Modelling mitigation measures for smolt migration at dammed river sections

There is no generic solution to establish safe passage of downstream‐migrating fish passed hydropower facilities, and mitigation measures are species and site specific. Development of solutions is thus often based on “trial and error,” and modelling‐based approaches may significantly reduce cost and time to arrive at successful mitigation. Here, we explore such an approach by combining data on fish migration and hydraulic modelling. First, we performed a positional telemetry study at a dammed section of a Norwegian river, where 100 Atlantic salmon smolts were tagged to track their downstream movement at the vicinity of a hydropower intake channel and bypass gates. An explanatory model was developed to explore mechanisms of migration route, into the intake towards the turbines or through the bypass gates. Next, flow conditions during the smolt run was numerically modelled to explore the physical environment of the tracked smolts. The joint results from the two approaches supported the general assumption that downstream migration is strongly influenced by flow patterns and showed that fish entering the study site closer to the riverbank where the intake channel is located were more likely to enter the intake due to the strong currents towards the intake. Finally, a suite of measures to guide salmon smolts past the hydropower intake were proposed based on the findings and local conditions and tested by hydraulic modelling. We found that most of the measures that were likely candidates for field trials would most likely fail at improving safe passage, and only a rack‐type guiding boom was promising. The presented combination of telemetry migration data and hydraulic modelling illustrates the value of evaluation of mitigation measures prior to implementation.


| INTRODUCTION
Hydropower plants (HPPs) on rivers are often detrimental to the fluvial fauna, and migratory fish species are particularly influenced by river fragmentations (Bunt, Castro-Santos, & Haro, 2012;Metcalfe & Craig, 2012). During their migration towards to the ocean, Atlantic salmon smolts (Salmo salar L.) may face flow diversion at water intake sites where route choice is crucial for successful downstream passage (Larinier, 2008). Different mitigation measures have been applied to guide the fish away from hydropower intakes (Albayrak, Kriewitz, Hager, & Boes, 2018;Boes, Albayrak, Kriewitz, & Peter, 2016;EPRI & DML, 2001) or attract them towards a bypass facility (Calles, Karlsson, Hebrand, & Comoglio, 2012;Scruton et al., 2007). A general solution does not exist, and mitigation measures are usually case specific and a function of local morphological, hydrological, operational, and ecological factors. Typically, extensive monitoring at each site is needed to identify the key criteria for successful mitigation (Wilkes et al., 2018), followed by monitoring of passage success of implemented measures . Downstream migration solutions thus often depend on a "trial and error" approach , sometimes followed by revisions of the measure. The "trial, error, and revision" approach is typically expensive and time consuming due to the necessary monitoring before and after implementation. To reduce time and cost, there is a need for better tools to test different measures without actual implementation.
Computational fluid dynamics (CFD) in 3D has been used to explore relevant hydraulic conditions for target species in a fishway (Fuentes-Perez et al., 2018) and to assess the attraction flow at a hydropower plant (Gisen, Weichert, & Nestler, 2017). Moreover, Goodwin, Nestler, Anderson, Weber, and Loucks (2006) developed an algorithm that was able to forecast response of downstream migrating individuals and fish schools to specific hydraulic conditions in order to evaluate alternative bypass designs. Nestler, Goodwin, Smith, Anderson, and Li (2008) explored hydrodynamic cues used by outmigration juvenile salmon and concluded that their swimming path can explained by fluid dynamics and geomorphology and linked to sensory capacities of the fish. Khan, Roy, and Rashid (2008) utilized the same approach and demonstrated how CFD models could help to assess complex hydraulic engineering problems in relation with juvenile fish migration over dams. However, the potential for in situ combination of telemetry data with 3D hydraulic modelling to obtain successful downstream migration solutions is largely unexplored.
In the present study, we combine telemetry migration data and CFD to evaluate a suite of different methods to guide migrating Atlantic salmon smolt past a hydropower intake. For that, the telemetry data are used to explore migration patterns and route choice in relation to governing flow patterns as revealed by CFD. Next, we use these patterns to explore different mitigation measures in the CFD model and evaluate to what extent the measures are likely to guide fish past the hydropower intake. By doing so, we illustrate the value of using combined modelling to test mitigation measures prior to implementation and to reduce the need for the "trial, error, and revision" practice.

| Study site
This study was carried out at the Bjørset intake to the Svorkmo hydropower plant in the River Orkla in Central-Norway (63°03′18.7″N, 9°39′47.8″E). During parts of the smolt migration period, the majority of the flow goes towards the HPP intake, and high fish mortality is expected in the high-head HPP with Francis turbines. The intake is controlled by a dam with four identical gates (closed height and maximum operational water surface elevation: 129.50 m above sea level [m a.s.l.]; minimum water surface elevation throughout the field surveys: 129.10 m a.s.l.) and pool and weir-type fishways near each bank to allow passage of upstream migrating Atlantic salmon ( Figure 1). Due to the low head (1.80 m), the fishways and gates represent safe passage opportunities for smolts. From May 1 to October 31, the northernmost gate (Gate 1) is open to release the stipulated minimum flow (20 m 3 s −1 ) whereas the other gates are used during flood conditions. For the remainder of the year, minimum flow (4 m 3 s −1 during winter) is released through the northern fishway. Approximately 100 m upstream from the dam an intake channel is located at the north side of the river. The intake area provide water through the intake channel and tunnel to the Svorkmo hydropower plant with a maximum capacity of 55 m 3 s −1 . A concrete wall has been placed at the entrance of the intake channel with two openings at the bottom (1.5 × 25.8 m each) to prevent smolt (and ice or debris) to enter the intake channel. The top of the openings are 2.0-2.5 m below the water surface, depending on the river water level. Due to the relative small area of the openings in the wall (77.4 m 2 ) combined with the flow capacity, racks that prevents smolt entry (recommended 10-to 15-mm bar spacing; Fjeldstad, Pulg, & Forseth, 2018) cannot be installed in front of the wall without risk of impingement and mortality at the rack (DWA, 2005). This study was carried out in the dammed intake area, which is approximately 500 m long and 80-100 m wide with an average water depth of 2.5 m (Figure 1).

| Fish telemetry, positioning, and statistical analyses
During the smolt migration from late April to early June 2016, Atlantic salmon smolts were trapped 1,800 m upstream of the study site, and 100 individuals were tagged with acoustic transmitters (Lotek,200 kHz,7.5 × 17 mm, Lotek Wireless Inc., Newmarket, Ontario, Canada) with a 2.01 s burst interval. Fish were anaesthetized by immersion in an aqueous solution of 2-phenoxy-ethanol (0.7 ml l −1 , Sigma Chemical Co., St. Louis, MO, USA) and then placed ventral side up onto a V-shaped surgical table. An incision (~1 cm) was made with a scalpel on the ventral surface posterior to the pelvic girdle. The transmitter was inserted through the incision and pushed into the body cavity in front of the pelvic girdle. The incision was closed with two independent sutures (5/0, Ethicon, Prolene). During surgery, a 25 mg l −1 solution of 2-phenoxy-ethanol was circulated over the gills.
Body mass (M), length (L), and smolt index (which indicates the parrsmolt transition of an individual, I S , Johnston & Eales, 1970) were recorded as biotic parameters for all tagged fish (Table 1). After tagging, fish were left to recover from the surgery before being released.
Twenty-seven acoustic receivers (Lotek 200 kHz WHS 3050) were installed in the study site ( Figure 1) and positioned using a GNSS receiver with a VRS-service providing 2 cm accuracy (CPOS-service from the Norwegian Mapping Authority). Receivers were either fixed on the concrete structures with the hydrophone at 70 cm depth or on a pole.
The detections of each smolt from the different hydrophones were processed by the software package YAPS (Baktoft, Gjelland, Økland, & Thygesen, 2017) to obtain movement trajectories including associated error estimates (standard deviation). This procedure produced migration tracks of each smolt with a position every 2 s. In total, migration tracks were obtained for 91 out of the 100 tagged smolts. Among the remaining nine smolts, one provided invalid data (apparently suffered predation), and eight smolts never appeared at the study site. Esti- Watson's two-sample test (Mardia & Jupp, 2010) was used to test differences between the smolts migrating into the intake and through the gates (Smolt Groups A and B, Table 1). The smolt migration route was analysed using generalized linear models. A binomial model with logit-link function (Zuur, Hilbe, & Ieno, 2013) was used to derive a relationship between migration route (0 = gates, 1 = intake) and different environmental variables. The aforementioned W elev , Q in , Q intake , and Nr Gates variables and the ratio of the flow through the dam and intake channel (Q dam /Q intake ) were added as hydrological and   Table 2.

| CFD modelling
Flow properties at the computational domain were captured by using one-phase pimpleFoam solver from OpenFOAM (Version 4.1.0, Greenshields, 2015), which discretise Reynolds-averaged Navier-Stokes equations and was associated with the standard k-ε turbulence model. The finite volume method to solve the equations by using the PIMPLE algorithm (Higuera, Lara, & Losada, 2013) for the pressure- As the majority of the tagged smolts moved through the study area when only one gate was open, such condition was modelled for low discharge (LQ) and high discharge (HQ) using OpenFOAM (see Table 2 for hydraulic boundary conditions). Simulated water velocity magnitudes (also known as resultant velocity, U mag [m s −1 ]) at 0.5 m below water surface were used to characterize the flow environment where the smolts were expected to travel (Thorstad, Whoriskey, Rikardsen, & Aarestrup, 2011).
After calibration, the two actual cases and 10 additional scenarios were prepared and simulated according to the desired mitigation conditions with the added elements. Simulated times were determined based on the flow development at the study site. The scenarios were set to represent 33 and 27 min of flow by simulation LQ and HQ conditions, respectively. Velocity field by CFD were used for further evaluation of the different mitigated conditions.
At the cases representing the LQ and HQ conditions, the release of five particles were modelled 0.5 m below water surface per section

| Hydraulics
Both at the LQ and HQ, horizontal velocities (x and y directions) dominated the modelled area (Figure 2a,b), except at the immediate proximity of the intake entrance where strong downward velocity occurred due to the openings at the bottom (Figure 2c,d).
The maximum velocity magnitude (U mag ) was approximately 0.64 m s −1 at LQ and 1.0 m s −1 during HQ and appeared at the vicinity of the intake. Sediment deposition at the upstream part of the modelled area and outside of the telemetry study area has significant impact on the velocity field as it yielded a split streamflow with high velocities appearing along both riverbanks (Figure 2a,b). The

| Telemetry and passage success
A total of 91 smolts were detected at the study site during the period of May 5 to 21, of which 17 (19%) left the area through the   (Figure 3).

| The migration route model
Migration route modelling were performed with all (both biotic and abiotic) explanatory variables: logit(π) = log 10 [π(1 − π) −1 ] = intercept +β 1 V 1 +β 2 V 2 +…+β 8 V 8 , where π is the probability that fish leaves the domain through the intake, β 1 ,β 2 …β 8 are the estimated coefficients, and V 1 ,V 2 …V 8 are the different variables listed in Section 2.2. Variables in the full model were sequentially removed using the AIC (Akaike, 1974) value and by eliminating all variables that contribution was minor. The final model contained three variables: the discharge through intake (Q intake ), the number of open gates at the dam (Nr Gates ), and the distance of the starting position from the south bank (D S ) of each smolts as logit(π) = log 10 [π(1 − π) −1 ] = β 1 Q intake +β 2 Nr Gates +β 3 D S (Table 3).
According to the model a smolt is more likely to end up in the intake channel when the intake discharge (Q intake ) is high and if the smolt appear at the intake area close to the northern bank (high distance values, D S ). In contrast, the probability for intake passage is Indeed, modelling five particle tracks at each sections at both flow conditions, produced a pattern where no particle entered the intake at the south section (one uncertain), followed by 10% (one uncertain; one particle at LQ), 70% (five particles at LQ and two at HQ), and 100% of the particles entering the intake in the sections towards the north.

| Mitigation measures
The two simulated scenarios revealed complex flow patterns at the entire study site, particularly at the proximity of the intake entrance    On the basis of these four major mitigation types, 10 different mitigation cases (different combination of measures at two flow conditions, see Table 4) were simulated to evaluate their impact on the flow field and potential impact on smolt migration route. The latter was based on qualitative assessment of the flow fields and their directions (towards the intake or dam).
The first mitigation measures (1.1 and 1.2) simply involved changing the gate operation with the aim of shifting the main current closer to the southern bank, away from the intake area. It had only a marginal effect on the flow field in front of the intake (Figure 5a,b in comparison with Figure 2) but a large effect on the flow between the intake and the dam. Here, the main current was widened, the recirculation area at the south side was reduced, and a new recirculation area was formed at the upstream side of Gate 1. Due to the continued strong currents against the intake, this mitigation is unlikely to improve passage efficiency.
The second mitigation measure (2.1) was more extensive, involving the construction of spurs on both riverbanks at the upstream side of the intake, with the aim to change the flow pattern away from the intake area. This solution was also combined with alternative gate operation (2.2). The main effect was the joining of the two high velocity flows at the two banks into one major flow towards the lower part of the intake (Figure 5c,d). Velocities exceeding 1.0 m s −1 were seen at the new structures, but they abated as the current deflected towards the intake. Even higher velocities appeared locally at the downstream end of the intake, which dampened towards the dam. Changing the gate operation had no effect on the flow towards the intake, and   Floating fish guidance booms were the fourth major mitigation types tested (4.1-4.4). Rather than directing the flow, the aim was to guide smolts from the northern to the southern bank. However, both solid and permeable floating booms were expected to influence flow fields, and this effect was explored (Figures 6 and 7).

| DISCUSSION
In this paper, we combined general knowledge on Atlantic salmon smolt migration, local telemetry migration data, and migration route modelling with river section scale hydraulic modelling by CFD. The aim was to evaluate a suite of mitigation measures to prevent smolts from entering a hydropower intake and their migration through the turbines. Our primary aim was not to provide a solution for the particular case but rather to illustrate the value of combining knowledge on migration patterns with river reach hydraulic modelling as a novel tool to reduce the need for "trial and error" in designing downstream migration solutions in general.
The general assumption that downstream migrating salmon smolts follow the main flow (Rivinoja, 2005;Williams et al., 2012) was supported with several findings from the telemetry study. First, migration tracks showed that the smolts tended to enter and follow the two main streamflows along the two riverbanks, whereas the recirculation and dead-water zones were barely visited. Second, the developed migration route model showed that the route (intake or gates) was strongly dependent on the position of entry for the smolts to the study site. In agreement with simulated particle movement in the model, fish arriving closer to the southern riverbank were more likely to leave through the gates compared with smolts entering at the opposite bank where the intake is located. The southern flow goes towards the gates, whereas the northern flow goes into the intake. Although several of the modelled measures likely would have been regarded as promising solutions, even by hydraulic and fish migration scholars, the hydraulic modelling deemed them at best as of minor use.
The floating booms designed to guide fish rather than the flow also influenced the velocity fields, and particularly, the solid boom combined with changing the gates from the northern to the southern created a distinct flow away from the intake and towards the gate.
However, this solution is challenged by strong vertical flow under the boom that the fish is likely to follow (they do so at the intake with a wall extending even deeper), and it is unlikely that such a floating device could actually withstand the drag forces and maintain position and function during flooding conditions, typically appearing during smolt migration.
The rack type permeable boom solution analysed appears as the most promising solution. It was modelled as a 75 m long rack with horizontal bars and angled 30°relative to the bank and the flow direction.
Because the surface water (1 m) flow through the rack, only small downwards velocities emerged for the fish to follow. Because the fish cannot pass through the gaps and smolts generally avoid both the structures and the resulting turbulence (Enders et al., 2012;Nestler et al., 2008;Williams et al., 2012), they may migrate along the angled boom to reach the southern flow towards the gates. It has been shown that fish guidance systems such as fish-friendly trash-racks or trash-booms placed upstream of the intake are viable solutions to guide fish away from HPP intakes (Albayrak et al., 2018;Boes et al., 2016;Calles et al., 2013;de Bie, Peirson, & Kemp, 2018;Nestler et al., 2008;Tomanova et al., 2018). In particular, Nyqvist et al. (2017), who analysed migration data from a very similar angled rack with horizontal bars, documented high guiding effectively for salmon towards the bypass channel at a HPP dam and intake facility in Sweden but with a full depth rack.

| CONCLUSION
The present study illustrated the value of using river reach CFD modelling as a tool for early evaluation of different mitigation measures to prevent fish entering water intakes to hydropower turbines or other installations and particularly so when combined with high resolution positional telemetry. Further analyses of migration tracks and work towards more detailed and general fish migration models may further improve the value of the approach. Modern measurement instrument (such as ADCP) provide rapid and cost-effective mapping and commercially available software and faster computers now allow efficient CFD modelling at river reach scales. The present modelling was rather extensive (involving the use of high performance computer) because the grid was designed for research purposes, but more coarse grids and modelling on ordinary PCs should be sufficient for more applied purposes. Obviously, the cost of such modelling is far lower than actual construction of measures and assessment of its effects, particularly if the mitigation measures fails and must be revised. We advocate that fish migration challenges at hydroelectric facilities should be explored in advance by a similar approach to provide wellfunded engineering solutions for effective fish passage. Although the actual effect of the chosen mitigation measure must be evaluated after implementation, the number and cost of trials are expected to be strongly reduced.