A Simulation-Optimization System to Assess Dam Removal With a Focus on Environmental Degradations at Downstream

7 The present study proposes and evaluates an integrated framework to assess dam removal in which 8 downstream river habitats and reservoir operation were simulated in three different statuses including 9 conventional reservoir operation optimization, optimal release considering environmental aspects in the 10 structure of the optimization model and natural flow. Fuzzy physical habitat simulation was used to 11 assess physical habitats. Moreover, ANFIS based model was utilized to simulate thermal tension and 12 dissolved oxygen tension at downstream habitats. Particle swarm optimization was applied in the 13 optimization models. Results of the optimization models and habitat suitability in the natural flow were 14 compared by applying different measurement indices including reliability index, vulnerability index, 15 The Nash – Sutcliffe model efficiency coefficient (NSE) and root mean square error (RMSE). Based on 16 the results in the case study, reliability of water supply might be reduced in the optimal release for 17 environment and demand. Furthermore, optimal release for environment is not able to protect 18 downstream aquatics habitats properly. Thus, if protecting downstream habitats is aimed, dam removal 19 might be logic in the case study. The main limitation of the proposed method is high computational 20 complexities.

responsible for water supply. Moreover, they might have other rules such as electricity supply and flood 26 control (Altinbilek, 2002). Increasing population might raise environmental challenges for the river 27 ecosystems (Postel, 1998). On the one hand, required water demand is increased that means more 28 offstream flow is needed. On the other hand, instream flow is decreased that means increasing 29 environmental degradation in the river ecosystems. Environmental impacts of the dams have been 30 highlighted in the literature in the recent decades (e.g Tahmiscioğlu et.al, 2007). Owing to importance 31 of the river ecosystems and other possible impacts of these structures, dam removal is one of the 32 challenging issues that has been focused in the recent years. Previous studies discussed regarding 33 importance and feasibility of the dam removal projects (e.g Bednarek, 2001;Foley et.al, 2017;Major 34 ert.al, 2017). It seems that different aspects must be considered in the dam removal. Thus, complex 35 modeling might be needed in this regard. An integrated framework is helpful to assess how dam removal 36 might be advantageous for the environment. Many aspects might be considered in the analysis of the 37 dam removal. However, impacts on the aquatic habitat at downstream is one of the effective aspects in 38 the assessment of the dam removal projects. 39 Dams are very expensive hydraulic structures. Hence, optimal reservoir operation was highlighted in 40 the literature as the most critical task after construction of the reservoirs (Ahmad et.al, 2014). In fact, a 41 reservoir should have an optimal operation to maximize benefits. Hashimoto et,al 1982 developed an 42 applicable form of the loss function that has widely been used for the optimization system of the 43 reservoirs. Moreover, Datta and Burges, 1984 highlighted adding storage loss to the optimization 44 system as well. This form of loss function has been used even in the recent studies due to applicability 45 and efficiency (Ehteram ert.al, 2018). On the other hand, optimization solution is another principal 46 aspect in the reservoir operation models. Many methods such as linear programming (LP), non-linear 47 programming (NLP) and dynamic programming (DP) have been applied in this regard (e.g Reis et.al, 48 D., 2006;Arunkumar and Jothiprakash, 2012). However, LP methods are not efficient due to non-linear 49 nature of the reservoir operation. NLP and DP might not be efficient due to complexities of the defined 50 objective function. Hence, advanced computational methods such as metaheuristic algorithms have 51 been utilized and recommended in many recent studies. A long list of the classic and new generation some minor changes and technical consideration related to the case study. Some simulations, 106 assessments and optimizations are required to implement the proposed framework that are presented as 107 follows. 108 109

2-1-physical habitat simulation 110
Physical habitat simulation is used to assess physical habitat suitability in different developed scenarios 111 as explained in the previous section. Previous studies indicate applicability of this simulation for the 112 river ecosystem management (Sedighkia et.al, 2017). Figure 1 displays workflow of the physical habitat 113 simulation in the present study. We applied 1D fuzzy physical habitat simulation, which is one of the 114 most efficient methods for physical habitat assessment. It should be noted that previous studies 115 corroborated the applicability and efficiency of the one-dimensional hydraulic modelling for the 116 physical habitat simulation (Jowett and Duncan, 2012). Development of the fuzzy habitat rules is the 117 most important step in the physical habitat simulation. We used a combination of the field studies and 118 expert opinions to develop fuzzy rules. Field studies consist of two main parts including ecological 119 studies or fish observations and surveying the river cross-sections that is required for hydraulic 120 simulation. Several methods have been developed for the fish sampling in the rivers including direct 121 and indirect methods. Discussion on advantages and disadvantages of each method have been presented 122 in the literature (Harby et.al, 2004). Electrofishing was used in the present study due to its advantages 123 and previous experiences of the research team regarding this method. Moreover, depth and velocity 124 were measured in the microhabitats simultaneously by the propeller and metal rule (Harby et.al, 2004). 125 It should be noted that the Brown trout was considered as the target species in the case study. A 126 representative reach with length of 1000 meters was considered in the physical habitat simulation. 127

2-2-thermal habitat assessment 130
Thermal habitat assessment is another requirement in the proposed framework to analyse dam removal. 131 The first need to achieve this purpose is to simulate stream temperature. Hydrodynamic models such as 132 SSTEMP has been developed to simulate stream temperature (Bartholow, 2000). However, they are not 133 flexible to apply in the structure of the optimization model. Hence, we need a flexible and applicable 134 method. Data driven models are one of the efficient options in this regard. We developed a new data 135 driven model to simulate stream temperature at downstream of the reservoir. Using neural networks is 136 one of the most applicable methods to develop a robust data driven model in the water quality simulation 137 (e.g Chatterjee et.al, 2017). Owing to inherent disadvantages of the artificial neural networks (ANN) 138 such as working as black box, combining fuzzy inference system and neural network might generate 139 the higher efficiency (Dumitru and Maria, 2013). Hence, we utilized an adaptive neuro fuzzy inference 140 system (ANFIS) to simulate stream temperature in the proposed framework (developed by (Jang, 1993). 141  More details on the responsibility of each layer is available in literature (Awan and Bae, 2014). Table  143 1 displays main characteristics of the developed ANFIS based model to simulate stream temperature. 144 As displayed, several inputs are considered in the model that demonstrates data driven stream 145 temperature model might be complex. In the next step, it is required to convert stream temperature to the thermal habitat tension. Sedighkia et,al 2019 used a biological based model to asses thermal habitat 147 tension of the Brown trout that has been adopted in the present framework. It should be noted that using 148 other target species might need its own thermal tension model. Figure 3 displays thermal tension curve 149 for the Brown trout that was used in the case study. It should be noted that optimization system was 150 used in the monthly scale. Thus, we applied recorded monthly data in different stations at the 151 downstream. Moreover, using an index is essential to measure the robustness of the data driven model. 152 The Nash-Sutcliffe model efficiency coefficient (NSE) was used in this regard which has initially been 153 developed for measuring performance of hydrological models (Gupta et.al, 2009). However, it is 154 useable for any types of the simulation to compare results with recorded data or observation for 155

2-3-Dissolved oxygen modelling 172
Dissolved oxygen is an important water quality factor for the river habitats that might shows suitability 173 of water quality for the aquatics. Because, other pollutants might affect the concentration of the DO in 174 the river. Thus, we selected DO as the representative for all of the water quality parameters. We used 175 ANFIS based model to simulate concentration of the dissolved oxygen as well. Monthly scale was 176 considered in this data driven models. Table 2

2-4-optimization model 191
In the both scenarios, we need optimization models that are able to optimize reservoir operation. In the 192 first scenario that generates the highest benefits from the reservoir, equation 2 was used as the objective 193 function. This loss function minimizes different between target (maximum requested water demand) 194 and release for demand. Each optimization model might need some constraints in its structure. We 195 considered some constraints in the proposed optimization model for the scenario 1 including minimum 196 operational storage, maximum storage and maximum water demand. In fact, storage in the reservoir 197 should not be less than minimum operational storage and more than capacity of the reservoir. Moreover, 198 release for demand should not be more than maximum requested water demand. We focused on the 199 metaheuristic optimization as an efficient solution for the optimization process. Penalty function is one 200 the known methods to handle constraints in the metaheuristic optimization that has been used in many 201 previous studies (developed by Agarwal and Gupta, 2005). Three penalty functions were considered in the optimization model for the scenario 1 as follows. In fact, these penalty functions increase penalty 203 for the model when the constraints are violated in the optimal solution. Equation 6 displays final form 204 of the objective function for the optimization model in the scenario 1. C1 to C3 are constant coefficients 205 that were determined based on initial sensitivity analysis of the optimization model. 206 (2) 207 where Dt is maximum water demand, Rt is release for demand or total release for environment and 212 demand in the scenario 2, St is storage, Smax is maximum storage or capacity of the reservoir and Smin 213 is minimum operational storage in the reservoir. Moreover, it is necessary to update reservoir storage 214 in each time step that was carried out by the equation 7. Overflow was calculated by the equation 8 215 where Et is evaporation from the reservoir, At is area of the reservoir, It is inflow of the reservoir and T 216 is time horizon. 217 We developed a novel optimization model for the scenario 2 in which water demand loss, storage loss 220 and environmental degradation are minimized. Hence, loss function used in the previous optimization 221 model is one of the terms in this optimization model. In fact, four main terms were considered including 222 water demand loss, physical habitat loss, and thermal habitat loss and dissolved oxygen loss as follows .PHN, THN and DHN are normalized weighted useable area, thermal suitability and dissolved oxygen 224 suitability in the natural flow respectively. PHM, THM and DHM are normalized weighted useable 225 area, thermal suitability and dissolved oxygen suitability in the optimal release for environment 226 respectively. 227 We utilized penalty function method in this optimization model as well. Used penalty functions were 229 the same. Final form of the objective function for the optimization model in the scenario 2 is displayed 230 in the equation 10. 231 Particle swarm optimization (PSO) is one of the classic and widely used metaheuristic algorithms in the 234 optimization models (Eberhart and Kennedy, 1995). Owing to previous successful application of the 235 PSO, we utilized this optimization algorithm for the both optimization models. More details regarding 236 the methodology of this algorithm has been addressed in the literature (Eberhart and Kennedy, 1995). 237 However, Figure 5 displays flowchart of this algorithm to find the best solution. 238 (NSE) and RMSE were utilized in this regard as displayed in the following equations. It should be noted 252 that optimal suitability is compared with natural condition at the downstream river reach. 253

2-6-Case study 263
Tajan river is one of the most important rivers in the southern Caspian Sea basin in Iran where is a 264 valuable habitat for many fish species. Owing to importance of the irrigation demand, Rajaei dam has 265 been constructed at upstream of this river. Water demand is being pumped from the reservoir or before 266 valuable downstream river habitats. In fact, release for environment is very low which raises challenges 267 between regional water authority and regional department of environment. On the one hand, dam 268 removal might be helpful to reduce severe environmental degradation at downstream. On the other 269 hand, environmental optimization might decrease environmental degradation at downstream habitats. 270 Owing to these ambiguities in the environmental management of this river basin, an integrated 271 framework was necessary to assess dam removal and complexities of environmental management in 272 this basin. Figure 6 displays land use, river network and location of the Rajaei reservoir in this basin.

3-Results and Discussion 276
In the first step, it is required to present and discuss regarding results of the physical habitat simulation. 277  In the second step, validation result of the thermal data driven model should be presented.  In the third step, results of the optimization should be presented and discussed. We selected a 72 months 324 period as the simulation period in the present study that was a challenging period for the reservoir and 325 river basin due to low inflows of the reservoir in some months. Figure 11 displays inflow of the 326 reservoir, optimal release for water demand and maximum water demand. It should be noted that 327 maximum water demand was considered based on recommendations by the regional water authority in 328 the case study. In fact, this is maximum water demand for the river basin that might be supplied by the 329 surface and ground water resources. However, it was considered to supply water demand by the  loss needs related measurement indices that will be discussed. Figure 13 displays release for demand, 353 reservoir inflow and maximum demand in the scenario 2. It seems that considering environmental 354 aspects with a focus on environmental flow in the optimization model changes supplied demand 355 especially in the first years of the simulated period due to low inflow of the reservoir. Hence, low 356 supplied flow in some time steps might be a challenge for using the proposed optimization framework 357 in the scenario 2. It should be noted further analysis on this issue is depended on the challenges and 358 certain issues of each case study that means it might be different case by case. In our case study, release 359 for demand in each time step was not very effective. In contrast, total release in the simulated period 360 was important. Thus, using reliability index seems logic in this case. However, engineers might use 361 other indices in the other case studies. More water is supplied in the third, fourth and fifth year of the 362 simulated period. Figure 14 displays storage time series in the scenario 2 that indicates changing the 363 storage in the reservoir is considerable compared with the scenario 1. Less storage in some simulated 364 time steps is obvious. Hence, storage loss might be increased in the optimal release for environment 365 and demand. However, accurate analysis needs using measurement indices. 366 Moreover, environmental results should be presented and discussed in the Scenario 2. Figure 15  for interpretation of the results in the further applications. There is a significant change compared with 370 natural flow. Generally, it seems that using optimal reservoir operation considering environmental 371 aspects is not able to increase useable area as much as natural flow. It might corroborate the opinion of 372 the environmental advocators to remove dam. Reducing NWUA is a serious concern for some time 373 steps in the simulated period. It should be noted that in some time steps, NWUA in the scenario 2 is 374 more than natural flow. In fact, optimal operation might be able to increase suitability of the physical 375 habitats compared with natural flow as a positive aspect of the dam. However, it was occurred for rare 376 time steps in the case study. 377 Thermal tension is another aspect that should be discussed in the present study. Figure 16  results of the optimization model might be reliable. 396 As mentioned, full discussion on the results and making a decision in the case study need using 397 measurement indices. Figure 18 displays reliability index of the water demand in the both scenarios. 398 Decreasing reliability of water supply is remarkable in the case study. In fact, reliability is changed 15% 399 approximately that demonstrates economic benefits from the reservoir might reduce considerably due 400 to using an optimal operation considering environmental aspects. Moreover, figure 19 displays 401 measurement indices for the storage loss including RMSE and VI. Results demonstrate that changing 402 storage loss in the scenario 2 compared with scenario 1 is not tangible. Thus, using and optimal reservoir 403 operation considering environmental aspects in the structure of the optimization model might not 404 weaken storage benefits in the reservoir of the case study. 405 Table 4 displays NSE and RMSE for three environmental aspects in the proposed framework including 406 physical habitat loss, thermal habitat loss and dissolved oxygen loss. It should be noted that only 407 positive difference between simulation and optimization in the scenario 2 and natural flow was 408 considered to compute indices. In other words, if physical habitat loss is more than natural flow in the 409 scenario 2, it will be considered equal to NWUA in the natural flow. This consideration was applied 410 regarding thermal tension and dissolved oxygen tension as well. NSE for physical habitat loss is -3.43 411 that indicates optimization model is not able to provide physical habitat loss as suitable as natural flow. 412 Moreover, RMSE for physical habitat loss is 0.41 that demonstrates considerable difference between 413 suitability of physical habitats in two statuses. Moreover, NSE for thermal tension shows that 414 performance of the optimization model is acceptable in terms of thermal tension. However, RMSE is 415 considerable. Because, 29% of the thermal tension might make habitat unsuitable especially for the 416 sensitive species such as the Brown trout. Furthermore, performance of the optimization model is not 417 defensible in terms of the DO tension based on computed indices. 418 Finally, it is possible to judge on the feasibility and advantages of the dam removal project in the case 419 study based on all of the results. It seems that optimal operation considering environmental flow is not 420 applicable in the case study. On the one hand, reliability of water supply was reduced which is 421 problematic for the regional water authority. On the other hand, optimal operation in the scenario 2 was 422 not able to provide environmental suitability at downstream of the reservoir. Hence, it seems that using 423 other options in the case study for supply of water such as storage in the pools of the farms should come 424 into picture, if protecting environmental values at downstream is aimed. In fact, dam removal seems 425 necessary in terms of suitability of downstream habitats. However, implementing dam removal might 426 need other simulations and optimizations. The main advantage of the proposed method is to provide a 427 practical simulation-optimization framework to simulate environmental aspects of the dam removal. 428 However, dissolved oxygen results indicates that reducing water pollutant at downstream of the 429 reservoirs should be considered in the habitat restoration projects as well. 430 Each novel method might have some limitations that should be discussed. The main limitation of the 431 proposed framework is high computational complexities. This term might be defined as required time and memory for using an algorithm in the optimization. Owing to using ANFIS based model in the 433 structure of the optimization model, computational complexities is considerable. Thus, using this 434 framework for long-term period or numerous simulations might need powerful computers and it might 435 be time consuming. Moreover, we considered water demand in the proposed framework based on the 436 case study. It is needed to consider other aspects such as hydropower in the future studies. Furthermore, 437 it is essential to consider uncertainties of the ANFIS based model for further applications of the results.

4-Conclusion 469
Present study proposed a novel framework to simulate environmental aspects of the dam removal for 470 making decision on the necessities and advantages of the dam removal projects. Two scenarios were 471 considered including a conventional reservoir operation optimization and optimal operation considering 472 environmental aspects in the structure of the optimization model. First scenario and second scenario 473 were compared in terms of the water supply and storage losses. Moreover, the outputs of the scenario 474 2 was compared with the natural flow in terms of physical habitat loss, thermal habitat loss and 475 dissolved oxygen habitat loss by applying NSE and RMSE as measurement indices. Based on the results 476 in the case study, reliability of water supply was remarkably reduced in the scenario 2 that demonstrates 477 optimal release considering environmental issues might be challenging for the water demand 478 management in the river basin. Moreover, optimal release considering environmental aspects is not able 479 to reduce tensions and losses as suitable as natural flow. Thus, dam removal is recommendable in terms 480 of environmental suitability at downstream aquatics habitats. Simple structure of ANFIS based data driven model   Land use, location of the Rajaei reservoir and river network map of Tajan basin. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.   Validation results of the data driven stream temperature model Figure 10 Validation results of the data driven dissolved oxygen model Figure 11 Reservoir in ow, release for demand in the scenario 1 and maximum water demand in the simulated period Figure 12 Storage of the reservoir in the simulated period for the optimization model in the scenario 1 Figure 13 Reservoir in ow, release for demand in the scenario 2 and maximum water demand in the simulated period Figure 14 Storage of the reservoir in the simulated period for the optimization model in the scenario 2   Reliability index for water supply Figure 19 Measurement indices for storage loss