The catastrophe of the Niedów dam – the dam break causes, development and consequences

Due to extreme rainfalls in 2010 in the Lusatian Neisse river catchment, a flood event with a return period of over 100 years took place leading to the failure of the Niedów dam. The earth type dam was washed away with a result of a rapid release of nearly 8.5 million m of water volume and flooding of the downstream area with substantial material losses. The paper analyses the conditions and causes of the dam’s failure, while special attention is given to the mechanism and dynamics of the compound breaching process. Several empirical formulas for the dam breach prediction are tested with regards to their 5 usefulness in assessing breach dimensions and outflow peak discharge. The paper also describes a numerical approach to simulate the flood propagation downstream the dam with the use of a two-dimensional hydrodynamic model. Considering the specific local conditions and available data set, an iterative, reversed solution of the unsteady state problem is made. This approach enabled to deliver realistic flood propagation estimates, verification of the dam breach outflow, and several important answers on the dam failure consequences. 10

. The catchment of the Witka river on the major Lusatian Neisse river (IMGW, 2011). In most hydrological gauge stations, observed water levels significantly exceeded historical maxima. Remarkably, a number of gauge meters was destroyed during the floodwater passage, making it more difficult to subsequently assess the quantitative data of the flood. The return period of the flood is estimated to be within 100-200 years. 125 The flow on the Witka river -its name on the Czech territory is the Smeda river -is monitored at four gauge stations. On the Polish section from km 0.0 (river mouth) to km 8.0, there are two stations: Ostróżno (km 7.98), upstream the reservoir, and Ręczyn (km 1.8), downstream the reservoir (Fig. 1). On the 7 August at the gauge Ostróżno the highest water level occurred at 16:40. The gauge station Ręczyn recorded the water levels until 15:20, as it was destroyed due to high release of water from the reservoir, yet before the dam break. During the 45-year period of continuous flow monitoring at Ostróżno gauging 130 station, the flood discharges were less than 70 m 3 s −1 , which still is within the limit of bankful flow. There was just a single higher flow in August 2001, equal to 171 m 3 s −1 . That event also featured a rapid ascent and descent of the wave, typical for the flash flood. On the critical 7 August the estimated flood rate was 615 m 3 s −1 , but this estimation is still burdened with 5 https://doi.org/10.5194/nhess-2020-372 Preprint. Discussion started: 23 November 2020 c Author(s) 2020. CC BY 4.0 License. significant uncertainty and was a subject to discussion. A direct reliable estimation of the peak flow rate was not possible, as the water level substantially exceeded the measuring range. In addition, the topography makes it more difficult due to locally 135 wide floodplain. Between the Ostróżno cross-section and the reservoir, there is an increase of the catchment area from 268 to 331 km 2 , including the Koci Potok stream, which also severely flooded and delivered a significant direct inflow to the reservoir.
This stream was not monitored, but based on the field survey after the flood, the peak flow rate was estimated to ca. 70 m 3 s −1 .

Field observations
A field survey was carried out past the flood to collect data on the flood wave passage along the Lusatian Neisse river and 140 its major tributaries (IMGW et al., 2010). A number of eye witnesses of the flood were interviewed, including several local authority representatives. Water marks were searched and fixed for further geodesy recording. Maximum water elevation marks were recorded in over 50 locations, including upstream and downstream the Niedów dam. It is to be pointed out that post-factum determination of the exact maximum water elevation is not a simple task. In some cases, clear water lines were found on walls in a form of sediment and residues marks or washed out dirt, but in several locations high water marks were approximate, 145 based on residues found on threes, bridge piers or decks. Therefore, the error of maximum water elevation may vary from a few millimeters to ca. 0.3 m. Apart of that, the limits of flooded area were also explored and brought on the map, consecutively digitalized. This later served also as a reference for the verification of the flood hazard maps elaborated in the context of the EU Flood Directive.
An important information is the timing of the flood on the Lusatian Neisse river at the section of the Witka river inflow as 150 no gauge is located nearby. Based on inhabitants reporting it has been found that culmination of the flood occurred between 2 and 3 a.m. on the 8 August 2010, hence significantly later than the peak outflow from the Niedów dam. This information helps explaining the effect of the coincidence of the flood waves from the two rivers and to find proper upper boundary conditions for the hydrodynamic model. It is also worth mentioning here about a good communication and cooperation between Polish and German (Land of Saxony) 155 water services and authorities to put under discussion and gather the needed information, and reaching an agreement on selected important data. In this respect, a bilateral experts group was set up to deliver a common set of data to be used in hydraulic modelling. As a result, a common 2D modelling platform is being developed for consistent flood hazard mapping for this transboundary territory (ICPO, 2019).   The structure and geometry of the earth dam are depicted on Fig. 5. The maximum height of the embankment in respect to the base ground level was 11.6 m. The body of the dam was well compacted sand. The slope upstream was of a ratio 1:3, while the slope downstream of 1:2.5. The total volume of earth dam was ca. 61,000 m 3 . Because the sand had a high filtration rate of 2.8×10-3 ms −1 , the upstream slope was shielded with a double layer of concrete slabs with dimensions of 1.5×1.5×0.1 m, sealed with a bituminous material. This shield from upstream was supported by a vertical reinforced concrete filtration screen, 170 reaching down the basement rock. The downstream slope was covered with grass on humus. In the lower part of the slope a

Dam Wash out Mechanism and Breach Characteristics
The process of destruction of the Niedów dam followed a specific pattern as the overflow and washing out was different for the 175 left and right dam side, both, in terms of timing, and dynamics. An S-shape plan of the reservoir just upstream the dam (Fig. 1) caused a low angle of inflow direction to the dam axis. As a result, the water level at the left side was higher several centimetres than on the other side, leading to uneven overflow. The overflow started at 17:00 over the left dam near the bank because of a slight crest inclination towards it. The water passing over the crest caused the erosion first around the lighting foundations (additional turbulences) and progressive washing out of the downstream, grass grown slope. This action took approximately 180 half an hour resulting in a damage of the dam crest, by disintegrating it part by part, since the road concrete slabs lost support of the eroded sand. Further, the dam breach moved regressively upstream causing upstream slope to disintegrate from the top.
Remarkably, the concrete slabs when losing support broke in series like a chocolate, and were swept away by intensified flow.
This phase was slowed down for a while as two concrete slabs resisted longer, hydraulically acting as a sharp crested weir.
Next, another important moment took place. As the support of the earth embankment vanished, the left training wall flanking 185 8 https://doi.org/10.5194/nhess-2020-372 Preprint. Discussion started: 23 November 2020 c Author(s) 2020. CC BY 4.0 License. the central concrete dam collapsed due to upstream water pressure. This resulted in a further rapid outbreak. This phase was relatively short but intense, with a result of torrential flood wave downstream documented by Fig. 6. After next 80 min. the earth dam was almost completely swept away (Fig. 7).
The overtopping of the right dam begun approximately 15 min. after the left one. The breaching in this case developed in a similar but in less dynamic fashion as above. The wash-out started at the central part of the right dam, evolving towards the 190 right bank, conforming with the water inflow direction. As a results of the fall of left abutment and a rapid lowering of the water level in the reservoir, the right bank washing out decelerated. In addition, the concrete slabs resisted to fail and worked as a weir for some 20 min. It is difficult to explain the origin of it. Possibly, the slabs jammed or the concrete debris temporally hindered the erosion progress. Fig. 6 again portrays this situation of hindered breaching of the right dam side. Finally, the concrete slabs got down. Nevertheless, the outflow here was not that intense any more, since the upstream water level had A part of the dam adjacent to the control structure remained. Table 1 has the crucial moments of the breaching development, established via blending of observations, records, and interviews. 3 Methods

Empirical formulas for Breach Calculation
The breaching development prediction is essential for making the estimates of flood propagation and flood hazard. Therefore, 205 with own data a number of empirical and regression formulas are put to the test to calculate breach characteristics, including the   a Difference between natural ground lowest point behind dam and dam crest level. b Average depth along the left and right dam axis respectively. c Breach formation times provided by (Froehlich, 2008). Considered to be "length of time needed for the final trapezoidal breach to form, which takes place after the breach initiation phase". d Times from overflow to empty reservoir.    The peak of total outflow through the spillway and two sections of breaching dam (individual peaks are shifted in time) is taken for comparison with the formulas results due to the statistical methodology used for their development. As presented in Table 9, also in this case there are significant discrepancies with the real life estimates. Most of the calculated peak outflows 285 are overestimated, with an error in a range of 3% (Froehlich, 2008) to 122% (Ashraf et al., 2018).  The origin of the discrepancies is attributable to specific features of the studied dam wash-out and may be listed as follows:

2D Hydraulic Model for Flood Routing
1. The inflow direction relative to the dam axis plays a relevant role. study highlights the need for more work in this subject, especially relating the breaching dynamics with the physical properties of the dam and hydraulic parameters in place of rather robust regression analysis. A unified methodology and general approach 310 is highly desirable.

2D Flood Routing
The flooding along the Lusatian Neisse in the studied reach is a combination of two major flood vawes (including the catastrophe of the Niedów dam) at the upper boundaries (see Fig. 10) with hydrographs to be reconstructed. The solution of the problem is iterative, relying on a series of computations executed with adjusted shapes of hydrographs, 315 timing and volume, conforming with Eq. 1. In this iterative approach, a simultaneous sensitivity analysis was performed on identifying and analyzing the influences of change of the roughness Manning coefficients on peak flows and on wave front propagation for a given breaching outflow hydrograph. The consequent calibration process also included local bathymetry modifications of the main channel, which appeared necessary due to two reasons: i) rectangular grid is not always properly representing relatively narrow and curved channels; ii) the cross-sections taken during low or medium flows may not be 320 representative for high flow conditions and local scouring needs to be taken into account (local roughness coefficients would be beyond an acceptable range). Yet, one also needs to bear in mind that high water marks are not absolute values in terms of accuracy, some of them are just indicative. The author's being a field surveyor and modeler in one person is an advantage here. computation is considered as a consistent reconstruction of the flood in terms of water levels, flooding extend, discharge rates, timing and water volumes. The simulated water levels correspond well with the high water marks, given the complexity of the modelled domain and the inherent uncertainties; the differences are from just a few to about thirty centimeters. Remarkably, the 2D model delivered a close to reality inflow to the Berzdorfer lake due to embankment overtopping. The resulting dam breach hydrograph Q N (t) was determined with a peak discharge of 1380 m 3 s −1 appearing at 18:20, see Fig. 11. The total volume of 330 water released due to the dam failure is 22 million m 3 , which is about 5 mln m 3 more than the inflow to the reservoir. the Zgorzelec city. The Lusatian Neisse hydrograph shows two peaks -the first one is caused by the Niedów dam break, and