Individual‐based modelling of hydropeaking effects on brown trout and Atlantic salmon in a regulated river

We developed an individual‐based model (IBM) to understand the effects of hydropeaking on growth, survival and distribution of age 0+ to 1+ juveniles for high‐conservation value populations of native brown trout (Salmo trutta) and Atlantic salmon (S. salar) in river Gullspång, Sweden. We parameterized and applied inSTREAM (7.2‐SD) and calibrated the model by comparing predicted versus observed growth under the current hydropeaking regime (n=>1,200 model fish for 365 days). Our objective was to model growth, survival and distribution under flow scenarios with and without hydropeaking. We observed that hydropeaking generally resulted in modest (~10%) negative effects on growth and survival of both species. Survival was more affected than was growth, smaller fish more affected than larger fish. On‐peak (high) hydropeaking flows resulted in less profitable feeding conditions (less growth) and higher predation (lower survival). Thus, inSTREAM 7.2‐SD appears to capture ecologically‐relevant behavioral patterns under hydropeaking, for example, habitat selection, in response to rapid flow changes. Understanding such patterns for large rivers via manipulative field studies, even if possible, would be time‐consuming and costly. Our study demonstrates the potential of IBMs as powerful tools for testing research questions and assessing and prioritizing alternative management strategies in regulated rivers.


| INTRODUCTION
Habitat degradation and fragmentation caused by hydroelectric dams and flow regulation is a leading driver of population declines in freshwater systems (Tickner et al., 2020;WWF, 2018). Indeed, freshwater biodiversity is declining rapidly on all continents and in every major river basin on earth, with losses occurring far faster than in terrestrial and marine ecosystems (Albert et al., 2020). At the same time, demand for fossil-free hydroelectric power continues to grow, and many hydroelectric power plants are operated with sub-daily changes in flow (i.e., hydropeaking) to follow daily cycles in energy demand.
There is a great need, therefore, to improve river management to protect and restore freshwater biodiversity and ecosystem services (Tickner et al., 2020), while continuing to provide clean energy to end society's dependence on fossil fuels. In this article, we demonstrate how individual-based models (IBMs) can be an effective tool for understanding how fish populations may respond to river management strategies such as hydropeaking.
The general paucity of long-term, comprehensive field studies on fish behavior and consequent population-level responses under hydropeaking, combined with the variable results of these studies, suggests that ecological modelling approaches might be fruitful for developing a better mechanistic understanding of hydropeaking effects. Advanced mechanistic IBMs have received practically no attention for assessing the effects of hydropeaking on the growth, survival and distribution of individual fish. Our study pioneers the IBM approach by using a model that explicitly represents the full circadian cycle to understand the effects of hydropeaking on salmonid populations. By simulating the full daily light cycle, IBMs are capable of modelling how individual fish respond at any time during a day to substantial changes in flows caused by hydropeaking flow regime. In addition, they can model how fish behaviour (e.g., feeding or hiding) and habitat use depend on light and its interaction with flow.
The theoretical underpinning for stream fish IBMs has been developed and tested over the past 30 years (Hughes & Dill, 1990;Piccolo, Frank, & Hayes, 2014;Railsback, Lamberson, Harvey, & Duffy, 1999). IBMs for salmonid populations (e.g., Railsback, Harvey, Jackson, & Lamberson, 2009;Van Winkle et al., 1998) rely on the bioenergetics-based net rate of energy intake theory and have the advantage of linking physical stream habitat with realistic measures of fishes' ecophysiology and ecology, which make these models robust for assessing river habitat quality and production potential (Railsback, 2016). By considering the important temporal and spatial processes through which river regulation affects fish populations, for example, predation, foraging and reproduction, IBMs capture dynamic responses to river management scenarios such as changes in the flow regime (Anderson et al., 2006;Railsback et al., 2009). Developing IBMs for species of high conservation value to understand their population responses will be valuable when planning management scenarios to mitigate the effects of habitat alteration caused by river regulation. Model outputs could help river managers to prioritize management decisions and biodiversity conservation actions.
InSTREAM 7 is the most recent of a family of individual-based salmonid models that have been in development since 1998 (Railsback, Ayll on, & Harvey, 2021). Since its initial release (Railsback & Harvey, 2001), inSTREAM's capabilities have been steadily developing. The model simulates the fitness-seeking process in which fish select habitat, using a trade-off between food intake and mortality risks to maximize their probability of surviving and reproducing. Mortality risks are functions of habitat and fish variables. Fish growth depends on factors such as prey availability and hydraulic conditions and is modeled as proportional to net energy intake rate, the difference between the energy intake from feeding and metabolic costs (Piccolo & Watz, 2018). One of the most important among the unique characteristics of inSTREAM as an assessment tool is that it considers such individual-level processes as feeding and energetics, competition and predation risk to make population-level predictions.
InSTREAM's ability to assess the effects of hydropeaking operations was first verified by the application of an earlier sub-daily version of the model, which simplified the circadian cycle to only two phases: day and night, to assess problems for fish related to flow fluctuation (Hayse, LaGory, & Railsback, 2006). InSTREAM has also proven useful for designing and evaluating restoration projects and for prioritizing alternative management actions (e.g., Bjørnås, Railsback, Calles, & Piccolo, 2021;Railsback, Gard, Harvey, White, & Zimmerman, 2013). Because of its mechanistic nature, inSTREAM is transferable for different species and geographical applications. The parameter values should be reconsidered for each application of inSTREAM. In addition, it is easily extended, and many previous applications of inSTREAM were developed by modifying the model to address various species-specific ecological research questions (e.g., Ayll on et al., 2016;Ayll on, Nicola, Elvira, & Almod ovar, 2021;Hajiesmaeili, 2019;Harvey & Railsback, 2012).
In this study, we made the first application of inSTREAM 7.2-SD (also referred to as inSTREAM 7-SD) to assess the effects of different flow scenarios on populations of sympatric brown trout (Salmo trutta) and Atlantic salmon (S. salar), hereafter trout and salmon, in the lower Gullspång River, Sweden. We simulated the growth, survival and distribution of age 0+ to 1+ trout and salmon under various hydropeaking and steady-flow scenarios, as well as under a natural flow regime.
Our goal with this study was not to give direct management advice, but rather to demonstrate the development and application of an IBM in a large regulated river under hydropeaking flows. We illustrate the potential value of the IBM by presenting an example application, which demonstrates that the model is capable of comparing the effects of alternative peaking levels and flow regimes. Such comparisons may be useful for evaluating management scenarios for various hydropower projects. Assessing such population-level responses for entire cohorts of fish under multiple flow scenarios would be prohibitive; IBMs may be powerful tools for assessing and prioritizing research questions and alternative management strategies such as different flow scenarios in regulated rivers. The methods used in this study would also be applicable to other regulated rivers to address spatially-explicit research questions. We, therefore, anticipate that our study will have broad applications for the sustainable management of regulated rivers. (c) behavior, including the selection of habitat cells and three activities: drift feeding, search feeding, and hiding. Model outputs can be spatially and temporally explicit, that is, fish distributions can be output at any given time step.
Here, we summarize the most important and relevant characteristics of inSTREAM 7-SD to the simulation experiments of this study.
The model we used, and complete documentation, are available at: https://ecomodel.humboldt.edu.
The main purpose of inSTREAM 7-SD ("SD" referring to the sub-daily flow fluctuations) is to model how individual trout respond to substantial sub-daily changes in stream flow such as those from peaking hydropower operations, and the resulting effects on measures such as fish growth and abundance. The standard version of inSTREAM 7 simulates four" & "time steps per day (dawn, day, dusk, and night) to represent the daily light cycle.
InSTREAM 7-SD differs from previous versions in its ability to represent how trout are affected by flow changes that occur within one of the four daily time steps. On each time step, each fish selects a habitat cell and activity (feeding vs. hiding) with the objective of maximizing their expected future growth and survival.
Habitat and activity selection is modelled throughout the entire circadian cycle in inSTREAM 7-SD, by starting new time steps at any time a substantial change in flow occurs (thereby modelling more than four" & "time steps per day). Time steps are triggered when the flow read from an input file changes by a user-defined threshold (typical value = 10%-20%). This feature gives its simulated fish greater ability to adapt to changes in habitat and biological conditions than in previous versions of the model (Railsback, Harvey, & Ayll on, 2021b). This approach has been tested thoroughly and shown to reproduce a variety of observed patterns in salmonid behavior (Railsback, Harvey, & Ayll on, 2020).
Detailed documentation of the model's formulation is provided in the inSTREAM 7 user manual (Railsback, Harvey, & Ayll on, 2021b).

| Study site and model inputs
The regulated Gullspång River (58 59 0 14.0''N, 14 06 0 41.4 00 E) runs from Lake Skagern to Lake Vänern, Sweden, and is home to native landlocked populations of migratory brown trout and Atlantic salmon of high conservation value (Piccolo, Norrgård, Greenberg, Schmitz, & Bergman, 2012). Spawning and rearing are limited to three river sections: the Gullspång Rapids (A: Gullspångsforsen. GF), and the Årås Rapids (B: Lilla and C: Stora Åråsforsen); (Figure 1). An inSTREAM model was recently completed for the non-hydropeaking section of GF (Bjørnås et al., 2021). Our focus area in this study is Lilla Åråsforsen (LÅ). Between 20 August and 19 April, the Gullspång hydroelectric power (HEP) station is allowed to use a hydropeaking flow regime, which varies between 9 (minimum flow) and 230 m 3 /s (maximum capacity). During high discharges, a diversion weir located upstream of LÅ damps the high flow peaks, allowing a maximum of 80 m 3 /s to follow the natural river course into the Årås Rapids ( Figure 1). We developed ten flow scenarios for this river reach to assess the effects of hydropeaking on fish growth, survival and distribution. The purpose of including steady and natural flow scenarios was to provide management-relevant baselines for comparisons with the hydropeaking flow scenarios.
We first mapped the riverbed geometry and topography of LÅ River reach. We created hexagonal habitat cells for inSTREAM in InSTREAM simulates how temperature affects fish both directly by influencing the mortality of fish or eggs at extreme temperatures and indirectly by changes in behavior and bioenergetics. Although inSTREAM represents how turbidity decreases the ability of fish to capture food and makes the fish less vulnerable to predators, timeseries for turbidity data are not available for river Gullspång; spot checks throughout the year demonstrated that turbidity is generally very low so we ignored turbidity in our application. To avoid an unrealistic lack of light attenuation in the model, we assumed turbidity was constant at 2 NTU.
Initial fish population characteristics (Table 1) were estimated using data from the last 10 years from the open Swedish National Electrofishing Register (Svenskt ElfiskeRegiSter, SERS, 2021). We introduced an initial population of age 0+ trout and salmon based on observed mean lengths and densities of fish from SERS. Observations (unpublished) of spawning demonstrate that trout spawn on average 1-2 weeks earlier than salmon. Thus, 0+ trout emerge earlier in the spring and maintain a larger average size than salmon during the first year. In our inSTREAM application, morphological interspecific differences between trout and salmon leading to different swimming abilities and foraging niches were not considered.

| inSTREAM simulations and analysis methods
All simulation experiments were based on 1-year model runs, starting on September 1, with five replicate simulations for each flow experiment. The response variables were model-simulated final mean length and abundance of trout and salmon (starting as age 0+ and ending as age 1+), as well as fish distribution which is only simulated and not observed. Both species were included together in each model run. In considering the effects of flow regime on the response variables, we focused on the trends of the responses, positive or negative, rather than on the absolute values; results trends are generally more robust than absolute values when applying IBMs to various potential management scenarios.

| Model calibration
Sensitivity analyses of earlier inSTREAM applications have shown that outputs are sensitive to food availability and predation parameters, making these parameters suitable for calibrating the model (e.g., Railsback et al., 2009). Therefore, we adjusted these parameters to reproduce observed length values ( Table 1) that we obtained from the SERS database and reasonable survival rates. Calibration was done by varying the parameter values for the baseline flow scenario (see Flow scenarios below). Our calibration process had two steps: (1) we varied food availability (i.e., drift food concentration and search food production) and compared the final mean lengths of modelled fish to observed fish. We selected the parameter values that produced the lowest differences between model results and (observed) electrofishing data. Further, (2) we varied the parameters that adjust fish predation risk (aquatic predators, such as pike Esox lucius, are common in river Gullspång) to balance mortality rates from terrestrial and fish predation. We assumed that only non-salmonid fish are predators of salmon and trout in the Gullspång River because there are few or no resident adult trout or salmon in these migratory populations. In this step, because of the lack of survival data for age 0+ trout and salmon, the calibrated model used the smallest differences between model results versus the target range estimated by Gibson (1993) for age 0 to age 1 survival.

| Flow scenarios
We defined a baseline scenario for the simulation experiments based on the current HEP operating protocol for our case study river. We

| Parameter uncertainty analysis
We conducted a parameter uncertainty analysis of the model to investigate how robust management recommendations based on it are to different values of the three parameters: (a) food availability, and survival from (b) fish and (c) terrestrial predation (Railsback, Harvey, & Ayll on, 2021a).
Each parameter was set to three values below, at, and above its standard calibrated value. Values of the food availability parameter were set to 50%, 100%, and 150% of its standard value, parameters related to fish and terrestrial predation were each varied in steps of 1%.
We  Mortality outputs for both flow 9 and 21 m 3 /s scenarios indicated that hydropeaking flow regime will increase fish predation, as well as the risk of poor condition which represents mortality due to starvation compared to steady flows, leading to reduced survival ( Figure 6).
Sub-daily flow fluctuations resulting from hydropeaking flow regime also affected the distribution of fish in the river (Figure 7). For both species, individuals packed more tightly and moved more during the peak at the baseline hydropeaking scenario, compared to the natural flow regime. Although salmon were more abundant than trout, there were more trout in deeper water.

| Daily light cycle
Most individuals feed during dusk when feeding was more efficient than at night and safer than during the day (Figure 8). For both steady F I G U R E 4 Results of the model runs (n = 5) for the effects of different flow scenarios on (a) brown trout and (b) Atlantic salmon final predicted mean length after 1 year. Initial modelled mean length (±SD) of trout and salmon was 11.33 ± 0.13 and 9.81 ± 0.05 cm, respectively. White background indicates hydropeaking scenarios, light grey a natural flow regime, and dark grey steady flow scenarios and hydropeaking flows, there was more feeding during daytime and less at night when the minimum flow was 21 versus 9 m 3 /s. Both species exhibited similar trends in their diel feeding activity in response to flow changes, although salmon fed overall more frequently than trout, but not during the day when this pattern was reversed.

| Parameter uncertainty
The rankings of flow scenarios by predicted trout and salmon length varied little among the 27 parameter combinations (Figure 9, left panels). There was more variation in rankings by survival (Figure 9, right panels), which reflects that differences in survival among flow scenarios were smaller and more stochastic than differences in length (Figures 4 and 5).

| DISCUSSION
In this study, we employed the inSTREAM 7.  Drift feeding was modelled using an approach similar to other drift-feeding models (e.g., Naman et al., 2020). Fish in inSTREAM are assumed to capture the food items that pass within a reaction distance that increases with fish length but decreases with water velocity, whereas the amount of food passing within the reaction distance increases with velocity and drift concentration. Therefore, higher flows deliver more food and resulted in higher growth. In addition, drift-food intake peaks at an optimal velocity that is higher for larger fish and lower when there is less light. Flow can also affect survival by creating feeding places where fish can feed productively and safely, and, therefore, reduce their risk of predation without increasing the risk of starvation. Increasing the minimum discharge of hydropeaking and magnitude of steady flows appeared to provide less habitat for successful night feeding, which is not surprising because simulated growth is higher in lower velocities at night.
In river Gullspång trout spawn earlier than salmon, trout fry emerge earlier in the spring, and trout maintain a size advantage during the first year of growth. The different responses of trout vs. salmon to alternative minimum flow releases for hydropeaking and steady flow scenarios in our models were mainly driven by differences in fish size, and partly related to the geomorphology of the river. Predicted survival for trout peaked at higher flows than did survival for salmon, because increasing flow resulted in increased depths and velocities, reducing the risk of terrestrial predation. Increased depths and velocities are preferentially beneficial for larger fish because of their better swimming and foraging abilities (Railsback et al., 2009).
Fish size probably plays a relatively larger role in foraging success in deeper and faster water than do interspecific differences between salmonid species (Piccolo, Hughes, & Bryant, 2007, 2008. Incorporating differences in ecophysiological parameters between salmon and trout (e.g., swimming performance, metabolism and aggressive behavior), in The exceptional positive effect of hydropeaking flow regime on trout growth compared to the steady flows at flows of 9 and 21 m 3 /s was likely caused by length-related differences in foraging success between the two species. Trout probably dominated salmon due to their larger size and therefore occupied the more preferred habitat cells with profitable feeding sites and gained a competitive advantage over salmon from hydropeaking flow scenarios. This mechanism can also explain the reason that trout survival rate was generally higher than salmon, that is, they were larger in size and selected safer habitats with less predation. In addition, inSTREAM assumes fish adapt quickly and without energy cost to flow changes, which may exaggerate the ability of real fish to adapt. Both negative and positive effects of hydropeaking compared to steady flow regime predicted in our study were also found in other studies (e.g., Korman & Campana, 2009;Puffer et al., 2015;Puffer et al., 2017).
Perhaps the most important mechanism in inSTREAM that relate growth to survival is the choice between feeding and hiding. Better feeding conditions allow fish to feed in safer habitats and at times with little predation risk (low light intensity) and to spend more time hiding instead of feeding. Our results also showed that whereas the increases in flow from 9 to 21 m 3 /s produced more growth in each of steady and hydropeaking flow scenarios, both trout and salmon reduced feeding with increasing flow and therefore experienced less predation and higher survival. Therefore, these results can represent overall feeding and growth conditions at the study site. For example, at higher flow (21 vs. 9 m 3 /s) the fraction of fish feeding during day increased, while the fraction feeding at night decreased. This transition indicates that higher minimum flow reduced the availability of habitat where feeding is profitable at night. Therefore, they fed more during the day, when drift feeding is more efficient because prey is more visible (Watz, Piccolo, Bergman, & Greenberg, 2014). Our results also showed generally higher growth at flow 21 versus 9 m 3 /s.
In shallow areas, fish are likely visible to predators during the day.
In the Steady-Min flow scenario, there was much shallow habitat. Therefore, fish avoided daytime feeding in this scenario to avoid predation, and they mainly feed at night and during the crepuscular phases. In general, feeding was more frequent for salmon than trout, presumably because salmon were smaller in size and usually individuals with higher fitness motivation for energy acquisition tend to eat more to meet their metabolic demands.
Harvey and White (2016)   Such modelling studies have the potential to prioritize field studies to make the most efficient use of limited resources to collect field data that will best guide management decisions.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.