred on 6th August 1986. Difficulties arose from abrupt changes in bottom slope and channel cross-section. These irregularities were 'smoothed out' and the program was executed successfully. However, the results of water elevations were not considered sufficiently close to the observed values for prediction purposes. Some 'straightening' of the river was deemed necessary and so Manning's coefficients were varied within realistic bounds. It was observed that, in general, the initial estimates of Manning's n in the main river channel were close to the final values used. However, in the floodplains differences of up to 30% between initial and final values were occurred, the final values normally being higher. It was also found necessary to increase the value of Manning's n on reaches which described bends in the river. A summary of the comparison between final computed and observed flood peaks and associated times is presented in Table 1. Location km. 0.01 14.50 18.50 21.62 24.74 27.85 Table 1 - Flood of August 1986 - Recorded & Computed Stages Max. Recorded 51.48 50.45 16.40 13.00 8.53 7.22 Elevation (m OD) Computed 51.63 50.49 16.50 13.17 8.58 7.50 Difference (m) +0.15 +0.04 +0.10 +0.17 +0.05 +0.28 From Table 1 it is considered that DAMBRK can model just' overbank flow with sufficient accuracy. Figure 3 shows a comparison between recorded and computed hydrographs immediately downstream of Leemount Bridge. In general, the comparison is considered quite good particularly since this is a complex stretch of the river due to the Shournagh inflow, a bridge just upstream and an opening out of the valley into a flat flood plain. Flood of August 1986 Stage (m) 72 Time (hrs.) Figure 3: August 1986 Hydrographs at Leemount Bridge
130 Hydraulic Engineering Software Dam Breach Analyses The characteristics of a flood wave and the extent of associated downstream flooding are obviously dependent on the nature of a dam breach and its time of formation. Structural analyses of both dams has indicated that excessive cracking, sliding or overturning of either dam due to an extreme flood event, an extreme earthquake or a combination of both is highly unlikely. Nevertheless, in the interest of safety, it was decided to perform the following three dam breach analyses. Dam breach of Carraigadrohid Dam The breach of Carraigadrohid Dam was assumed to be caused by the complete failure of the largest gravity block, 17m wide and 20.5m deep, within half an hour. It was also assumed that maximum normal operating water levels prevailed when failure began and that there was a continuous inflow of SOmVs, equivalent to full load discharge, into Carraigadrohid Reservoir. The ensuing inundation downstream depends on how Carraigadrohid Bridge behaves during the flood. Thus two simulations were performed, the first assuming the bridge remained intact, the second simulating a collapse of the bridge. The analyses show that the peak levels are similar except in the vicinity of the bridge. Very little land would be inundated between the two dams because of the area of the intervening reservoir. The discharge through the spillways at Inniscarra dam due to this breach would be very close to that resulting from the 10,000 year flood and would cause considerable flooding downstream of Inniscarra dam as described below. 10,000 year flood through Inniscarra Dam Since the discharge from Inniscarra Dam due to a breach at Carraigadrohid Dam is almost identical to the 10,000 flood discharge from Inniscarra Reservoir, the inundation due to the latter was simulated. The inundation caused by this flood is considerably more extensive than that caused by the flood of August 1986 and the levels throughout the reach were significantly higher, as hsown in Figure 4. It appears from this analysis that neither of the two bridges over the River Lee in the lower valley would be overtopped and it was considered that neither bridge failed. However, the approach roads to the bridge would be rendered impassable. Peak Flood Level m OO Bn 30 -h 10000 Area Flooded (ha) 360 500 620 Distance Downstream from Inniscarra Dam (km) Figure 4: Downstream Inundation Profiles and Flooded Areas Dam breach of Inniscarra Dam Two failure mechanisms were simulated during this part of the project. The first mode of failure considered was the instantaneous open failure of all three gates - a potential mode of failure common to all dams with gates. The extent of inundation caused by this breach was found to be similar to the inundated caused by the 10,000 year flood. The second mode
Hydraulic Engineering Software 131 of failure considered was sudden failure of a 13.7m wide block of the dam. It was considered that this event has a very low probability of occurrence and hence it was decided that annual average inflows to Inniscarra Reservoir prevailed when the breach occurs. The maximum flow eminating from the dam during this event is about three times than that of the 10,000 year flood. The inundated areas caused by this breach are shown in Figure 4. A far greater area would be under water due to this event downstream of Leemont Bridge, where the floodplain opens out and the peak levels throughout the lower valley would be much greater than those predicted for the 10,000 year flood. The investigation has recently been extended to assess the consequences of a dam breach on the tidal reaches of the River Lee through Cork City, downstream of the Waterworks Weir. APPLICATION OF DIMENSIONLESS GRAPHS Sakkas [1] developed a procedure for determining the propagation of waves from ruptured dams for the U.S. Army, Corps of Engineers. A full description of the technique and its development, which predicts time of arrival of wave front, Tf, maximum flood depth, H^x, and time of maximum flood depth, T^%, is given by Sakkas. In this study the following data were used to calculate wave propagation characteristics for two values of Manning's n using the above graphs: (i) Average bed slope 0.0016 (ii) Characteristic length 52,500 ft. (iii) Downstream valley shape parameters C = 5.4; m = 0. (iv) Froude number 0.17 (v) Manning's n values: (a) n = 0.1 and (b) n = 0.15 Tables 2 presents the results from the above analyses alongside results from the DAMBRK simulation of a breach at.inniscarra Dam. Location km 3.2 6.4 12.8 Table 2 - Wave Characteristics Computed by DAMBRK and Graphs Tf 0.25 0.8 1.7 DAMBRK _T " max 9.8 9.0 8.0 T* max 1.1 2.4 5.0 Tr n O/ O.M 0.2 03 (134 0.5 1.36 21 GRAPHS H_x n O./ 0./.S 12.3 12.4 11.2 11.3 8.6 8.7 T* max n 0.7 0.7J 0.77 0.91 1.54 1.82 2.64 3.12 When n = 0.1, the dimensionless graphs predicted faster times of wave front arrival, earlier times of peak flooding and higher peak water levels at each location. Faster times to peak and higher levels appear compatible with each other and are predicted because the method assumes a prismatic channel, which cannot account for flood plain storage and retarding effects such as bridges, bends and groves of trees. The results of the analysis for n = 0.15 represent a
132 Hydraulic Engineering Software slower propagating downstream, however, the peak levels were found to be very similar to the first analysis. Both values of n are artificially high but are necessary to account for the assumed prismatic channel. From the results presented in Table 2, it is considered that dimensionless graphs may be useful to obtain an order-of-magnitude estimate of likely inundation during preliminary design or feasibility studies. However, more complete studies such as those provided by DAMBRK, are necessary during detailed design and implementation stages. CONCLUSIONS Many different techniques exist for estimating the extent of dam breach inundation, however, it must always be remembered that the predictions are not precise and that the techniques can be correctly applied only within their inherent limitations. It is considered by Fread [2] that DAMBRK can estimate flood levels to within 1m. Given this level of accuracy and the calibration work carried out by the study team, it is concluded the results obtained during this study are sufficiently accurate to allow emergency plans be drawn up for potential dam breaches at Carraigadrohid and Inniscarra. The comparison of the results between the DAMBRK and dimensionless graph analyses suggest that the latter method, in general, provides conservative results and is useful during initial appraisal stages. The authors are of the opinion that programs such as DAMBRK are useful to dam-owners and others involved in emergency flood planning since their application makes it is possible to predict with acceptable accuracy areas likely to be inundated and the timing of the flood wave as it propagates down the valley. ACKNOWLEDGEMENTS The authors would like to thank ESB for permission to publish data provided by them and, in particular, would like to thank Mr. Martin Quinn, Chief Civil Engineer, ESB REFERENCES 1. Sakkas, J.G. Dimensionless Graphs of floods from Ruptured Dams, U.S. Army Corps, of Engineers, 1974. 2. Fread, D.L. Channel Routing, Chapter 14, Hydrological Forecasting, ed M.G. Anderson and T.P. Burt, pp 437-503, John Wiley and Sons Ltd, 1985. 3. Cowan, L.W. Estimating Hydraulic Roughness Coefficients, Agricultural E/zgrnffrmg, 1956,37, 473-475. 4. Chow, V.T. O/7^C/267/iW/y)Wr^//c\s, McGrath-Hill, 1959. NOMECLATURE Q = Discharge A = Cross-sectional area of flow q = Lateral inflow / unit channel length / unit of time x = Longitudinal distance along the channel t = Time p = Water density P = Momentum coefficient V = Channel velocity V\= Velocity of lateral inflow g = Acceleration due to gravity SQ= Channel bottom slope Sf = Friction slope