Assessment of optimal empty flushing strategies in a multi-reservoir system

operation, thus a feasible strategy of empty flushing should prevent significant increase of water shortage risks. This paper presents a framework of performing empty flushing in a multi-reservoir system, where flushing is carried out in a primary reservoir and the other reservoirs provide backup storage for stable water supply during flushing. A network flow programming-based model is employed to simulate daily joint operation of reservoirs. 15


Introduction
Empty flushing is the most effective method for removing deposited sediments from reservoirs (Fan and Morris, 1992; Morris and Fan, 1998;Shen, 1999).This process requires complete drawdown of reservoir storage to allow "inflows to pass through at riverine depths" (Atkinson, 1996).The drawdown of storage is usually carried out by releasing water 5 through bottom outlets, such as sluiceways.During this process, the accelerated flow near the inlet may partially reactivate and scour out the depositions to generate a flushing cone in the vicinity of the inlet.By completely emptying the reservoir and maintaining the riverine flow condition, retrogressive erosion may be induced from the rim of the flushing cone extending to the upstream to create a flushing channel.The formation of the flushing channel 10 usually leads to hyper sediment concentration of the bottom release and thus effectively recovers partial deposited capacity of the reservoir.This operation has been used to pursue sustainable utilization by many reservoirs worldwide (Atkinson, 1996;White, 2001; Chaudhry and Habib-ur-rehman, 2012), some examples of which are presented in Table 1.
The other side to desilting, however, is that draining the storage of a reservoir 15 counteracts its water supply function.Hence empty flushing is generally limited to reservoirs that operate solely for hydropower generation, flood mitigation, or irrigation.These purposes usually do not require reservoir storage during certain periods of the year, during which empty flushing can be implemented without impairing the original design function of the reservoir.However, for reservoirs with municipal or industrial water users that rely on 20 sufficient storage for steady water supply, the implementation of empty flushing is relatively rare.
The conflict between water supply and empty flushing has been addressed by Chang et al. (2003) and Khan and Tingsanchali (2009).Chang et al. (2003) developed the operating HP: hydropower generation, FC: flood control, IR: irrigation, ID: industrial water supply, S: single reservoir system, M: multi-reservoir system, Capacity-inflow ratio: the ratio between the effective capacity and the annual inflow volume of the reservoir, HP: hydropower generation, FC: flood control, IR: irrigation, ID: industrial water supply, S: single reservoir system, M: multi-reservoir system, Capacity-inflow ratio: the ratio between the effective capacity and the annual inflow volume of the reservoir, HP: hydropower generation, FC: flood control, IR: irrigation, ID: industrial water supply, S: single reservoir system, M: multi-reservoir system, Capacity-inflow ratio: the ratio between the effective capacity and the annual inflow volume of the reservoir,

Qualitative analysis: key factors for successful operations of empty flushing
Two performance indices, expected desilting volume and the induced increments of water shortage, are used in this study to evaluate an empty flushing strategy.An optimal strategy should maximize the desilting volume while maintaining the incremental shortage 5 under an acceptable threshold.According to the cases in Table 1, key hydrological and operational factors for succeeding in these indices are described as follows: 1. Qualitative conditions for water supply (QCWS) (1) QCWS  However, if the inflow exceeds the capacities of the outlet works, then the WSL in the reservoir will begin to rise.This leads to decreased flow velocity in the reservoir, which reduces the empty flushing efficiency.Atkinson (1996) suggested 20 the use of the drawdown ratio (DDR) to measure the flushing efficiency.This index is defined as 1 minus the ratio between the depth of WSL during empty flushing and the depth of normal pool level of the reservoir.Atkinson (1996) and White (2001) defined incomplete drawdown flushing as situations in which DDR is less than 0.66, wherein the depth of the water during flushing is greater than a third of 25 the maximum depth.In such circumstances, the efficiency of empty flushing is significantly reduced and it is recommended to switch the operation to the regular mode of water supply.

2 Quantitative derivation of the optimal empty flushing strategy
As stated in the introduction section, this study focuses on implementing empty 5 flushing of a single primary reservoir within a multi-reservoir system.While the comprehensive discussion of the previous subsection is generally applicable to multi-reservoir systems, the proposed quantitative methodology as well as the following case study apply specifically to those systems without means to artificially generate flushing inflow to the primary reservoir.In addition, we focus on event-based operation.This means that the timing  The following proposed method for deriving optimal strategy adopts the simulation-

Estimate the flushed sediment discharge
Fig. 1 The procedure to derive the optimal empty flushing strategy

Joint operating rules for a multi-reservoir system
According to Oliveria and Loucks (1997), the rules to jointly operate multiple 5 reservoirs for water supply include the following two phases: 1. Determination of total water supply amount: The total amount of water supply is determined based on the total storage of reservoirs in the system.If the total storage does not suffice, a discount of total water supply may be applied by the system-wide release rule.Fig. 2 presents the joint operating rule curves, a form of the system-wide release rule, reservoirs is below the critical limit, only 80% of the public demand and 50% of the agricultural and industrial demands will be satisfied.When the total storage is between the lower and critical limits, the public demand should be fulfilled and 75% of the agricultural and industrial demands need to be satisfied.When the total storage is between the upper and lower limits, all demands should be fulfilled.In the event that the storage in the  measures the total storage in the system, and the two curves represent the suggested target storages for the respective reservoirs with regard to various total storage amounts.These curves vary during each ten-day period within a year to facilitate efficient storage allocation according to the pattern of water demands and reservoir inflow.The first part of the proposed method requires appropriate adjustment of the storage balancing curves before and during the periods feasible for empty flushing.This adjustment 10 would prioritize the water released from the primary reservoir while preserving storage in the other.This complies with the aforementioned QCWS1 and QCFS2, and creates a favorable

Conditions for initiation of an empty flushing operation
Water supply simulation of historical daily reservoir inflow records is sequentially performed according to the joint operating rules.During the simulation, empty flushing operation is activated when all of the following conditions are satisfied: 2. The storage of the primary reservoir is lower than a threshold U T .This ensures the satisfaction of QCFS2.Theoretically, a higher value of U T allows for the initiation of drawdown flushing at higher primary reservoir storage levels, thus increasing the range of 10 opportunities for empty flushing.Nonetheless, a higher U T incurs the risk that, if subsequent reservoir inflow falls short of predicted values, the emptied storage may not be replenished.
3. The total storage in the backup reservoirs is greater than a threshold D T .This ensures meeting QCWS1.A higher value of D T elevates the stability of water supply during 15 empty flushing.In cases where either this or the above condition has not been met, demand should be supplied from the storage of primary reservoir as much as possible, or storage should be diverted from the primary reservoir to the others.However, this storage reallocation may be limited by the water transmitting capacities between reservoirs.Such that the conditions for initiating empty flushing may not be met within the pre-specified 20 feasible period for flushing.Therefore, a higher

Estimation of the flushed sediment discharge
Once the activation conditions are met, the gates of the bottom outlets of the primary reservoir are fully opened to empty the storage and route the inflowing water and sediments.
The release from the primary reservoir may cause blockages of the downstream water diversion or water treatment facilities due to its high sediment concentration.Thus the water where QC t and Q t denote the sediment discharge (T/s) and water discharge (m 3 /s) flushed from the primary reservoir during the t-th simulating day, respectively; S f represents the energy slope associated with the flow in the primary reservoir during empty flushing; W is the width of the flushing channel (m), which can be estimated using the empirical formula

Conditions for termination of empty flushing operation
Empty flushing operations should be terminated if either of the following circumstances occurs:

Evaluation of optimal empty flushing strategies
The storage thresholds for activating and terminating an empty flushing operation as described in subsections 2.2.2 and 2.2.4 are regarded as parameters.These parameters are where n t is the total number of days within the simulating horizon; QC t is the simulated sediment discharge from the primary reservoir by empty flushing on the t-th day.It is determined by substituting the release of the primary reservoir during the flushing period into Eq.( 1), and  is the maximum acceptable monthly water shortage ratio induced by empty flushing.The BOBYQA, a nonlinear optimization algorithm of Powell (2009), is used to 5 solve the problem.The details of BOBYQA can be found in Powell (2009) and the barrier function approach to handle the constraint of Eq. ( 5) can be found in Chou and Wu (2015).

Case study and experimental setup
The joint operating system of the Tsengwen and Wushanto Reservoirs in southern Taiwan is selected for case study.

Determination of the feasible period for empty flushing
Fig. 5 illustrates both water demand of this system and average inflow to the Tsengwen 10 Reservoir in ten-day increments over a year.As can be seen, inflow to the reservoir generally begins increasing between late May and early June, as precipitation rises during the beginning of the wet season.This is also the period in which the irrigational water demand, which constitutes the majority of total demands, is lower.The first semiannual rice crop is harvested, and the second semiannual irrigation just begins.As shown in Fig. 2, between May 11 and June 30, the lower limit of the operating rule curves is below the effective capacity of the Wushanto Reservoir.Even if the Tsengwen Reservoir is empty, as long as the Wushanto Reservoir is full, the total storage of the system would still exceed the lower limit of the operating rule curves, such that the demand for water could be satisfied.All of these   2 and 3.The results show that in May, there is a 52 % probability that the storage of the Tsengwen Reservoir will drop below 20 million m 3 and an 8 % probability that the Tsengwen Reservoir storage drops below 20 million m 3 while the 5 Wushanto Reservoir storage simultaneously exceeds the lower limit.In June, the two probabilities are 31 % and 14 %, respectively.These two months present the highest probabilities during a year.The respective storage levels of the Tsengwen and Wushanto Reservoirs each satisfy the abovementioned conditions only between May 11 and June 20, which is thus selected as the feasible period for empty flushing in the Tsengwen Reservoir.WSLs are not available currently; therefore, we referred to Atkinson (1996), who suggested using  = 60 when the capacity of bottom outlets is limited.Atkinson (1996) also suggested that when the water depth of a reservoir exceeds 30 % its maximum depth, the flushing efficiency will decrease significantly.The 30 % depth of the Tsengwen Reservoir is 15 approximately at the elevation of 185 El. m, which corresponds to an impoundment of 21.81 million m 3 .To prevent overestimating the effectiveness of empty flushing, it is assumed that if a flood during an empty flushing operation raises the WSL of the Tsengwen Reservoir to exceed 185 El. m, then the flushed sediment volume from the PRO is set to be 0. In addition, the assumption of uniform flow condition during empty flushing allows the use of thalweg 20 slope, which is 0.0032 according to the measurement in 2011, to represent the energy slope as required in Eq. (1).Then, according to the simulated PRO release during the empty flushing operation, the flushed sediment discharge as well as the desilting volume during the simulating time horizon can be estimated using Eq.(1).The results are presented in Fig. 9.The simulations are then repeated by applying the 10 modified storage balancing curves in Fig. 7 to the period between April and June, the results of which are displayed in Fig. 10.A comparison of Figs. 9 and 10 shows that the modified storage balancing curves effectively enhance the effectiveness of desilting.For instance, strategies with R d max between 0.17 and 0.23 correspond to a maximum annual desilting volume of 0.06 million m 3 /year in Fig. 9, whereas the same strategies in Fig. 10 4), ( 9) and (10) under a specific value of the maximum acceptable monthly shortage ratio,  .In the case study, three values of  , including 0.1, 0.2 and 0.3, are tested.The corresponding optimal storage thresholds to activate and terminate an empty 5 flushing operation are presented in Table 5.The average annual desilting volume and maximum monthly shortage ratio induced by empty flushing are also marked in Fig. 10.Due to the frequency of drought in this system, the optimal strategy associated with  =0.1 is selected due to its conservativeness.Table 6 displays the simulated events of empty flushing based on this calibrated strategy.10 Table 6 presents the monthly shortage ratio in July following the empty flushing operations in 1989 and 2009 both reach 0.41.However, the corresponding shortage increments are both 0; therefore, they did not violate the constraints of Eqs. ( 9) and (10).Figs.
11 and 12 present the hydrographs of inflow to the Tsengwen Reservoir and the total system storage during these two years.As shown in Table 6 and these figures, the Tsengwen 15 Reservoir is nearly empty and the Wushanto Reservoir is nearly full before the initiation of empty flushing operations.Thus, the empty flushing operations only consume the inflow of Tsengwen Reservoir during a 2 to 3 days period.These water consumption volumes are too insignificant to induce the subsequent water shortage seen in July.The primary reason for the subsequent shortage is the delayed arrival of the first typhoon in late July or early August, by 20 which time the total storage falls below the critical limit of the joint operating rule curves and water rationing is applied.Following the arrival of the typhoons, however, the total reservoir storage exceeds the lower limit and even the upper limit of rule curves, thereby alleviating the water shortage.

Validation analysis of the optimal strategies
The optimal strategies in Table 6 are derived by linking the optimization algorithm to  respectively, present the hydrographs of reservoir inflow and total system storage from May to December of these years.Clearly, following the initiation of empty flushing operations in early June of 2010, the monthly water shortage ratio during July is 0.18, which is higher than the 0.12 that would have been the case without empty flushing.The increased shortage ratio is induced by drawdown and empty flushing, which cause the total storage to fall below the critical limit earlier in July.Empty flushing thus necessitates a longer water rationing period. 5 Nonetheless, torrential rains in late July elevate the storage to exceed the lower limit, thereby resolving the shortage crisis.The major impact of water shortage during this period is on the second semiannual irrigation operation, which requires large quantities of water during July.
One of the mitigation measures is to postpone the beginning irrigation schedule no later than August 10.For example, in May and June of 2004, the total storage in the Tsengwen and

Conclusions
This study aims to optimize the performance of empty flushing of one primary reservoir within a multi-reservoir system.Prior to empty flushing, the total available storage in a 5 system is allocated from the primary reservoir to the others in order to create favorable initial conditions and prepare backup water supply to be used during empty flushing.The activation and termination conditions of an empty flushing operation are determined according to whether storage in the primary and auxiliary reservoirs satisfy applicable thresholds.
Optimization analysis calibrates these storage thresholds to maximize the desilting volume 10 without inducing intolerable water shortage.The case study of the water resources system of the Tsengwen and Wushanto Reservoirs of southern Taiwan verifies the effectiveness of the derived optimal empty flushing strategy.
For the sake of clarity, the proposed method only discusses systems without means to artificially create inflows for flushing discharge to the primary reservoir.This simplification may not always be valid.For example, there is currently a trans-basin tunnel under 5 construction that will divert surplus water from adjacent basin to the Tsengwen Reservoir in order to enhance the efficiency of regional water utilization.The diverted water could also serve to scour the depositions of Tsengwen Reservoir as well as to replenish the emptied storage following empty flushing.Application of the proposed method to such a system will require optimization of the transferred discharge as a parameter for maximizing the desilting water supply, it also provides more adequate water to allow recovery and enhanced desilting of existing reservoirs, thus allowing the entire system to advance toward the goal of sustainability.

2 .( 1 ) 9 5 ( 2 )
Qualitative conditions for flushing sediments (QCFS): Compliance with these conditions promotes efficiency of sediment flushing.The key is to identify and take advantage of opportunities with both high inflow and low WSL of the reservoir by performing empty 15 flushing.QCFS1: High inflow during empty flushing High inflow is required to maximize the flushing efficiency by more effectively scouring the depositions of the reservoir.Atkinson (1996) and White (2001) indicated that empty flushing should only be initiated when the inflow is at least double the 20 inflow in normal conditions.The experience with flushing the Zemo-Afchar Reservoir of the former USSR (Chaudhry and Habib-ur-rehman, 2012) suggests that empty flushing is most effective with inflow between 400 to 500 m 3 /s, which is 2 to 2.5 times the average inflow (Bruk, 1985; Singh and McConkey-Broeren, 1990).Long-term observation of Jianshanpi Reservoir in southern Taiwan also revealed that the 25 Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.efficiency of empty flushing peaks during heavy rainfall events when daily rainfall on the reservoir watershed is between 40 to 60 mm.This condition can also be artificially achieved.For instance, during the empty flushing of the Mangahao Reservoir in New Zealand, water was released from another upstream reservoir to enhance the scouring of depositions and thus maximize desilting volume (White, 2001).QCFS2: Low WSL before and during empty flushing a.Before empty flushing is started: During the regular operation, operators could take advantage of periods when the reservoir WSL is low to perform drawdown and initiate empty flushing.In cases where the reservoir has outlets with sufficient capacities, timely drawdowns can be performed shortly prior to expected floods so 10 that the flood inflow can effectively scour and flush out depositions.One example is the Dapu Reservoir, of which WSL is generally the lowest in mid-May.This timing is thus considered as the ideal time to empty the reservoir, with the expectation that subsequent abundant floodwater from May to August can also fulfill the QCWS 1, QCWS 2 and QCFS 1. 15 b.After empty flushing is initiated: Once empty flushing is initiated, the reservoir should remain as close to empty as possible to maintain high flushing efficiency.
10 and duration of empty flushing is flexible according to real-time hydrological and operational conditions.If these conditions are not favorable, the primary reservoir could resume regular operation.This feature distinguishes the present method from previous related studies (Chang et al., 2003; Khan and Tingsanchali, 2009), which mandatorily empty reservoir storage during predetermined periods within a year.This paper also assumed that the water demands in the 15 system require constant supply, thus rendering empty flushing infeasible during parts of the year.To facilitate determination of feasible periods for empty flushing, the following criteria are provided: 1. Meeting QCWS 1 requires periods of low water demand during which backup reservoirs can provide adequate supply during empty flushing.

20 2 .
QCFS 2 dictates that the most opportune time to begin empty flushing is at the end of the dry season.At this time the storage levels of reservoirs are usually at their lowest levels of the year.This ensures that storage can be effectively and efficiently drained by drawdown flushing through the capacitated bottom outlets of the primary reservoir.3. Meeting QCFS 1 requires that the feasible duration for empty flushing should be extended 25 11 Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.into the wet season to ensure adequate inflow for scouring depositions.4.Meeting QCWS 2 requires that the feasible duration for empty flushing should be ended before the end of the wet season to ensure adequate replenishment of reservoir storage after the flushing operation.

5 optimization 1 t = t+ 1
linkage approach.It requires a model to simulate the operations of water supply and empty flushing, thus allowing for quantifying the desilting volume as well as the incremental water shortage generated by a given strategy.The model simulates the process of water supply according to a set of joint operating rules as presented in subsection 2.2.1.When specific quantitative conditions presented in subsection 2.2.2 are achieved, empty flushing in 10 the primary reservoir is activated and the approach in subsection 2.2.3 is employed to estimate the desilting volume.The empty flushing terminates when the conditions presented in subsection 2.2.4 are reached, and the simulation is switched to regular water supply operation until the next time activation conditions are satisfied.The simulation model is linked to an optimization algorithm to calibrate optimal parameters in the activation and 15 termination conditions, according to the formulation presented in subsection 2.2.5.Fig. 1 depicts a flowchart of the analyzing procedure.Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.Are the conditions to activate empty flushing satisfied by the inflow and initial storage of the t th day t = Is the t th day the last day of the simulation time horizon?simulated carryover storage, set the initial storage of reservoirs for the next day Are the conditions to terminate empty flushing reached by the inflow and initial storage of the t th day Yes Perform regular water supply simulation according to the joint operating rules No Perform simulation of empty flushing in the primary reservoir and backup water supply in the other reservoirs.
10 for the Tsengwen and Wushanto Reservoirs in southern Taiwan.The location and associated water resources system of these reservoirs are depicted in Figs. 4 and 6 in the case study section.The release rules stipulate that when the total storage of the two 13 Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.

Fig. 2
Fig. 2 Joint operating rule curves of the Tsengwen and Wushanto Reservoirs

Target
Storage for Indivisual Reservoir (million m 3 ) the curve which indicates target storage of Tsengwen Reservoir curve which indicates target storage of Wushanto Reservoir C U : the storage capacity of Tsengwen Reservoir C D : the storage capacity of Wushanto Reservoir

5 1 .
The current simulating date falls within the pre-evaluated feasible timeframe for empty flushing.

5 supply
may rely solely on the storage preserved in the other reservoirs.During empty flushing, the empirical formula developed by the International Research and Training Center on Erosion and Sediment (IRTCES) in Tsinghua University, Beijing (IRTCES, 1985) is employed for the estimation of releasing sediment discharge from the primary reservoir.The formula is based on measurements from 14 reservoirs in China: 1996), and  is the flushing coefficient, associated with the characteristics of the sediment and topography of the reservoir.

20 1 . 2 .
The flushing should be terminated when the flood flow has raised the WSL of the primary reservoir and inflow subsequently recedes to be below the capacity of associated bottom Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.outlets.This situation indicates that the operation has been successfully timed to encounter a flood and should thus be ended when the flood ends.The flushing should be ended when the storage of backup reservoirs decreases to below a threshold d T .This condition prevents short-term water shortages following flushing operations resulting from insufficient storage.During the flushing operation, the primary 5 reservoir will remain empty in the absence of floods, so providing water supply will gradually reduce available storage in the other reservoirs.A higher value of d T will cause the storage below threshold more quickly, thus reducing the window of operation for empty flushing.Nonetheless, adequate reservoir inflow and proper storage reallocation after an earlier termination of one flushing operation will facilitate the re-initiation of a 10 subsequent operation during the feasible period for empty flushing.Thus, under conditions of a higher d T value, the pattern of empty flushing may be transformed from a few operations of longer duration into multiple intermittent operations of shorter durations.A generalized water allocation simulation model (GWASIM) developed by Chou and Wu (2010) is used to simulate the alternating operations of empty flushing and joint water 15 supply according to the aforementioned rules and conditions.The structure of this model is formulated in network flow programming.It has already been implemented in the planning and management of all major water resources systems in Taiwan.Details of its simulations regarding the operations of multi-reservoir systems, such as those in Subsection 3.1, can be found in Chou et al. (2006) and Chou and Wu (2014). 20

,
Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.calibrated to maximize the total desilting volume without inducing intolerable water shortage scenarios.Since empty flushing is restricted to a feasible period suggested to span from the end of the dry season to the early wet season, the occurrence of a subsequent flood which may cause reservoir spillage will fully compensate for the impact of probable water shortage after empty flushing.Thus the incremental shortage will be concentrated in a few months following 5 the feasible flushing period, during each of which the monthly shortage increment and ratio is calculated respectively: ,1,2,...,n m , n = 1,2,...,n y m = 0,1,...,n m , n = 1,2,...,n y the water shortage increment and ratio during the m-th 10 month following the feasible period of empty flushing in the n-th simulating year; m D denotes the water demand during the m-th month following empty flushing; shortages under conditions with and without empty flushing operations, n m is the number of months within which the impact of empty flushing on water supply is carried over; and n y is the number of simulating years.The formulation of the 15 optimization problem is as follows: Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.

5 characteristics
indicate that the climate and operating conditions during May and June are favorable for empty flushing of Tsengwen Reservoir.

Fig. 5
Fig. 5 Demand and inflow patterns of Tsengwen Reservoir during ten-day periods

Fig. 6 Fig. 7 Fig. 8 .
Fig. 6 Network of the joint operating system of Tsengwen and Wushanto Reservoirs

Fig. 8 , 1 , and I n d 1 ,
Fig. 8 Relationship between the flushing coefficient and WSL of Tsengwen ReservoirTo assess the impact on water supply following empty flushing, the ratio and

result in an 15 increase
of desilting volume reaching 0.54 million m 3 /year.

Fig. 9 Fig. 10 3 . 4 Optimization of empty flushing strategies 5 According
Fig. 9 Simulation results of various empty flushing strategies using the original storage balancing curves Fig. 11 The hygrographs of Twengwen Reservoir inflow and total system storage throughout 1989

5 the
model that simulates operations according to the records of daily reservoir inflow between 1975 and 2009.Following this calibration period, the records through the end of 2013 are used to verify the effectiveness of the established strategy.The results of the validating simulation indicate that two flushing operations could have been conducted during this period, one in 2010 and the other in 2013.

10WushantoFig. 13 Reservoir inflow and storage throughout 2010 5 43
Fig. 13 Reservoir inflow and storage throughout 2010 10 volume during empty flushing.The validity of the employed empirical formula to estimate the discharge of flushed sediments should also be investigated when more field measurements become available.One scenario to which this formula may not apply is drawdown flushing, during which the flow through the outlet is pressurized rather than adhering to open channel flow conditions.Numerical modeling may be employed to more comprehensively and 15 accurately simulate the flushing process in order to optimize the desilting volume.The operators of Tsengwen Reservoir currently oppose empty flushing due to the high pressure of water shortage, even though reservoir sedimentation imposes a more severe threat in the long term.Nonetheless, this perspective might change with the completion of the sediment sluicing tunnel that is currently under construction as well as the upstream trans-20 basin diversion tunnel.The design capacity of the sluicing tunnel is 815 m 3 /s, which enables pre-emptying the reservoir shortly before an expected flood with less uncertainty.It reduces the risk that the inflow is inadequate after the reservoir is emptied.The urgent need of desilting also endows a new role to the conventional projects of water resources development, such as the aforementioned trans-basin diversion tunnel.In addition to elevating the yield of 25 45 Hydrol.Earth Syst.Sci.Discuss., doi:10.5194/hess-2016-61,2016 Manuscript under review for journal Hydrol.Earth Syst.Sci.Published: 7 March 2016 c Author(s) 2016.CC-BY 3.0 License.