REGIONAL FLOOD HAZARD MODELLING FOR DAM SAFETY ESTIMATION - A BULGARIAN CASE STUDY Jose Luis Salinas 1, Ulrike Drabek 2, Alberto Viglione 1, Günter Blöschl 1 1 Institute of Hydraulic Engineering and Water Resources Management, Vienna University of Technology, Vienna, Austria. 2 DonauConsult Ingenieurbüro GmbH, Vienna, Austria. ABSTRACT This study adresses the issue of estimation of design flood values in poorly gauged basins. An assessment of the safety calculations toward flood hazards has to be performed on existant hydraulic structures in the Tutchenitsa and Tchernialka catchments in northern Bulgaria. For this purpose a regional flood frequency analysis based on discharge data from nearby gauging stations and gridded precipitation data (on annual scale) was performed. With this approach, peak discharges for set of predefined return periods were estimated. РЕГИОНАЛНО МОДЕЛИРАНЕ НА ОПАСНОСТТА ОТ НАВОДНЕНИЯ ЗА ОЦЕНКАТА НА СИГУРНОСТТА НА ЯЗОВИРНИТЕ СТЕНИ ПРИМЕР ОТ БЪЛГАРИЯ Хосе Луис Салинас 1, Улрике Драбек 2, Алберто Вилионе 1, Гюнтер Бльошл 1 1 Институт по конструктивно водно строителство и управление на водните ресурси, Технически университет Виена, Виена, Австрия. 2 DonauConsult Ingenieurbüro GmbH, Виена, Австрия. РЕЗЮМЕ Това проучване се отнася за въпроса за определяне на проектните водни количества в слабо наблюдавани водосбори. Оценка на изчисленията за сигурността по 1
отношение на опасността от наводнения трябва да се извърши за съществуващите хидротехнически съоръжения във водосборите на р. Тученица и р. Чернялка в Северна България. За тази цел е извършен регионален анализ, който се основава на данните за оттока от близките хидрометрични станции и данни за годишните валежи. С този подход за определени върховите стойности на оттока за предварително избрани периоди на повторение. 1. Introduction Flood hazard estimation is the first step in dam design projects, and it is particularly challenging in a data scarce area. In this case, there are mainly two approaches that can be applied in order to estimate design floods for hydraulic structures: (i) rainfall-runoff modelling, and (ii) regional flood frequency analysis. In this study, a combined approach for both flood peak and flood volume was applied for the hydrological modeling of the Tutchenitsa and Tchernialka catchments. Here, only the regional flood modeling will be discussed. Several flood scenario hydrographs associated with different return periods (or annual exceedance probabilities) were estimated which in posterior stages were used as inputs for the hydraulic modeling and safety checks for existing dams in the catchments. Fig. 1. Calculation points for the Tutchenitsa case study. 2
2. Case Study, Data and Method A variety of flood waves are needed at specific locations, calculation points or nodes, constrained mainly by the boundary conditions of the hydraulic modeling. The nodes of the case study Tchernialka are the Nikolaievo II dam, and the Nikolaievo village (before and after the urban area) and the calculation points for the case study Tutchenitsa are show in Fig. 1. Given the 17 hydrometric gauging stations available shown in Fig. 2, the flood peak discharges for the two ungauged catchments (and associated sub-catchments) were calculated using the index flood method as in Hosking and Wallis (1997) [1]. From every station c.a. 20 years of peak annual discharges, catchment size and mean annual precipitation were given. 3. Results 3.1 Delineation of the homogeneous regions The delineation of the homogeneous regions was performed based on a priori knowledge of the physiographic characteristics of the catchments. This information was taken from the report Ninov et al. (2012) [2] from the Bulgarian NIMH. 3
Fig. 2. Hydrometric gauging stations available in the region (black triangles with numeric code) and location of the ungauged catchments where the peak discharges need to be estimated. The stations belonging to the Central Bulgarian Danube Plain region were identified (see Figure 3) and tested for homogeneity, giving a result of 2.46 after the heterogeneity measure proposed by Hosking and Wallis (1997) [1]. This defines the region as possibly heterogeneous and allows us to accept the cluster as in the literature values smaller than 3 are accepted for most practical applications. 4
Fig. 3. Specific mean annual flood versus catchment area of the stations from the Central Bulgarian Danube Plain. The number in brackets after the station id is the mean annual precipitation in mm/yr. Taken from Ninov et al. (2012) [2]. 3.2 Growth curve estimation The chosen statistical model for the non-dimensional flood frequency distribution, i.e. the growth curve, was the Generalized Extreme Value (GEV), based on the location on the L- moment-ratios diagram shown in Fig. 4 and goodness of fit measures. The distribution parameters estimation was performed with the L-moments method. Figure 5 shows in a flood frequency plot the empirical plotting positions of the single stations, the empirical plotting positions of the cluster as a homogeneous region and the fitted theoretical GEV distribution. 5
6th Bulgarian-Austrian Seminar Fig. 4. L-moments-ratio diagram for the Central Bulgarian Danube Plain region. Fig. 5. Growth curve for the Central Bulgarian Danube Plain region. 6
3.3 Index flood choice and estimation The index flood chosen was the mean annual flood, as it is more robust than the median for small samples (Flood Estimation Handbook, 1999 [3]). In Fig. 6, the relationship between the specific mean annual flood, i.e. the mean annual flood normalized by the catchment area, with catchment area and mean annual precipitation for all the stations available is shown. As expected, and observed in the literature, the specific mean annual flood slightly (as the scale is logarithmic at the horizontal axis) decreases with catchment size and dramatically increases with mean annual precipitation. Fig. 6. Specific mean annual floods (µ/a) in the Central Bulgarian Danube Plain region plotted against catchment area (left) and mean annual precipitation MAP (right). A log-linear regression was established between the variable to predict, in our case specific mean annual flood, and the explanatory variables or predictors, here catchment size and mean annual precipitation. The choice of this model and catchment attributes is supported by the scientific literature, as can be seen in the methods review Merz and Blöschl (2005) [4]. The result for this particular region was specmaf=2.11 10 9 Area 0.30 MAP 3.09, with specmafinm 3 /s/km 2 Areainkm 2, MAPinmm/yr 7
The model validation of the regression model was done via cross-validation or leaveone-out jackknifing simulation and the results are shown in Fig. 7, where the observed versus the simulated (in cross-validation) specific mean annual flood are plotted. As the plot shows, the points are gathered around the one to one line a perfect model would imply a perfect linear relationship and there is no appreciable bias. Two performance measures are given, the coefficient of determination (R 2 ), and the root mean square error (RMSE). The first one is the amount of variability explained by the model and the second one is an average dispersion estimate. Fig. 7. Specific mean annual flood versus catchment area (left) and mean annual precipitation (right) of all the stations available. 4. Summary and Conclusions Table 1 shows the peak discharges estimations at all the desired calculation points for different return periods. In red are marked the peak discharges associated to complete hydraulic scenarios simulations. 8
Table 1. Peak discharge estimates at desired calculation nodes for several return periods. The peak discharges associated to the presented return periods, together with additional rainfall-runoff simulations were used to obtain full flood waves. These design hydrographs acted as input and boundary conditions for both the hydrodynamical modeling performed in the area, and the safety calculations on the existent hydraulic structures in Tutchenitsa and Tchernialka catchments. R E F E R E N C E S [1] Hosking and Wallis (1997), Regional Frequency Analysis, by J. R. M. Hosking and James R. Wallis, pp. 240. ISBN 0521430453. Cambridge, UK: Cambridge University Press, April 1997. [2] Ninov, P. et al. (2012), Regionalization approaches for determination of flows with 1% probability of maximum runoff (or 100 years return period) for ungauged catchments of Tutchenica and Tchernialka rivers, Project Report from the NIMH, Sofia, Bulgaria, February 2012 (unpublished). [3] Flood Estimation Handbook (1999), Procedures for flood frequency estimation, Volume 3: Statistical procedures for flood frequency estimation. Reed, D.; Robson, A. Institute of Hydrology, Walingford (UK) 1999. [4] Merz, R., and G. Blöschl (2005), Flood frequency regionalisation spatial proximity vs. catchment attributes, Journal of Hydrology, Volume 302, Issues 1 4, 1 February 2005. 9