Dam Effects on Flood Attenuation: Assessment with a Distributed Rainfall-Runoff Prediction System

Transcription

1 Dam Effects on Flood Attenuation: Assessment with a Distributed Rainfall-Runoff Prediction System Takahiro SAYAMA, Yasuto TACHIKAWA, and Kaoru TAKARA Disaster Prevention Research Institute, Kyoto University Uji Gokasho, Kyoto, , Japan sayama@mbox.kudpc.kyoto-u.ac.jp Abstract This paper presents the development of a rainfall-runoff prediction system that enables the simulation of flood attenuation processes due to dam operation. The case study in the Yodo River basin demonstrated that the system can simulate complex dam operations such as preliminary release, peak attenuation, and cooperative operations by multiple dams. The assessment of dam effects on flood peak attenuation using the system developed indicated that in 1960 the flood discharge caused by the 30-year return period rainfall (Q 30 ) corresponds to the allowable maximum flood discharge at Hirakata, while in 2000, the Q 100 corresponds to this discharge as a result of the peak attenuation due to newly constructed dams. 1. INTRODUCTION Through the operation of dam reservoirs the flow regime of rivers are regulated to achieve a variety of objectives such as: flood attenuation, water supply, and power generation. These flow regulations form the existing flow in rivers. Hence it is difficult to find a river with natural flows in many parts of the world. Hydrological study has generally paid more attention to the understanding of the physics of natural water flow. The importance of this kind of scientific approach is obvious, but at the same time we should recognize the importance of understanding human impact on flow regimes such as highly regulated flow from dam reservoirs. In terms of flood attenuation by dams in Japan, on which this paper focuses, it is clear that the construction of dams has increased the safety level against flood disasters. On the basis of this, we are now expected to make decisions whether we continue putting more effort into increasing the safety level by structural measures, including dam construction, or we shift to more emphasis on preserving the watershed environment by accepting some floods. Understanding the current flood safety level, considering the effect of flood attenuation by existing dams, is essential to make the decisions. Understanding how these dams can attenuate different magnitude of floods is important. There are mainly two methods to assess the effects of flow regulations on the water cycle in river basins. One is to analyze observed discharge data and the other is to use a rainfall-runoff model that enables simulation of the regulated river flows. An example of the first method is to conduct flood frequency analyses with discharge data of several decades that cover the period before and after the construction of dams. Batalla et al. (2004) showed that the peak discharge corresponding to flood events of 2 and 10-year return period are attenuated by approximately 30%. This first method is useful when flow-regulating processes cannot be modelled because of their complications: however, it requires long-term data. In addition, it has the disadvantage that it cannot evaluate the effects of dams on severe floods outside the range of the observed data. On the other hand, by combining a distributed rainfall-runoff model and models of hydropower generation dams, Montaldo et al. (2004) assessed the effects of hydropower generation dams on flood attenuation. They concluded that the reservoir storage conditions before floods influence the level of peak attenuation. This type of approach has an advantage that it can simulate different hypothetical conditions. Although they modelled the dam operation rules, only hydropower generation dams were taken into account. To assess the effect of flood attenuation by multi purpose dams, which this paper deals with, it is necessary to model the operation rules in more detail.

2 This paper presents a dam operation model that can predict outflow and water level of multi purpose dams with the conditions of inflow, upstream rainfall and cooperative operations among several dams. The hydrologic prediction system, the combination of the dam models and a proposed rainfall-runoff model, is applied to one of the largest and highly regulated Japanese basins, the Yodo River basin (7,281 km 2 ). This paper is organized into sections as follows. In the next section we present the whole structure of the rainfall-runoff prediction system including the detailed explanation of the distributed rainfall-runoff model and the dam operation model used in this study. Then we apply it to the Yodo River basin and conduct test simulations. Finally, we assess dam effects on flood attenuation to investigate how the flood safety level has been increased since 1960 against different magnitude of floods. In the final section, the concluding remarks are given. 2. A DISTRIBUTED RAINFALL-RUNOFF PREDICTION SYSTEM 2.1 Whole structure of the system We construct the rainfall-runoff prediction system based on the concept of "Object-oriented Hydrologic Modelling System (OHyMoS)" proposed by Takasao et al. (1995). In this hydrologic modelling system, a user can just construct the necessary element models and specify the way to connect them by following some programming rules so that the system can combine the element models to construct the entire system. There are several significant advantages of using this system, for example: it can control the order of element models to be simulated, and it can also add or replace element models easily. Because of this flexibility, we use this system to construct the rainfall-runoff prediction system. The following four kinds of element models compose the whole system: 1. River element model. 2. Sub-catchment element model. 3. Lake element model. 4. Dam element model. Figure 1. Data processes and the whole structure of the system

3 Figure 1 shows the data processes and the element models used in this study. The processes to setup the model of the system are summarized as follows. 1. River element model 1) Prepare the digital river-network data that includes the location information of rivers and lakeshores. The solid lines in Figure 2 show the river-network. 2) Cut the river-network into segments of about 3 km length. The dark dots on the river-network in Figure 2 show the edge of each river segment. 3) Apply the kinematic wave model to the each river segment. Each model is regarded as being one river element model. 2. Sub-catchment element model 1) Use a digital elevation model to calculate the flow direction and to define the sub-catchment of each river segment. Different color zones in Figure 2 denote different sub-catchments. 2) Apply the saturated-unsaturated subsurface and surface runoff model (Tachikawa et al., 2004) to all grid-cells composing a sub-catchment. Note that each sub-catchment element model is a distributed rainfall-runoff model. 3. Lake element model Construct a lake element model, which is a simple mass-balance model, to simulate water level from inflow, outflow, and rainfall information. In the case study in this paper, we apply this model to Lake Biwa and its outflow is simulated with the dam element model applied to the Setagawa dam, which controls the outflow from the lake. 4. Dam element model Construct a dam operation model and apply it to dams in the basin. We refer to this site-specific dam model as a dam element model. 5. Combining element models Combine all element models on OHyMoS to obtain the whole prediction system. In the case of the Yodo River, there are 1707 river element models, 1707 sub-catchment element models, one lake element model and eight dam element models. Figure 2. Examples of river segments and sub-catchments 2.2 Sub-catchment element model Figure 3 (a) shows a schematic diagram of the soil layer of the model, which takes into account three types of flow: unsaturated flow in capillary pore, saturated flow in non-capillary pore, and surface flow on the soil surface. In this figure, the soil depth is D [m], the water stage corresponding to the

4 saturated water content is d s [m], and the water stage corresponding to maximum water content in the capillary pore is d c [m]. Figure 3 (b) shows the stage-discharge relationship of the model. Let k c and k a be saturated hydraulic conductivities in capillary pore and in non-capillary pore, respectively, and v c = k c i, v a = k a i, then the relationship between the discharge par unit width q [m 2 /s] and the stage h [m] are described as follows: Figure 3. Schematic diagram of the soil layer (a) and stage-discharge relationship (b) of saturated-unsaturated subsurface and surface rainfall-runoff model β h v ( ) cdc, 0 h dc dc q = v cdc + v a ( h dc ), ( dc < h ds ) (1) m v cdc + v a ( h dc ) + α( h ds ), ( ds < h) where β [-] is the parameter to describe the reduction of hydraulic conductivity in capillary pore as the water content reduces. β equals k a / k c so as to keep the continuity of the stage-discharge relationship between the capillary pore and the non-capillary pore layers. Combining this stage-discharge relationship (1) and the continuity equation, we simulate rainfall-runoff from each grid-cell. The simulated discharge will be inflow of downstream grid-cell. The water flow is routed until it reaches a river segment. 2.3 Dam element model By formulating the dam operation rules and decision-making processes used by dam operators, we develop the dam operation model. It predicts the outflow and water level of a dam with the input information of inflow, average rainfall in the dam catchment, and operation status of other related dams. All the dams located in the Yodo River basin, the case study basin in this paper, are multi-purpose dams. Although each dam has different operating rules, we can categorize the following six common operation processes for flood attenuation (Ichikawa et al., 1999). 1. Ordinary operation 2. Operation under flood warning 3. Preliminary release operation

5 4. Peak attenuation operation 5. Flood release operation 6. Post flood operation Each dam is always under one of the six operations, and we formulate the conditions to shift from one operation to another with if-then equations. Figure 4 (left hand side) shows how to shift the process from one to another, and Figure 4 (right hand side) shows specific water levels that appear in the dam operation rules. In the following six sub-sections, each operation process is explained in terms of the general objectives, its modelling methodology and conditions required to start the operation process. Figure 4. Dam operations and specified water levels Ordinary operation When a dam is under ordinary operation, or when a flood warning has not been issued, the dam is operated to keep the target water level. The target water level differs between flood season and nonflood season. The model decides the outflow to maintain the target water level Operation under flood warning Generally speaking, when a nearby meteorological observatory issues a warning related to heavy rainfall, dam administrators have to issue a flood warning as well. They have to control the water level to maintain the same level as when the flood warning was issued. The dam operation model judges if the dam is under the flood warning condition or the ordinary condition depending on the average rainfall in the dam catchment. It uses a parameter to denote the threshold of the rainfall amount to determine the issue of the flood warning. The model maintains the water level unless the flood warning is canceled or meteorological and hydrological conditions are severe enough to shift to the next operation process that is preliminary release Preliminary release operation Dam operation rules set up the preliminary release to draw water level to increase the capacity before a flood peak arrives. The rules require a preparation period before they actually start the release. The dam operation model lowers the water level until it reaches the Minimum Permissible Water Level. The maximum discharge of the preliminary release is decided by operation rules. In addition, the dam operation model uses a parameter to specify the maximum rate of change in discharge which can also be obtained from the operation rules. The conditions to start a preliminary release are somewhat arbitrary. Some descriptions of releases can be found from the operation rules. For example, if a typhoon comes to a certain location, dam administrators must start preliminary release. There are several conditions and different dams have

6 different conditions. It is, therefore, not easy to formulate all of these conditions. Since it is observed that all these conditions are related to current and forecast rainfall, the following simple condition to express the dam administrator s decision making processes was setup. Start preliminary release if the average rainfall in the dam catchment reaches Rpp mm in the last Tpp hours and if the forecast rainfall in Tpf hours is more than Rpf mm. where Rpp, Tpp, Tpf and Rpf are the model parameters, which are estimated based on observed rainfall-outflow data Peak attenuation operation Dam administrators start peak attenuation operation when inflow exceeds Flood Discharge. There are several peak attenuation methods, such as Constant discharge release and Fixed discharge-rate release. These two operation methods are taken into account in this study, and all operation methods of dams in the Yodo River basin can be classified into either one of the two. The dam model selects the outflow based on peak attenuation control rules, and peak attenuation control starts when the inflow exceeds Flood Discharge, which is a parameter of the model Flood release operation Dam administrators are to open the dam gate if the water level reaches the Flood Release Water Level. The dam model releases the same discharge as the inflow if the water level reaches this level Post flood operation When inflow becomes less than Flood Discharge after peak attenuation control, dam administrators can start post flood operation. This control aims at reducing the water level to Normal Water Level. The model reduces the water level with the amount of Flood Discharge, which is defined as a parameter of the model. 3. TEST SIMULATION 3.1 Study area The Yodo River basin is analyzed in this case study because it is a typical Japanese river basin highly regulated by multi-purpose dams. The total area of the Yodo River basin is 8,240 km 2. In this case study, since Hirakata is the main design target location for designing dams and other river works, we focus on the upper Hirakata basin (7,281 km 2 ), and refer to this basin as the Yodo River basin in this paper. There are eight multi-purpose dams inside the basin and five of these dams are located in one of the sub-basins, called the Kizu River basin (1,596 km 2 ). Setagawa dam controls outflow from Lake Biwa (670 km 2 ), which is the largest Lake in Japan. Administrators operate all dams in the basin so that integrated flow regime control can be undertaken. The mean annual precipitation of the Yodo River basin is about 1,600 mm.

7 Figure 5. the Yodo River basin (upper Hirakata: 7,281 km2) Dams Table 1. Properties of the dams in the Yodo River basin Operation started year Catchment area [km 2 ] Flood control capacity [10 6 m 2 ] Peak attenuation Setagawa Constant Amagase Constant Takayama Fixed disc.-rate Shorenji Constant Murou Constant Nunome Fixed disc.-rate Hiyoshi Constant Hinachi Constant 3.2 Method We conducted a test simulation using observed rainfall and discharge data during a typhoon event in The period of the simulation is from July 25 to July 29. We used the Thiessen polygon method to distribute rainfall data that is observed by 58 raingauge stations inside the basin. Estimated total rainfall during this event is 149 mm. We divide the whole basin into three zones depending on the land use, and assign different parameters in the different land use zones. There are three land use categories: forest, paddy field, and urban area, among which forest is the dominant landuse (63%). A rainfall-runoff model considering surface flow is used for the paddy field zone and the urban area zone by substituting zero in d c and d s in Equation (1), therefore, only the Manning coefficient n is a parameter of the surface rainfall-runoff model (paddy field zone : 1.0 [m-1/3s] and urban area zone : 0.3 [m-1/3s]). For the forest area zone, we used saturated-unsaturated subsurface and surface rainfall runoff model, and the parameters are as follows: n = 0.6 [m-1/3s], D = 1.0 [m], d s = 0.2 [m], d c = 0.1 [m], k s = 0.01 [m/s]. Six dams and Biwa Lake are also simulated in the rainfall-runoff simulation system. Two dams constructed after 1997 are not included in this simulation.

8 3.3 Results and discussions Figure 6 (a) shows the simulated and observed inflow and outflow at Shorenji dam. The good agreement between the simulated inflow with the observed inflow verifies the rainfall-runoff model at the upstream of the dam. In terms of the simulated outflow, the dam element model could simulate the preliminary release and peak attenuation operations well. Figure 6 (b) shows the simulated and observed water level at Shorenji dam. Though the simulated water level does not perfectly match the observed water level, it shows the water level is drawn down by the preliminary release and it is increased by the peak attenuation operation. Figure 7 (a) shows the simulated outflow from Setagawa dam, which is located at the outlet of Lake Biwa. Since there is not enough quantitative information or operation rules to model this dam, we were required to tune some parameters based on the observed data. It is, therefore, not surprising that the simulated outflow can fit with the observed outflow very well. However, there is one notable feature in this simulation result. We observe that the outflow is reduced from 300 cms to 200 cms temporarily around 35 hours to 50 hours after the simulation began. This reduction reflects the special operation rule: Setagawa dam has to keep its outflow less than 200 cms when Amagase dam is under preliminary release or peak attenuation operations including their preparations. Figure 8 shows the simulated and observed inflow and outflow at Amagase dam, and we realize that it was under preliminary release operation from 35 hours to 50 hours. This kind of complex operations by multiple dams could be simulated well because the dam operation models and the rainfall runoff models interact with each other. (a) Figure 6. Simulated and observed inflow and outflow (a) and water level (b) at Shorenji Dam (b) (a) Figure 7. Simulated and observed outflow at Setagawa Dam (a) and water level of Lake Biwa (b)

9 Figure 8. Simulated and observed inflow and outflow at Amagase dam Figure 9. Simulated and observed discharge at Kamo Figure 9 shows the simulated and observed hydrographs at Kamo. The simulation result excluding the dam element models is also displayed in the figure. The simulation result with the dam element models reproduced the observed hydrograph quite well. By comparing simulated hydrographs with and without dams, it was concluded that the dams located upstream of Kamo could attenuate the peak flood by around 500 cms. 4. ANALYSIS OF THE DAM EFFECTS ON FLOOD ATTENUATION 4.1 Objectives Construction of large-scale dams is believed to have improved the safety level of a catchment against flood disasters. However it is not clear to what extent these dams have improved the safety level or which range of flood magnitudes dams can regulate efficiently. In terms of the Yodo River basin, there are currently eight large-scale dams in the basin and seven dams were constructed after Conducting rainfall-runoff simulation using the developed rainfall-runoff prediction system, we evaluate how the flood safety level of the Yodo River basin has been progressively increased over the period 1960 to 2000 with respect to rainfall events of different magnitude. 4.2 Method Rainfall-runoff simulations were implemented considering the dams that were operated at the beginning of the following years: 1960, 1970, 1980, 1990, and The simulated peak discharges at Hirakata are examined to discuss the effect of the dams. The rainfall event used in this study is the typhoon event observed in 1982 (August 1 to 3). This event was chosen because it was the largest event since 1980 when enough rainfall-discharge sequences were available to conduct the simulation. Furthermore, in order to examine which flood magnitude the dams can attenuate the peak, some factors were applied to the 1982 rainfall pattern. The factors are selected so that the modified rainfall events correspond to the following return periods: 30, 50, 100, 150, 200, and 300 years. The same parameters and the initial conditions used in the test simulation are used for this assessment.

10 4.3 Results and discussions Figure 10 shows the simulated peak discharge at Hirakata. The horizontal axis of the figure represents the return period of the input rainfall, and the vertical axis of the figure represents the simulated peak discharge. Different lines in the figure represent different years: for example, the line of 1980 represents the result simulated with the dams of Setagawa, Amagase, Takayama, Shorenji, and Murou because all of these dams commenced operations before Firstly, looking at the results of 1960 when only Setagawa dam existed, it can be observed that peak discharge caused by the 30-year return period rainfall (Q 30 ) exceeds 12,000 cms, which is the allowable maximum flood discharge at Hirakata. By 1970 two other dams, Amagase dam and Takayama dam, had been constructed. The comparison between the lines for 1960 and 1970 indicates these two newly constructed dams succeeded in attenuating flood peaks from relatively smaller rainfall events up to around the 50-year return period. In other words, if rainfall exceeded the 100-year return period or so, these two dams could not reduce the peak discharge.. By 1980, two more dams, Shorenji dam and Murou dam, had been constructed. These dams enabled the peak discharge to be reduce by about 2,000 cms compared with the discharge from the 100-year and 150-year return period rainfall. On the other hand, these two dams were not effective in attenuating the peak discharge at Hirakata if the magnitude of rainfall is less than the 30-year return period. This is shown by the fact that the lines for 1970 and 1980 are quite to close each other for these return periods. This is because Takayama dam located at the downstream of Shorenji dam and Murou dam had a flood control capacity large enough to regulate floods of this magnitude. By 2000, a further three dams, Nunome dam, Hinachi dam, and Hiyoshi dam, had been constructed. It is noteworthy that, in 2000, the allowable maximum flood discharge was exceeded by the discharge from the 100 year return period rainfall (Q 100 ), whereas in 1960 this value was exceeded by the Q 30 peak discharge. Figure 10. Simulated peak discharges at Hirakata with different magnitude of input rainfall. Years (1960, 1970, etc.) denote that the dams existed in the year are included in the simulation.

11 5. CONCLUSIONS A rainfall-runoff prediction system was developed for the Yodo River basin. The combination of the dam operation models and the rainfall-runoff models enabled the simulation of the highly regulated rainfall-runoff processes in the basin. The conclusions of the assessment of dam effects on flood attenuation in the basin are summarized as follows: 1. The dams constructed in the 1960s were effective in attenuating relatively small flood peaks caused by the rainfall event with the magnitude of the 50-year return period or smaller. On the other hand, the dams constructed after 1970 were effective in attenuating relatively larger flood peaks caused by the rainfall events with the magnitude of the 100-year return period or larger. 2. Q 30 corresponds to the allowable maximum flood discharge at Hirakata in 1960, while Q 100 corresponds to it in Thus, the safety level at Hirakata has been increased by the construction of the dams. However, we found that it has not achieved the initial design target: the discharge caused by smaller than the 200-year return period rainfall cannot exceed the allowable maximum flood discharge. It is still necessary to verify these results with different rainfall patterns under different conditions. In addition, assessing the flood safety at several locations in a basin is also necessary to evaluate flood risk in a catchment. 6. ACKNOWLEDGEMENTS A part of this study is supported by "System Modelling Approaches for Assessment of Interaction Between Social Changes and Water Cycle (PI: Prof. Kaoru Takara, Disaster Prevention Research Institute, Kyoto University)", which is conducted under the CREST (Core Research for Evolutional Science and Technology) Program by the Japan Science and Technology Agency. 7. REFERENCES Batalla, R.J., Gomez, C.M. and Kondolf, G.M. (2004), Reservoir induced hydrological changes in the Ebro River basin (NE Spain), Journal of Hydrology, 290, Ichikawa Y., Hirano K., Shiiba, M., Tachikawa, Y. and Takara, K. (1999), Development of a dam model based on structural modelling system (in Japanese), Proceedings of the 54 th Annual Conference of JSCE, Hiroshima, September 22-24, 1999, Montaldo, N., Mancini M. and Rosso, R. (2004), Flood hydrograph attenuation induced by a reservoir system: analysis with a distributed rainfall runoff model, Hydrol. Process., 18, Tachikawa, Y., Nagatani, G. Takara, K. (2004), Development of stage-discharge relationship equation incorporating saturated-unsaturated flow mechanism (in Japanese), Annual Journal of Hydraulic Engineering, JSCE, 48, Takasao, T., Shiiba, M. and Ichikawa, Y. (1999), A runoff simulation with structural hydrological modelling system (in Japanese), Annual Journal of Hydraulic Engineering, JSCE, 39,