Development of behavioral rules for upstream orientation of fish in confined space.

Improving the effectiveness of fishways requires a better understanding of fish behavior near hydraulic structures, especially of upstream orientation. One of the most promising approaches to this problem is the use of model behavioral rules. We developed a three-dimensional individual-based model based on observed brown trout (Salmo trutta fario) movement in a laboratory flume and tested it against two hydraulically different flume setups. We used the model to examine which of five behavioral rule versions would best explain upstream trout orientation. The versions differed in the stimulus for swim angle selection. The baseline stimulus was positive rheotaxis with a random component. It was supplemented by attraction towards either lower velocity magnitude, constant turbulence kinetic energy, increased flow acceleration, or shorter wall distance. We found that the baseline stimulus version already explained large parts of the observed behavior. Mixed results for velocity magnitude, turbulence kinetic energy, and flow acceleration indicated that the brown trout did not orient primarily by means of these flow features. The wall distance version produced significantly improved results, suggesting that wall distance was the dominant orientation stimulus for brown trout in our hydraulic conditions. The absolute root mean square error (RMSE) was small for the best parameter set (RMSE = 9 for setup 1, RMSE = 6 for setup 2). Our best explanation for these results is dominance of the visual sense favored by absence of challenging hydraulic stimuli. We conclude that under similar conditions (moderate flow and visible walls), wall distance could be a relevant stimulus in confined space, particularly for fishway studies and design in IBMs, laboratory, and the field.

Introduction

Worldwide, fishways play a key role in efforts to restore upstream-directed fish migration at dams. The widely differing effectiveness of existing projects [1, 2] illustrates that their design is still challenging, especially for species other than Pacific salmon [3]. Improving fishway effectiveness for all species requires a better understanding of fish behavior near hydraulic structures, especially of orientation and navigation [4, 5]. One of the most promising approaches to better understand such behavior and “a high research priority” [5] is the development of behavioral rules. Behavioral rules are the logical and/or algebraical formulation of behavioral hypotheses. They can be implemented and tested in individual-based models (IBMs, [6]).

IBMs enable one to implement the concepts of orientation (directional response to local conditions, [4]) and navigation (movement towards a goal outside the local sensory environment, [4]) rather accurately. For example, in a number of recent fish movement IBMs, orientation is the outcome of behavioral rules using local stimuli, while navigation emerges from the sum of orientation decisions over time (e.g., [7-11]). Together with their ability to integrate CFD (computational fluid dynamics) model results, IBMs are therefore highly suitable to support fishway design.

A preferred object of research for fishway design are hydraulic stimuli, because they can be manipulated to optimize passage efficiency. The most widely used stimulus in fish orientation IBMs, regardless of whether the model considers upstream or downstream fish migration, is velocity direction [8, 11–14]. It is a fundamental stimulus, as it informs positive and negative rheotaxis, which are crucial for guiding upstream and downstream navigation, respectively [15-17]. Velocity magnitude is also commonly considered [8, 10–14] and especially important for upstream migration, since the energy demand for swimming against the flow is proportional to the cube of relative fish velocity [18].

Apart from velocity direction and magnitude, stimulus selection to formulate behavioral rules remains difficult [8]. To integrate information on turbulence into IBMs, the turbulence kinetic energy (TKE) is a common choice [12, 13]. Turbulence may also be integrated as velocity fluctuation [11]. Other approaches to capture hydraulic information in IBMs include using spatial acceleration [8-10] and the related concepts of spatial velocity gradient and strain rate [7, 13].

Fish studies from laboratory flumes offer further potential stimuli for modelling upstream orientation, for example the horizontal Reynolds shear stress (τ), turbulence intensity, and eddy orientation and scale (see [19] for a framework). However, results are manifold and could depend on the specific setup and/or species tested. In different studies, dace preferred adapted turbulence levels (TKE and τ) [20], adult Iberian barbels primarily avoided areas of high τ [21], and brown trout preferred low drag areas [22]. Eddy orientation and scale did influence Iberian barbel behavior [21], but are still difficult to define in common vertical-slot fishways [23] and are not used in IBMs to date. As for IBMs, the selection of hydraulic stimuli for laboratory studies remains controversial [22].

Non-hydraulic external stimuli for fish orientation, such as visual, olfactory, acoustic, thermal, and magnetic stimuli, are rarely investigated in flows that are comparable to a fishway flow and the authors are not aware of corresponding fish orientation IBMs that include such stimuli. This is particularly notable for vision, as it is considered important for orientation [24, 25]. Published laboratory and field studies on upstream orientation using visual stimuli as explanatory variable are limited to general phototaxis, i.e. attraction or repulsion [16, 26].

Basic motivation to swim against the flow is crucial for successful upstream migration. However, such motivation may vary widely between species and even between individuals [3]. Existing upstream orientation IBMs avoid modeling motivation as an internal state by determining that fish constantly try to move upstream [9, 11–14]. However, this approach prevents the potentially important ability to drift, recover, and repeat passage attempts after initial failure [27]. Modeling variable internal states to elicit differing behavioral responses to identical situations could be one way to approach this problem.

Furthermore, existing IBMs for upstream migration still await quantitative validation. They are either tested qualitatively, e.g. using visual track comparisons, or untested due to a lack of observation data. This is a remarkable contrast to the comprehensive quantitative testing available for downstream migration IBMs [7, 8, 10].

In the present study, we thus developed a new IBM for fish upstream orientation and navigation in confined space to support fishway design. It was tested against detailed data of brown trout movement in a laboratory flume, and is the first such model enabling holding behavior, recovery, and thus repeated passage attempts. Five external stimuli were selected for testing based on the literature:

positive rheotaxis as a fundamental behavior in upstream migrating fish,

velocity magnitude as being fundamentally related to energy demand,

TKE as one of the most common and general turbulence variables,

spatial acceleration as having been tested against the largest data sets by far,

wall distance as a potential proxy for visual orientation.

The present paper describes how we (1) defined movement patterns from brown trout flume data, (2) developed and tested a CFD model to generate stimuli data, (3) developed and tested a behavioral model to reproduce the movement patterns, and (4) examined which of the five external stimuli would best explain upstream brown trout orientation and navigation in confined space.

Methods

Flume setups

For developing and testing the behavioral model, we used unpublished brown trout movement data from a previous, unrelated study (wild Salmo trutta fario, n = 66, mean body length ± standard deviation (SD) = 0.27±0.04 m, Schütz et al. [28]). Fish were caught in a close-by river during their spawning period in October/November, when an increased motivation to migrate can be expected [28]. The study was conducted in a wide indoor flume (Fig 1). The external walls and bottom consisted of smooth acryl glass and metal and the internal walls consisted of coated wood. Water temperature was 16.2–18.2°C. Water was clear and evenly illuminated. All care and procedures involving handling and holding fish were conducted as stated and permitted by the district government Karlsruhe (license AZ 35–9185.82/A-6/16).

Fig 1

Plan view of flume setup 1.

Total length and width = 16.38 m x 2.50 m, water depth = 0.60 m. Flow direction was from left to right. The slot was removed in setup 2 to alter the flow field. Transparent gray areas mark left and right zones (in flow direction) for pattern P1 (defined below, section “Movement patterns”). A and D mark control lines to filter inactive fish and to define pattern P5.

We used two hydraulically different setups from Schütz et al. [28] to test the application range of our model. Setup 1 included a jet created by a slot; setup 2 had no slot and consequently, no jet. The total discharge of Q = 1.00 m3/s (used in both setups) produced a jet velocity of U = 1.5 m/s, which is typical for slots of large multi-species fishways in Germany. The flume was wide enough to enable distinct lateral movement, but narrow enough to influence behavior through the walls and screen.

Fish were released unmarked in batches of three at the downstream end (staging area in Fig 1) and were used only once to avoid learning effects. Fish positions were manually recorded in real time independently by two biologists upon notable change. A change of position was defined as either:

crossing a meter line of the flume in longitudinal direction

moving between three zones in lateral direction: close to left wall, close to right wall, far from wall (close meaning max. ~ 0.25 m)

moving between three zones in vertical direction (close to bottom, close to surface, in between (close meaning max. ~ 0.15 m)

Moreover, it was recorded if fish swam alone or in groups of two or three. Records were attributed to the current test minute. During postprocessing, multiple position changes within the same minute were distributed evenly over the seconds (S1 and S2 Appendices). Data were verified qualitatively using video records.

To ensure that all fish were actually motivated to move during the experiment, only fish that crossed line A (Fig 1) within 30 min after removing the staging mesh were considered valid. A trial was finished either after all three fish passed line D or 60 min after the first crossing of line A. Data points following a passage of line D were excluded. After filtering, we obtained n = 25 tracks for setup 1 and n = 24 tracks for setup 2.

Movement patterns

We processed the trout track data using patterns in the sense of pattern-oriented modeling [29] using MATLAB R2018b. We defined five patterns, P1-P5, to capture the most striking spatial behaviors observed during the original experiment. They included a preference for wall and bottom proximity as well as frequently swimming back and forth (turns):

P1, lateral distribution

The flume was divided into a left, middle, and right zone (looking downstream, Fig 1). The dividing lines were located at a lateral distance of Δy = 0.25 m (10% flume width) to one of either side walls or the screen. Time spent by single fish (i.e., center of mass) was summed for each zone. To account for the differences in total track duration [see 30], resulting time sums per fish and zone, t, were divided by the particular fish’s track duration, t, to get relative track fractions, s. Finally, s was averaged by the fish count, n, to get: where i was a fish iterator. The resulting three lateral zone averages constituted pattern P1. As it is the most direct outcome of upstream orientation, matching it was a major goal of model development.

P2, vertical distribution

Vertically, the flume was divided into a surface, middle, and bottom zone. The dividing lines were located at a vertical distance of Δz = 0.15 m (25% water depth) to either the water surface or flume bottom, respectively. Tracks were averaged using Eq 1 to obtain three vertical zone average values. Due to the shallow water, we expected P2 to be of less significance for orientation.

P3, turn zone and P4, few turns

Turns were defined as changes in the longitudinal movement direction, x, leading to a displacement Δx ≥ 2 body lengths (BL) before the next direction change. About half of the fish in both data sets performed 4 or less turns. To avoid bias, the data set was divided into turning (> 4 turns, P3, n) and few-/no-turning fish (≤ 4 turns, P4, n). For computing P3, the flume was divided longitudinally into three zones of equal length, in which the relative turn shares were computed and averaged by n analog to Eq 1. P4 was defined as n/n.

P5, arrival rate

The count of fish navigating the total length of the experimental flume area and crossing line D, divided by the count of fish crossing line A, formed P5: n/n ∙ 100 [%].

For use in discussion, we also calculated the lateral distribution of fish at the start position and the proportion of track time fish spent without other fish nearby (i.e., not shoaling).

Computational fluid dynamics model

A CFD model was employed to calculate the velocity, TKE, and acceleration fields of both setups. We applied the free-surface solver interFoam of OpenFOAM 2.3.1 [31] to solve the 3D incompressible URANS (unsteady Reynolds-averaged Navier-Stokes) equations.

A hexahedron-dominant unstructured mesh was generated using snappyHexMesh. Base resolution was uniform Δ = 5 cm, with local refinement to Δ = 1.25 cm around the slot and screen posts (Fig 1). Horizontal screen bars were omitted, as their influence on the velocity and turbulence field was negligible with respect to behavioral data accuracy [32]. Total cell count for setup 1 was 848,694.

Inlet flow rates and outlet fixed water level were set to the laboratory values. For turbulence closure, we chose the k-ω-SST model. At the inlet, TKE (≙ k) was set to k = 0.001 m2/s2 and the specific rate of TKE dissipation was set to ω = 1 Hz. No-slip conditions (velocity vector U = (0,0,0)) and a small sand-equivalent roughness coefficient (k = 1·10-5 m) were set at the walls and bottom.

The fields of velocity U and water/air distribution α were initialized with a largely converged solution and ran for 80 s of simulated time. U was averaged over the final 20 s to obtain the steady flow velocity vector field can be expressed either as three-component vector (u,v,w) or as magnitude U along with its horizontal and vertical angles γ and β.

TKE is defined as half the sum of the velocity components’ variances (squared turbulence intensities): using Reynolds decomposition, .

Spatial acceleration a is caused by changes in the flume cross-section and by friction, in contrast to temporal acceleration caused by transient processes. Spatial acceleration is the product of velocity and velocity gradient (which comprises strain rate and rotation rate). In 3D Cartesian coordinates, it reads where the superscript denotes transposition from column to row vector. Its magnitude |a|, U, and TKE were used as hydraulic stimuli variables in the behavioral model.

Behavioral model description

A complete, detailed model description, following the ODD (overview, design concepts, details) protocol [33, 34] can be found in S4 Appendix. Here, we provide a summary. The overall purpose of our model was to find behavioral rules for fish that could enable predictions of fishway attraction and passage efficiency. Specifically, we addressed the following question: How well do different external stimuli explain orientation and navigation of upstream moving brown trout?

The entities of the model were fish and mesh cells. Fish were represented as mesh-independent, spatially explicit points in 3D space. Their key state variables were position, motivation, and fatigue. Fish movement was modeled kinematically, i.e. there was no influence of fish on the flow field. Key state variables of the mesh cells were the cell label number and the hydraulic variables |a|, U, TKE, and α. Spatially, the behavioral model covered the fish-accessible areas of the CFD model (i.e. without the air phase and without areas upstream of the screen). The temporal extent was 60 min of clock time, equal to the maximum laboratory trial duration, divided into constant time steps of Δt = 0.5 s.

Each model run used n = 100 fish of BL = 0.27 m (as measured), each being assigned to either a “fast” or “slow” category. In this way we tried to model individual traits which could influence migratory tendencies, such as being bold or shy [35] or being physically strong or weak. Model fish were positioned only in the left and right zones of the staging area (Fig 1) as per observations. Their exact distribution was controlled by a parameter.

Every time step, the model computed a new position for every fish. First, a sensory ovoid consisting of one point in the fish center and six surrounding points was determined [7]. The surrounding points were placed at an imaginary skin/water interface, approximately where real fish detect water motions by their lateral line system [36]. Hydraulic variables were interpolated to these points. Wall distances were computed towards the front and sides.

Next, we aimed to summarize all fish characteristics possibly acting on longitudinal movement by two opposing variables: motivation (to swim upstream), M, and fatigue, F (both range 0.0–1.0). The variables resembled the typical internal states “need” and “cost” [4]. Their balance determined the following movement decision. We chose this rather basic representation of internal stimuli because our focus was on orientation and external stimuli. Still, it enabled model fish to drift, recover, and swim upstream again, a behavior which we consider crucial for modeling fish migration through fishways [27]. Its success was measured foremost by patterns P3 and P4.

M was increased proportionally with time when the model fish made no spatial progress, presuming there was a principle motivation to move upstream. The maximum increase rate was reached after k = 20 s for “fast” fish and k = 130 s for “slow” fish (default values, varied later). F was computed proportionally to local flow velocity and swim speed. It reached the maximum value at a burst speed of k = 25 BL/s [37]. We note that more realistic speed-fatigue relations are available [37, 38], but they were not required for our purpose of modeling orientation. From the balance of time-averaged M and F, one of three horizontal behavioral rules was selected (Fig 2):

Fig 2

The process of computing the swim vector.

Based on the balance between motivation and fatigue, a horizontal behavioral rule is selected. Migrating is the key behavioral rule for which five versions are contrasted. Holding means maintaining the position. Drifting includes swimming against, but slower than the flow. In combination with the vertical behavioral rule, the final swim vector is determined.

migrating (moving upstream),

holding (maintaining position),

drifting (moving downstream with limited random deviation from the flow vector).

Migrating was the key behavioral rule. In its baseline version it only depended on the velocity direction. It set the horizontal and vertical swim angles against the velocity direction, as we judged rheotaxis to be a fundamental behavior in fish that is always active (e.g. in Mexican tetra [16]). To also capture smaller, unpredictable behavior variations, a limited random angle was added to or subtracted from the horizontal swim angle. Migrating was expected to have the most direct effect on pattern P1.

Our guiding question (“how well do external stimuli explain upstream orientation?”) was addressed by adding four different stimuli for orientation to the baseline migrating behavioral rule. That is, the horizontal swim angle could be adjusted in either one of five alternative stimulus versions:

Baseline (rheotaxis): Against velocity direction & towards random side by a limited random angle as described above.

Velocity: Baseline & towards side of lower velocity magnitude. Rationale: Avoiding zones of high velocity to save energy (low-velocity seeking) is a common strategy for fish and was observed e.g. in laboratory flumes for longnose dace [39] and brown trout [22] and modeled in an IBM for carp and sturgeon [11].

TKE: Baseline & towards front or side with the smallest TKE difference towards the fish center (constant turbulence level). Rationale: As it is still unclear how turbulence affects orientation in a specific situation or for a single species and fish size, we chose one of the most common and general 3D turbulence variables, containing the turbulence intensity. Orientation along a constant, adapted turbulence level was observed before in dace [20] and modeled in an IBM [12].

Spatial acceleration: Baseline & towards side of higher |a| magnitude. Rationale: Acceleration is the hydraulic stimulus tested against the largest data sets by far, although for downstream migration [8, 10]. As it also contains the strain rate [13] within the velocity gradient tensor, it is an information-rich representation of the flow field and we consider it potentially useful also for upstream migration [e.g., 9]. We chose attraction towards higher acceleration to mimic our laboratory results.

Wall distance: Baseline & towards side of shorter wall distance. Rationale: Within the confined space of our model, wall distance can serve as a potential proxy for the little investigated, but potentially important [24, 25] visual orientation. Again, we chose attraction towards shorter wall distance to mimic our laboratory observations.

An independent vertical behavioral rule was always active. It limited vertical movement using an elevation difference [7, 8] and was primarily gauged by pattern P2.

The output of the process described was a 3D swim vector for each fish and time step. The new position was determined from this swim vector, the flow vector at the fish’s position, and the time step width. Overall navigational success was measured by pattern P5.

Code and speed

The code of the behavior model and its software framework can be found in S5 Appendix. The software framework was based on the Fortran 90 code for downstream migrating smolts used in Goodwin et al. [8]. It accounted for computer simulation dynamics such as variable storage, sensory point creation, vector transformation from/to Cartesian coordinates, and pseudo-random number creation. We modified it to work with unstructured, polyhedral meshes in OpenFOAM 4.1 using C++. Computation time varied depending on how fast fish exited the domain. A typical runtime for a single model run with n = 100 fish was about 85 s, using 8.1 GB RAM for up to 7200 time steps on one core of an Intel E5-2660-v3 (“Haswell”) CPU.

Model function test

Besides behavioral rules, parameters (i.e., variables which do not change during a model run, but can change between model runs) are the second main component of a behavioral model. Our methods for testing their influence are described in this section before returning to our actual purpose of testing different stimuli in section “Stimulus version test”.

Evaluation metric RMSE

To rate the quality of each model run, we computed pattern values P1-P5 in analogy to the laboratory patterns from n = 100 fish. As an evaluation metric, we computed a weighted root-mean-square error (RMSE) between all model and laboratory pattern mean values (Table 1 in Results) for each model run as where a was a pattern weight factor chosen per pattern importance (a = 0.3, a = 0.1, a = 0.2), i = 1..5 was a pattern iterator, and j = 1..3 was a pattern value iterator. As P1-P3 consisted of three pattern values each, these were averaged. A value of RMSE = 0 implies perfect agreement between model and laboratory results. We chose RMSE as it is both simple and penalizes large differences, which makes it useful for filtering outliers and achieving an overall good agreement.

The different weights of patterns P1 and P2 reflected their expected importance for modeling orientation (section “Movement patterns”). We also tested equal weighting per pattern average value and equal weighting per pattern value, but obtained negligible effects on the stimulus version ranking. We focused on comparing the mean instead of SD values to reduce the complexity of model development, accepting that SD agreement cannot be evaluated.

Random seed sensitivity

The behavioral rules used pseudo-random numbers which were generated from a random seed number fixed for each model run. For identical seeds, results were identical. To minimize and to quantify influence of the random seed choice, we tested each parameter set with N = 10 different seeds. Then, we averaged the resulting RMSEmodel-run values to obtain one RMSE value per parameter set. As seeds, we used ten true random numbers from www.random.org in the range from 1–999: 656, 36, 849, 934, 679, 758, 743, 392, 655, 171.

Parameter set generation

In total, the model had 7 fixed and D = 20 variable parameters (Table 2 in Results and Table 3 in S4 Appendix, p. 8). The latter could be split into D = 14 parameters for general movement, D = 3 parameters for initialization, and D = 3 individual parameters that were used only for single stimulus versions.

For parameter sensitivity testing (next section, “Parameter sensitivity test”) and stimulus version testing (section “Stimulus version test”), we needed to generate a number of parameter sets with systematic variation in their values. The values of an initial default parameter set were established by trial-and-error. For variation, we chose another 5 values in equal distances around each default value to test a total of p = 6 values for each parameter. The default values were either at the 3rd or 4th place of the range (Table 3 in S4 Appendix, p. 8).

The full parameter space would comprise p ≈ 3.7 ∙ 1015 parameter sets. We sampled it using the revised Morris method ([40], r = 1000 random trajectories, T = 50 optimum trajectories) implemented in Python 3.7.4 with SALib 1.3.8. We obtained n = (D+1) ∙ T = 1050 systematically varying base parameter sets (Fig 3, box 1). They consisted of D parameter values each and each set differed from the next one by exactly one parameter value. Sets differing only in a stimulus-dependent parameter value (D parameters) were used only for the corresponding stimulus version.

Fig 3

The process of testing our behavioral model.

The process starts with the generation of varying parameter sets, which are used for testing parameter sensitivity and contrasting stimulus versions in the following. Each box corresponds to one text section.

Parameter sensitivity test

To identify parameters of strong and negligible influence, we analyzed parameter sensitivity using the wall distance stimulus version (Fig 3, box 2). As the acceleration/TKE stimulus threshold parameter was not required for this version, the number of parameter sets decreased by T variations to n = 1000 parameter sets. As described in section “Random seed sensitivity”, for each parameter set we ran the model with N random seeds and averaged the resulting RMSEmodel-run values to obtain one RMSE value per parameter set (Fig 3, box “Random seed sensitivity”).

Results were analyzed by means of revised Morris screening [40, 41]. This method uses the effect of single parameter variations on the RMSE value to determine an influence measure, μ*, and an interaction measure, σ, for each parameter. We used μ* for ranking. For further interpretation, the values of the measures needed to be classified as “high” or “low”. This is usually achieved graphically [e.g., 40]. We took a more quantitative approach through normalizing μ* and σ by their respective maximum values and defining (arbitrary) thresholds for “high” values at μ*/ μ* > 0.6 and for “low” values at μ*/μ* ≤ 0.1 in dependence of the results (section “Parameter sensitivity”). We neglected three parameters with “low” influence in the subsequent stimulus version tests.

Stimulus version test

To address our guiding question (“How well do different stimuli explain upstream orientation?”), we compared the five stimulus versions of the “migrating” behavioral rule in both flume setups (Fig 3, box 3). All other behavioral rules of the model (including behavior selection, holding, drifting, and vertical behavior) were kept constant and were only tested implicitly (for a protocol, see [29]).

Parameter influence

Model parameters required special attention for the stimulus version test due to their large influence on the RMSE result. To eliminate the risk of favoring a stimulus by a certain parameter set, we applied a large number of parameter sets to each of the stimulus versions. We re-used the varying base parameter sets generated for parameter sensitivity testing (section “Parameter set generation”), but removed the three negligible parameters identified in sensitivity analysis. Using the same parameter sets for all versions ensured comparability. Only minor differences from the individual parameters (that were used only for single stimulus versions) were inevitable:

the threshold factor was used only by the velocity and acceleration versions;

the default swim angle for the wall distance stimulus was half the default swim angle of the velocity, TKE, and acceleration versions. It was not used in the baseline version.

These differences were also reflected in the slightly varying number of parameter sets tested per stimulus version, n. The average was n = 860 parameter sets in a range of 800–900 (Fig 3, box 3).

Running

For each parameter set, we repeatedly ran the model with N random seeds and averaged the resulting RMSEmodel-run values to obtain one RMSE value (section “Random seed sensitivity”). From the combination of parameter sets (n = 860), flume setups (n = 2), and stimulus versions (n = 5), we received a total of 8600 RMSE values (Fig 3, box 4).

Ranking and analysis

Since sampling of the parameters was optimized to cover the parameter space (section “Parameter set generation”) and not towards better RMSE values, the tests produced many irrelevant, high RMSE values which masked differences between the stimulus versions. We ranked versions by the 10th percentile of RMSE values to neglect the irrelevant values and to accent relevant low (optimum) values instead.

To evaluate if differences between the stimulus versions were statistically significant (p < α = 0.01), we computed the p-values between their RMSE values. We only considered values below the 10th percentile to cut off irrelevant RMSE values while preserving enough RMSE values for a statistically valid comparison. We used SciPy 1.0.1 with scikit-posthocs 0.6.6 to perform a Kruskal-Wallis test for all versions of a setup and a post-hoc Nemenyi test for pairwise tests between the wall distance version and the remaining versions.

During both ranking and statistical analysis, we treated the setups separately, as setup 1 was more demanding on the model due to the more heterogeneous flow field and the slot geometry.

Finally, to facilitate vivid understanding and discussion, we classified the versions qualitatively. The best version per setup was classified as “good”. Versions differing significantly (p < α = 0.01) from the best version were classified as either “moderate” or “poor”, depending on their 10th percentile value. Versions not differing significantly were classified as “good”.

Results

Analysis of the movement patterns provided several insights into observed trout behavior (Table 1). Pattern P1 indicated that brown trout spent the major part of trial time close to the lateral walls and screen, i.e. in a rather small share of the flume width. P2 showed that fish swam close to the bottom almost exclusively. P3-P5 were less distinct. Large SD values indicated wide individual differences in observed behavior for P1 and P3.

Table 1

Mean and standard deviation (SD) of brown trout patterns in the two flume setups.

Pattern	Flume zone		Setup 1 (jet)			Setup 2 (no jet)
	Name	Extent	n	mean	SD	n	mean	SD
	(facing downstream)	(m)		(%)	(%)		(%)	(%)
P1 Lateral distribution	Left	0.25	25	33	±37	24	38	±39
	Middle	2.00	25	10	±12	24	6	±10
	Right	0.25	25	57	±38	24	56	±40
P2 Vertical distribution	Surface	0.15	25	1	±2	24	4	±17
	Middle	0.30	25	1	±2	24	0	±1
	Bottom	0.15	25	99	±3	24	96	±17
P3 Turn zone	Upstream	3.25	9	24	±19	12	27	±17
	Middle	3.25	9	29	±17	12	40	±16
	Downstream	3.25	9	47	±16	12	33	±26
P4 Few turns	-	-	15	60	-	12	50	-
P5 nD/n	-	-	25	84	-	24	79	-

Values rounded to the closest integer. n and n, number of fish at lines D and A.

Brown trout start positions were distributed to the P1 left/middle/right zones 68% - 4% - 28% (setup 1) and 46% - 13% - 42% (setup 2). They spent an average of about 60% of track duration without other fish nearby.

CFD model results were in good agreement with acoustic Doppler velocimeter measurements from the laboratory [32]. Advection dominated the flow field; boundary roughness influence was limited to a few centimeters in both setups (Fig 4, S3 Appendix). Absence of the jet in setup 2 considerably reduced mean velocity U, acceleration magnitude, and TKE on the right-hand flume side (facing downstream). The flow field of both setups was asymmetric in lateral direction. The bulk flow velocity was U = Q/(width*water depth) ≈ 0.67 m/s. In large parts of the jet region of setup 1, flow velocity was about U ≈ 0.8 m/s.

Fig 4

Horizontal flow field slices of setup 1 (a) and setup 2 (b) illustrate the effect of the slot and jet and its removal. Flow from left to right. Plane at z = 0.07 m above bottom where trout preferred to stay. Water depth = 0.60 m. Sub-panels in (a) and (b) show velocity magnitude U (top panels), acceleration magnitude (middle panels), and TKE (bottom panels). Note logarithmic scale for acceleration and TKE.

Random seed and time step sensitivity

To evaluate the influence of the different random seeds used, we compared the median coefficient of variation (σ/μ) for all ten setup/stimulus combinations. The values were low and ranged from 0.02–0.04, suggesting that our results were independent of the random seed chosen.

Halving the default time step to Δt = 0.25 s for the best wall distance parameter set deteriorated the RMSE in setup 1 and improved it in setup 2. Doubling the time step deteriorated the RMSE in both setups.

Parameter sensitivity

Parameter sensitivity was computed for the wall distance stimulus version to identify parameters of strong and negligible influence. Setup 1 exhibited seven parameters of strong influence (μ*/μ* > 0.6, Table 2), while setup 2 exhibited four parameters of strong influence. The lowest ranks (17–19 in Table 2) comprised three parameters of negligible influence in both setups (μ*/μ* ≤ 0.1). There was no first-order (independent) parameter (indicated by a high μ* and low σ value).

Table 2

Parameters of the behavioral model with wall distance stimulus, ranked by normalized influence, μ*/μ*, in setup 1.

Setup 1 (Jet)			Parameter name	Setup 2 (No jet)			No.
Rank	μ*/μ*max	σM/σM,max		Rank	μ*/μ*max	σM/σM,max
1	1.00	1.00	Migrating behavior max. random angle	2	0.74	0.60	14
2	0.82	0.87	Holding behavior extent	5	0.55	0.56	9
3	0.77	0.89	Migrating behavior wall distance angle	4	0.63	0.59	16
4	0.70	0.86	Fatigue (decreasing) memory coefficient	1	1.00	1.00	6
5	0.68	0.72	Vertical behavior correction angle	16	0.12	0.21	12
6	0.67	0.68	Spot memory coefficient	3	0.68	0.89	5
7	0.66	0.79	Motivation denominator (“fast fish”)	7	0.47	0.69	3
8	0.56	0.64	Motivation memory coefficient	6	0.47	0.60	2
9	0.47	0.53	Stuck time threshold	10	0.22	0.34	13
10	0.41	0.46	Vertical behavior elevation threshold	13	0.13	0.17	11
11	0.33	0.41	Fatigue denominator	14	0.13	0.19	8
12	0.32	0.37	Motivation denominator (“slow fish”)	9	0.27	0.37	4
13	0.24	0.34	Motivation initial value	11	0.21	0.30	1
14	0.24	0.29	Initialization “slow fish right”/“right fish”	15	0.13	0.17	20
15	0.21	0.20	Initialization “slow fish left”/“left fish”	12	0.13	0.34	19
16	0.20	0.23	Initialization “left fish“/“total fish”	8	0.44	0.44	18
17	0.09	0.10	Drifting behavior straight drift probability	17	0.10	0.23	10
18	0.07	0.07	Fatigue (increasing) memory coefficient	18	0.05	0.08	7
19	0.04	0.04	Migrating behavior wall detection range	19	0.02	0.02	15

σ/σ is normalized interaction with other parameters. The threshold for “low” values is indicated by a horizontal line. The last column, No., is for reference to Table 3 in S4 Appendix, p. 8.

Further analysis showed that both swim angles included in the migrating behavioral rule had a high influence and interaction in both setups. This result confirms that modeling orientation was essential for agreement with the laboratory observations. The vertical behavior correction angle was distinctly more important in setup 1 than in setup 2. This was likely caused by vertical currents induced by the head drop at the slot in setup 1.

Parameter sets differing only in one of the three negligible parameters (Table 2, lowest ranks) were removed from the final stimulus version tests to reduce complexity. The sensitivity test was not repeated for stimulus versions other than wall distance as (a) the parameter “wall detection range” was not used in other versions, and (b) the other two negligible parameters were not related to orientation, i.e. were expected to not interact strongly with the chosen stimulus.

Stimulus versions

We contrasted five stimulus versions for upstream orientation (swim angle selection) in the model and ranked them by their RMSE 10th percentile (Table 3, Fig 5). The Kruskal-Wallis group test obtained p = 16.2E-54 for setup 1 and p = 1.1E-78 for setup 2, suggesting that the stimulus version did influence agreement with the laboratory observations significantly.

Table 3

Stimulus versions ranked by their RMSE 10th percentile, per setup.

Setup	Stimulus version	RMSE 10th percentile	p-Value	Rating
1—Jet	Wall distance	20	-	good
	Acceleration	21	1.2E-01	good
	Baseline	21	4.0E-01	good
	TKE	25	*4.1E-30	poor
	Velocity	26	*2.7E-27	poor
2—No jet	Wall distance	14	-	good
	Velocity	16	8.9E-02	good
	Baseline	18	*1.2E-12	moderate
	Acceleration	20	*9.5E-29	moderate
	TKE	26	*2.0E-60	poor

The p-values were calculated using a post-hoc Nemenyi test between wall distance and the particular stimulus version. p-values are rounded to two significant digits, percentile values are rounded to integer. Significant differences at p < α = 1.0E-02 are denoted by an asterisk (*) and cause the rating to be either “moderate” or “poor”. Underlying data in S6 Appendix.

Fig 5

Violin plots indicating the frequency of root-mean-square error values (RMSE) in both setups.

Ordering the stimulus versions by increasing 10th percentile reveals similarities and significant differences. Results are better if RMSE is closer to zero. The lower violin tip indicates the result of the best parameter set(s) for each stimulus version. Underlying data in S6 Appendix.

Wall distance was the best stimulus version in both setups. Acceleration and velocity differed widely in their ranks between setups, while baseline and TKE did not. In setup 1, the order of acceleration and baseline was not distinct, as well as the order of TKE and velocity (Fig 5). This is reflected in our qualitative rating used for discussion (Table 3). The estimated distributions of the RMSE values are mainly multi-modal due to unfavorable parameter sets (section “Ranking and analysis”), which was a major reason to rank them by their 10th percentile. The overall better results in setup 2 indicate that it was less demanding on the model than setup 1.

The wall distance version with its best parameter set reproduced all five patterns with high accuracy (RMSE = 9 for setup 1, RMSE = 6 for setup 2) (Figs 5 and 6). The finding shows that our model is able to reproduce the system characteristics that are captured by our patterns [6].

Fig 6

Two selected real and two selected model trout tracks in setup 1 (with jet).

Each sphere shows one noted position or model time step, respectively. Model results are from the best parameter set of the wall distance stimulus. Plane shows mean velocity, U, at z = 0.07 m above the bottom. Flow direction is from left to right, fish start from right on both flume sides. See S1 and S2 Appendices for observed trout movement data.

Discussion

Stimuli for orientation

In this work, we combined the very different fields of ethology, hydraulic engineering, and behavior modeling to approach a classical problem of behavioral ecology: How can orientation behavior be explained through external stimuli? The resulting movement patterns, flow fields, and significant differences between the model stimulus versions enable us to develop explanations with respect to the ambient conditions in our two setups.

RMSE rating results for the baseline version ranged from moderate in setup 1 to good in setup 2, indicating that rheotactic orientation and a random component may already explain not only orientation (pattern P1), but also large parts of the remaining observed behavior captured by P2-P5. Positive rheotaxis was expected to be important as both velocity direction and observed movement directions were largely uniform and aligned, fostering this basic behavior [17]. Also, this result confirms that the random component was chosen in the right order to cover effects not modeled explicitly. Still, it was unexpected that the other hydraulic variables commonly associated with behavior, i.e. velocity magnitude, TKE, and acceleration [e.g., 8, 20, 39], were not required for this result. The most apparent explanation that the flow field was not challenging enough to influence trout behavior will be discussed along with the alternative stimulus versions tested.

The velocity version yielded poor results in setup 1 and good results in setup 2. One reason for the failure in setup 1 was that model fish mostly avoided the jet of higher velocity on the right-hand side, while observed fish frequented this zone in both setups (P1). The laboratory observations indicate that the tested trout did not aim to minimize their energy consumption by following low-velocity paths upstream as could be expected [18, 22]. Hence, the good model results in setup 2 are not the result of a real mechanism. They are rather a model effect of the low-velocity zone on the right-hand side, which fostered wall attraction and thus agreement to the observations. Or, put differently, there is no apparent reason why trout would seek lower velocity in setup 2, but would not do so in setup 1 (with larger bulk velocity). In conclusion, our model and laboratory results suggest that velocity was not a relevant stimulus for our trout.

To find an explanation, we used the body length as a proxy for swimming ability and considered the velocity/body length relation. A trout (BL = 0.27 m) holding station experienced a relative flow velocity of about U = 3.0 BL/s in large parts of the jet in setup 1. At our water temperature, trout can maintain a sustained swimming speed of U = 4.5 BL/s for up to 200 min according to Ebel’s model equation for the rheophilic guild [38]. Thus, in theory, trout could hold or move upstream while experiencing almost no fatigue [42] for longer than the test duration, indicating that the flow velocity was indeed too low to trigger a reaction. Some support for this explanation comes from two studies, where low-velocity seeking was observed only at much larger relative velocities. Broadly estimated relative velocities were (0.37 m/s)/(0.07 m/BL) ≈ 5.3 BL/s for Duboulay’s rainbowfish [18] and (0.39 m/s)/(0.06 m/BL) ≈ 6.5 BL/s for longnose dace [39]. In contrast, [22] reported low-velocity seeking already at (0.40 m/s)/(0.14 m/BL) ≈ 2.9 BL/s for small hatchery brown trout, which is similar to our relative velocity. This discrepancy indicates that a general movement hypothesis has to include more factors than just relative velocity.

Our next test for constant TKE attraction scored poor results in both setups. As both walls and bottom were smooth, they did not provide a distinct TKE stimulus for wall attraction, resulting in large deviations in P1 and the RMSE. These results challenge the statement that TKE is “by far the best stimulus” for fishway models [12] and the hypothesis that a constant turbulence level can be employed for orientation [20]. We did not test attraction towards increasing or decreasing TKE, but would expect similarly poor results for P1 as the TKE field offers a decent stimulus only towards one side (lateral asymmetry), while the fish observed visited both sides.

Like for velocity, we compared TKE levels with literature values to determine if a relation to body length could explain our results. Our setups exhibited levels of TKE = 0.001–0.1 J/kg. Studies which explained behavior using TKE reported higher values for smaller fish: TKE = 0.1–0.3 J/kg for trout, BL = 0.20 m [12] and TKE = 0.015 J/kg for dace, BL = 0.07 m [20]. This qualitative comparison is in agreement with the explanation that the flow field did not cause behavioral reactions as it was not challenging for trout. However, TKE is less studied for orientation than velocity and further research is required.

Results of the acceleration version (good and moderate, respectively) were similar in absolute RMSE terms between setups, but differed widely in their in-setup rank because of better performances of the velocity and baseline versions in setup 2. Finding a theoretical explanation is difficult, as the few existing quantitative studies dealing with acceleration as an orientation stimulus focus on downstream movement [8, 43], which is associated with different responses than upstream movement. As for velocity (but with the better setup swapped), we suspect that a model effect, and not a real mechanism, is responsible for the different ranking: Acceleration amplifies the velocity gradient at the lateral walls and produces a distinct stimulus for model wall attraction in setup 1, but not in setup 2, thus failing to match P1 observations (which are similar in both setups). Taken together, the most probable explanation for our inconsistent results with acceleration, velocity, and TKE is that the live trout did not orient by means of the flow field (except for direction), but by other, non-hydraulic factors. Among these, the most appropriate candidate is non-hydraulic perception of wall distance.

Using the wall distance version, our model performed well in terms of absolute RMSE and significantly better than any alternative version in at least one setup. This result is in line with the assumed limited hydraulic influence on orientation in our conditions. As this model version is based on following a gradient pointing towards the wall, it carries the risk of imposing [6] large residence durations near the walls. This risk is mitigated by two mechanisms for moving away from the wall, namely the random overlay angle and inclined drift. The good RMSE results for this version suggest that these mechanisms are in the right balance and that wall attraction by distance was a real behavior. Potential reasons include guidance, cover, or hydraulic advantages from the walls, but further experiments are required to evaluate their probability.

For the physiological mechanism of wall attraction, multiple explanations are conceivable. To us, visual distance estimation is the best explanation, considering that vision is a predominant stimulus in fish [16, 25, 44]. Given the clear visibility of the walls, it seems unlikely that another potential far-field stimulus (e.g. acoustic) dominated distance estimation. Distance estimation by means of a hydraulic signature is also unlikely according to our CFD model results: as discussed before, they show that wall influence was limited to some centimeters distance and/or was inconsistent between setups in hydraulic variables (velocity, TKE, acceleration) which are commonly associated with behavioral responses [e.g., 8, 20, 39]. As a study on blind cave fish shows, hydrodynamic imaging by the lateral line close to smooth walls is limited to a small fraction of the body length [45].

Taken together, our results indicate that the rarely considered wall distance can be a relevant stimulus for orientation in fishways and similarly confined migration paths. It requires, of course, ambient conditions which enable its perception, e.g. clear and light water. Further, it may be overruled by hydraulic stimuli if they are relatively larger (in relation to fish swimming ability) than in our conditions.

Stimulus combinations

In our model, we limited orientation stimuli to the baseline velocity direction and one additional stimulus at a time. This is a simplification of reality, where most complex behavior is unlikely to depend on a single stimulus/sense alone [20, 46], and is more likely to have a polysensory background [47]. While making a model more realistic is not the same as making it more useful, combining stimuli to act on the same behavior is a worthwhile research direction for improving the generality of fish orientation IBMs.

A recent example is the use of calibrated weight factors for three hydraulic stimuli in an upstream orientation IBM for carp [13]. A further step would be to model shifting influence of stimuli depending on the ambient conditions, which was shown to exist in different species [16, 48]. However, adapting our model to use e.g. acceleration in setup 1 and velocity in setup 2 would still yield results not as good as the wall distance version (two times rank 2 vs. two times rank 1). In addition, both stimuli were likely not real stimuli (as discussed above) and hence such a model version would reduce robustness and explanatory power for our experimental conditions. Therefore, we decided not to run our model with stimuli combinations.

Movement patterns in the laboratory

Transfer of our model also depends on how universal the underlying movement patterns are. The observed tendency to swim close to the walls (pattern P1) matched descriptions of rainbow trout behavior [49] and barbel behavior in a confined model fishway [48]. It was not observed in faster flow and a more narrow flume for brook and brown trout [37]. This discrepancy indicates a potential dependence on the ambient conditions. For our slow, shallow flow, we conclude that P1 is not an untypical behavior.

Swimming at the bottom is reported frequently, e.g. in a natural stream for brown trout [50], and in flumes for rainbow trout [49], barbels [21], brown and brook trout [37], and longnose dace [39]. The water column in our experiment was shallow (d = 0.60 m), hence it was unclear whether fish preferred the bottom for orientation, for shelter, or for another reason. Despite possible behavioral or physiological reasons, P2 can be considered a typical behavior in our flume conditions.

For patterns P3 and P4, we did not find matching descriptions in the literature, because they represent a new way of systematically dealing with the striking feature “turns”. P5 was also too specific to our setup.

From the literature comparisons, it seems plausible that patterns P1 and P2 would be universal enough to support model transfer to similar flows, e.g. to an altered flume geometry. Transfer to the real world could be challenging, e.g. due to limited wall distance estimation and/or larger flow velocities.

General limitations

Finally, we want to point out some general limitations of this work to facilitate critical reception. First, our results are limited in temporal and spatial scale to the estimated fish observation accuracy, which was coarser than the scale of changes in the hydraulic variables within a total distance of 1–2 m downstream of the slot. Thus, potential effects of sudden changes in this area on the behavior could not be incorporated into the model.

Second, we found that different parameter combinations produced similarly good results, which indicates that parameter values are not robust and transfer of the behavioral model to other species and/or hydraulic environments requires recalibration. This does not affect our reproducible testing procedure of complex hypotheses, which is one of the major assets of IBMs [4].

Third, the behavioral model was sensitive to changes in time step width. Main influences changing with the time step include accuracy of flow field perception [9] and the number of orientation and random decisions per time. In our view, even the larger time step tested (Δt = 1.0 s) should be sufficiently fine to meet our respective requirements. Hence, this result likely emphasizes sensitivity of the parameter calibration regarding the chosen time step rather than some physical meaning.

Fourth, matching P1 depended, among other effects, on the initial distribution of fish positions to either side. The share of fish starting on the left flume side in the best wall distance version was off the real distributions observed in the flume (33% vs. 68% with jet and 33% vs. 46% without jet). Producing matching results from differing initial conditions could point to a difference between model and laboratory behavior or it could be a sign of minor relevance of the initial conditions to the model result.

Fifth, our model is currently limited to steady flow fields. Modeling e.g. utilization of transient vortices would require considerable model modification and computational resources [14].

Sixth and last, although swimming in groups (shoaling) was not pronounced in our data set and not modeled, it can influence orientation behavior. Determining its influence would require laboratory experiments with single fish releases.

Conclusions

Our goal in this study was to better understand upstream fish orientation and navigation in confined space, e.g. in fishways, by means of behavioral rules. Our thoroughly tested IBM enabled representation of orientation choices with an accuracy in the order of degrees. It was able to reproduce five movement patterns of brown trout in two experimental flow fields, which represented a broad range of behavior in a wide flume. The first two patterns–preferring wall and bottom proximity–were very distinct, and should be investigated further to determine their dependency on water width and depth.

The significant advantage of the wall distance stimulus version illustrates that a focus on hydraulic stimuli for predicting fish orientation can be too narrow, especially in hydraulic conditions not challenging for the observed individuals. These can occur e.g. for strong swimmers in multi-species fishways. In such conditions, particularly when the walls are visible, wall distance should be considered more frequently as a stimulus in IBMs, laboratory experiments, and field studies.

The evaluation of observed patterns and key parameters indicates that transfer of our present model is likely limited to similar species and conditions, i.e. brown trout of a given size in a relatively slow, shallow flume flow. This does not affect the IBM’s ability to contrast different behavior hypotheses.

In summary, the interdisciplinary IBM technique facilitates testing of behavioral rules using alternative stimuli with flexibility and rigor. It is an important option to approach the many urgent questions of fishway designers.

Brown trout movement observations, setup 1.

Unpublished movement data from Schütz et al. [28].

(XLSX)

Click here for additional data file.

Brown trout movement observations, setup 2.

Unpublished movement data from Schütz et al. [28].

(XLSX)

Click here for additional data file.

Hydraulic field data.

ZIP file containing S3a_hydraulic_fields_setup1.csv and S3b_hydraulic_fields_setup2.csv. For coordinate system origin, see Fig 1.

(ZIP)

Click here for additional data file.

Behavioral model description.

(PDF)

Click here for additional data file.

Behavioral model code.

The code is written in C++ and Fortran 90. The Fortran code uses the FORTRAN 77 fixed format for historic reasons. https://github.com/baw-de/ELAM-flume

(TXT)

Click here for additional data file.

Root mean square errors.

The order is Wall distance (setup 1, setup 2)–Acceleration 1, 2 –Velocity 1, 2 –TKE 1, 2 –Baseline 1, 2.

(CSV)

Click here for additional data file.

22 May 2021

PONE-D-21-08612

Development of behavioral rules for upstream orientation of fish in confined space employing wall distance and hydraulic stimuli

PLOS ONE

Dear Dr. Gisen,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please ensure that your decision is justified on PLOS ONE’s publication criteria and not, for example, on novelty or perceived impact.

Please submit your revised manuscript by Jul 06 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see:  http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at  https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols .

We look forward to receiving your revised manuscript.

Kind regards,

Atsushi Fujimura

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The introduction is poorly organized and confusing. The discussion of hydraulic parameters is insufficient and unclear.

Why would learning effects matter in this scenario?

I don't understand why the fish swimming study was designed the way it was. Was it specifically designed to calibrate/validate the model?

I cannot provide a good review of the movement patterns section (2.2) because it is not something I've studied.

Please provide more information on CFD model setup, calibration and validation.

I would like more explanation for why decisions were made ....e.g. why use averaged/steady results from the CFD model?

I love this stuff and I think its interesting work, I just want more information in the methods that is clearly organized and suggest a full rewrite where you focus on streamlining terminology, writing good paragraphs, and correcting grammar errors. I'm a bit concerned that it is unclear how much of this work is Goodwin's and how much is new work.

Reviewer #2: Nicely thought out work. Please see manuscript PDF where I mention a few places where better word choice will help the reader. More broadly, I think the manuscript could be improved in terms of readability by improving the flow of information. Some detail and terms within the manuscript make it hard to follow at times, so perhaps consider moving some technical pieces/info together (or use more general terms where possible) to allow more portions of the manuscript to flow (read) easier.

Reviewer #3: In this manuscript, a multi-faceted study is used to develop a framework to model fish movement to test different behavioral rules to explain fish orientation and navigation. The study uses fish tracking data from an unpublished laboratory study of brown trout passage through a partitioned flume. Pattern orienting modelling was used to capture 5 distinct movement patterns. The patterns were used as a metric for which to evaluate the performance of the behavioral rules. The authors use the Eulerian-Lagrangian agent method (ELAM) developed by Goodwin et al. (2014) as a foundation for their individual based model (IBM). The authors IBM was used to evaluate three behavioral rules using five different guidance stimuli. The authors found that wall distance guidance stimulus performed best at replicating the movement patterns. Overall, the manuscript was well written and combined with the extensive detail in the appendices (e.g., ODD report) provides a reasonable account of a complex study. Several aspects of the study are not entirely novel. The laboratory study is addressed in a separate manuscript, the CFD analysis is standard practice, and the ELAM provides the general computational infrastructure for the IBM. Therefore, the novelty of this study lies entirely within the behavioral rule and guidance stimulus selection and analysis. While the sensitivity analysis and modelling output are rigorous, the findings are somewhat muted. As detailed later, I found the laboratory study to be incongruous with the movement hypotheses being tested, likely leading to the less explanatory stimulus of wall distance being the best fit. Despite this issue, I feel the manuscript is still an important finding in the field of predictive modelling of fish movement. Due to the number of comments and level of effort required to address, I recommend the manuscript undergo major revisions.

Detailed and line-by-line comments are provided below:

Introduction

1. Line 42 – A reference to Goodwin et al. (2014) would seem appropriate here.

2. Line 48 – While the introduction touches briefly upon aspects of the fuller study, it lacks sufficient detail to understand why certain selections of behaviors and guiding stimuli were chosen. In this instance, citing the IPOS framework to describe what aspects of turbulence are important to fish behaviors would be relevant.

3. What about motivation to move or context specific behaviors? The authors need to address the complexities caused by differences in the internal state of a fish to make different decisions to the same stimuli.

4. The introduction would further benefit from more details on how the current effort extends or differs from previous IBMs. Specifically, it is not clear how the model is differentiated from the ELAM used by Goodwin et al. (2014). The only reference to the ELAM in on Lines 73 and 84 stating the proposed model is “ELAM-type”. I feel the general reader is not going to understand what this means.

5. The authors need to more explicitly state what hypotheses they are testing and why. The preceding paragraphs lists evidence that points to a lot of hydraulic variables that could be influential to fish movement. The authors do not explain why they examine baseline rheotaxis, velocity magnitude, TKE, flow acceleration, and wall distance relative to other choices including turbulent intensity, velocity gradients, eddy sizes, etc.

Methods

6. Line 110 – Where does the “x=9.74” come from? Figure 1 indicates an observation point at x=7.49.

7. Figure 1. The figure caption is the first mention of patterns P1 and P5 without defining them. I suggest moving this statement into the main text after P1 and P5 are defined.

8. Line 109-119 – The methods and scales at which fish movement was tracked is unclear. Where observers tracking movements in real-time on paper as well as noting location relative to the wall or screen or position in a group? How often was the position recorded? The authors state that observations were verified qualitatively with video records, why not use the video to obtain more detailed tracks. Overall, the methods on how tracking was accomplished needs significant more detail.

9. How were the set-ups chosen for the laboratory tests and how does this relate to the central hypotheses being tested?

10. Line 121-147 – How were these patterns chosen and what hypotheses drove these decisions? Their selection seems somewhat random as written. For example, what details informed splitting the channel at a distance of 0.25 m? Does this distance coincide with an observed hydraulic feature or behavior?

11. Line 155 – Omission of the screen bars is not sufficiently detailed in the methods. I understand the reason for not modelling the bars explicitly, but why not model them as a permeable surface to replicate some of the fine-scale turbulence. See Ho et al. (2011).

Ho, J., Coonrod, J., Hanna, L.J., and Mefford, B.W. 2011. Hydrodynamic modelling study of a fish exclusion system for a river diversion. River Res. Applic. 27: 184–192. doi:10.1002/rra.1349.

12. Line 172 – Again, what underlying hypothesis drove the selection of just these 3 hydraulic variables?

13. Line 200 – Was distance to the floor included in the wall distance evaluations?

14. Since the behavioral rules and selection of stimuli are the main contributions of this work, I found the model description provided in the main text to be underwhelming. I generally understand the adherence to the ODD protocol, but this should not sacrifice the completeness of the main text to act as a standalone document.

Results

15. Table 1 – The general magnitude and relation of patterns appears to be nearly identical between set-ups. I would even doubt there is any statistical difference between values. This would indicate one of two possible failures in the experimental design: 1). The modifications to the laboratory set-up did not achieve a discernable change in behavior; or 2). The patterns do not adequately capture the behavioral changes caused by the modification to the set-up. Either way, since the authors do not provide any rationale as to what the set-up change was indented to do makes interpretation difficult.

16. Figure 6 – This figure was very helpful to understand the model results and laboratory data. Additional versions of this figure to compare the modelled movement against observed movement would be beneficial. Illustrations from both set-ups should be included.

Discussion

17. Considering that the observed movement patterns were not largely different between the two set-ups, the conclusion that velocity does not play a role in orientation and navigation is stated too strongly. Based on the available data, it would appear that fish largely exhibited exploratory behavior and followed the walls because the hydraulics did not require any modified behavior. The authors should at least comment that their rule of avoiding high velocities is rather simple. It would not appear to fit the behavioral data either, since passage required movement through an area of elevated velocity in set-up 1. Perhaps a more refined guidance rule surrounding velocity fields is necessary.

18. Line 530-532 – The rationale for not running the model with combinations of stimuli is lacking. I can understand the reason for not including this step in this study, but to dismiss their validity is premature. The authors do not explicitly indicate how multiple stimuli could be integrated into a single model, so I do not understand how they can claim the acceleration and velocity versions could not improve upon the wall distance version without explicitly testing this.

19. Line 589 – The authors repeatedly indicate that wall perception could be accomplished by vision or audition, but it is not entirely clear how or what role audition may play in wall distance perception. I would still expect near field hydrodynamics would override any true acoustical input for most fish. This just strikes me as an odd statement and should be further explained if it remains in the manuscript.

Appendix S4

20. Page 4, Para. 1 – The authors state that the relationship between fatigue and swimming speed is not well understood. I find this statement to be highly inaccurate. Numerous studies are available that examine brown trout (and similar salmonids) swimming performance, that is their swim speed vs. duration curves. The authors also site work by Castro-Santos et al. (2005 and 2013) that shows how swimming fatigue could be modelled and swimming performance data for brown trout.

21. Page 8, Table 3 – How was Ug estimated? Castro-Santos et al. (2013) clearly shows laboratory data on brown trout indicated an optimal ground speed of 5.65 BL/s.

22. Page 10, Section 7.4 – Energy usage is cumulative at the time scale of this study. Once fish expend energy to swim above their sustained swim speed, they are unlikely to recover within 60 mins. However, since the authors assume no fish became fully exhausted from swimming in the flume, their handling of fatigue as a behavioral metric is understandable. This caveat could be clearer in the main text.

23. Page 6, Section 4.7 – The authors should comment on whether the random effects encompass stochasticity in animal behavior and the hydraulic stochasticity (i.e., the real flow field is not static).

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: R. Andrew Goodwin

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: PONE-D-21-08612_reviewer_revAndy.pdf

Click here for additional data file.

5 Jul 2021

PLEASE SEE BELOW FOR ORIGINAL TABULAR FORM

Response to reviewers

First, we would like to thank the reviewers for their insightful comments, which helped us to improve the manuscript. Our particular responses can be found below. Generally, we tried to further streamline terms, correct errors, and to improve readability throughout the whole manuscript.

We note that the manuscript does not yet meet the style requirements of PLOS ONE, but decided to await the provisional Editorial Accept decision (per https://journals.plos.org/plosone/s/getting-started) to avoid confusion during review e.g. by changing the citation style.

Reviewer #1

Comment Addressed? How addressed or why not? Line (original)

1. The introduction is poorly organized and confusing. The discussion of hydraulic parameters is insufficient and unclear. Yes We’ve rewritten large parts of the introduction, focusing on readability and the motivation for our study. We’ve also deleted information on stimuli that went too much into detail or - where helpful - moved them to the model description to provide a rationale on why each stimulus was chosen.

2. Why would learning effects matter in this scenario? No Fish already familiar with the flume geometry and hydraulic field could deviate in their behavior unpredictably, e.g. navigate straight towards the upstream end or don’t move at all. Excluding such effects by testing every individual just once makes results more reproducible. 110

3. I don't understand why the fish swimming study was designed the way it was. Was it specifically designed to calibrate/validate the model? Yes (See also comments #9 and #15 of reviewer #3.)

The previous, unrelated study was primarily designed to investigate the influence of auxiliary water velocity coming from the screen on passage success and delay of five species.

Conditions were chosen to create hydraulic conditions typical for a large multi-species fishway in Germany. It is now published in Ecological Engineering, where more information is available: https://doi.org/10.1016/j.ecoleng.2021.106257

We used the both setups to establish an application range for our model, see also comments #9 and #15 of reviewer #3.

We’ve added information on design choice and the word “unrelated” in section 2.1 for clarification. 100

4. I cannot provide a good review of the movement patterns section (2.2) because it is not something I've studied. - - -

5. Please provide more information on CFD model setup, calibration and validation. No We feel the information provided on the model setup is sufficient for reproduction. Calibration in the classical sense was not performed, as boundary conditions are fully determined by the laboratory flume. Mesh, turbulence model, and screen negligence were validated using ADV probe data (line 362f). Comprehensive detail information can be found in (Gisen 2018, p. 33ff.). 153ff

6. I would like more explanation for why decisions were made ....e.g. why use averaged/steady results from the CFD model? Yes Using transient CFD results is a promising direction to capture behavior currently modelled stochastically. However, interpolating thousands of 3D fish positions on 7200 flow field snapshots (dt = 0.5 s for 1 h) is currently far beyond our simulation capacity, even using a high-performance cluster. We’ve added a brief mention to the limitations section 4.4.

7. I love this stuff and I think its interesting work, I just want more information in the methods that is clearly organized and suggest a full rewrite where you focus on streamlining terminology, writing good paragraphs, and correcting grammar errors. Yes We’ve added information to the methods section and tried to streamline terminology in the whole manuscript. We considered to change the order of subsections in section 2.5, but refrained from doing so, as there are too many dependencies in both ways, e.g. between RMSE, RMSE(model-run), and parameter set evaluation. We rely on figure 3 to explain these dependencies.

8. I'm a bit concerned that it is unclear how much of this work is Goodwin's and how much is new work. Yes We’ve tried to clarify in the introduction that Goodwin (2014) worked on downstream migration on dam-scale, while we work on upstream migration on fishway-scale, which required e.g. a completely new behavioral model. The CFD model was also replaced, of course, and the software framework was heavily modified (section 2.4.1).

Reviewer #2

Reviewer #2: “Nicely thought out work. Please see manuscript PDF where I mention a few places where better word choice will help the reader. More broadly, I think the manuscript could be improved in terms of readability by improving the flow of information. Some detail and terms within the manuscript make it hard to follow at times, so perhaps consider moving some technical pieces/info together (or use more general terms where possible) to allow more portions of the manuscript to flow (read) easier.”

Comment Addressed? How addressed or why not? Line (original)

1. Unclear, please restate another way. Yes Rephrased 50

2. Unclear. Yes Replaced “like for P1” with “using Eq. 1” 136

3. Unclear, please restate another way. Yes Rephrased 321

4. Unclear, please restate another way. Yes Rephrased 322

5. underlying ? Yes Deleted “a” 537

6. misplaced "e.g." ? No Other influences than width and depth are conceivable, such as smooth boundaries and illumination 584

7. demanding enough for No From our perception “enough” would suggest a need to create more demanding hydraulic conditions, which was not our intention in this study (see reviewer #1, comment #3). 586f.

Reviewer #3

Reviewer #3: “In this manuscript, a multi-faceted study is used to develop a framework to model fish movement to test different behavioral rules to explain fish orientation and navigation. The study uses fish tracking data from an unpublished laboratory study of brown trout passage through a partitioned flume. Pattern orienting modelling was used to capture 5 distinct movement patterns. The patterns were used as a metric for which to evaluate the performance of the behavioral rules. The authors use the Eulerian-Lagrangian agent method (ELAM) developed by Goodwin et al. (2014) as a foundation for their individual based model (IBM). The authors IBM was used to evaluate three behavioral rules using five different guidance stimuli. The authors found that wall distance guidance stimulus performed best at replicating the movement patterns.

Overall, the manuscript was well written and combined with the extensive detail in the appendices (e.g., ODD report) provides a reasonable account of a complex study. Several aspects of the study are not entirely novel. The laboratory study is addressed in a separate manuscript, the CFD analysis is standard practice, and the ELAM provides the general computational infrastructure for the IBM. Therefore, the novelty of this study lies entirely within the behavioral rule and guidance stimulus selection and analysis.

While the sensitivity analysis and modelling output are rigorous, the findings are somewhat muted. As detailed later, I found the laboratory study to be incongruous with the movement hypotheses being tested, likely leading to the less explanatory stimulus of wall distance being the best fit. Despite this issue, I feel the manuscript is still an important finding in the field of predictive modelling of fish movement. Due to the number of comments and level of effort required to address, I recommend the manuscript undergo major revisions.”

Comment Addressed? How addressed or why not? Line (original)

Introduction

1. Line 42 – A reference to Goodwin et al. (2014) would seem appropriate here. Yes The reference Silva et al. 2018 serves to underline the “high research priority” of behavioral rules. We did not intend to reference general studies using behavioral rules here, elsewise we would have to list much more than just Goodwin’s. We’ve tried to make this clearer by using a direct quote of Silva et al.

2. Line 48 – While the introduction touches briefly upon aspects of the fuller study, it lacks sufficient detail to understand why certain selections of behaviors and guiding stimuli were chosen. In this instance, citing the IPOS framework to describe what aspects of turbulence are important to fish behaviors would be relevant. Yes For choice of behaviors and stimuli see response to comment #1 of reviewer #1.

We’ve added a reference to IPOS (DOI:10.1002/rra.1584) for general understanding. However, for our IBM study (and others we are aware of), periodicity, orientation, and scale are still too detailed (see comment #5 below and new reference Roth et al. 2021).

3. What about motivation to move or context specific behaviors? The authors need to address the complexities caused by differences in the internal state of a fish to make different decisions to the same stimuli. Yes This is an important aspect and we’ve added a corresponding paragraph in the introduction. As we focused on orientation and fishway design, we modeled internal stimuli only rather basically (using proxies for motivation and fatigue). We now discuss this in the model description (2.4).

4. The introduction would further benefit from more details on how the current effort extends or differs from previous IBMs. Specifically, it is not clear how the model is differentiated from the ELAM used by Goodwin et al. (2014). The only reference to the ELAM in on Lines 73 and 84 stating the proposed model is “ELAM-type”. I feel the general reader is not going to understand what this means. Yes See response to comment #8 of reviewer #1.

We also point more directly to lacks of previous upstream IBMs (testing, resting behavior). We’ve removed the term “ELAM”, as it is not required to understand our work.

5. The authors need to more explicitly state what hypotheses they are testing and why. The preceding paragraphs lists evidence that points to a lot of hydraulic variables that could be influential to fish movement. The authors do not explain why they examine baseline rheotaxis, velocity magnitude, TKE, flow acceleration, and wall distance relative to other choices including turbulent intensity, velocity gradients, eddy sizes, etc. Yes See response to comment 1 of reviewer #1.

We’ve added a justification for each stimulus tested in the behavioral model description (2.4) and a summary in the introduction. Generally, we preferred the more comprehensive variable over specialized variables. E.g. all three turbulence intensity components are contained in the TKE (see e.g. IPOS paper, p. 432).

The velocity gradient tensor is contained in the acceleration metric.

Strain rate and rotation are contained in the velocity gradient tensor.

Reynolds shear stresses could be relevant to future work.

Eddy size, as well as eddy orientation, is an interesting, but advanced measure left for future work (see response #2 above). It is partly covered in the rheotaxis stimulus.

Methods

6. Line 110 – Where does the “x=9.74” come from? Figure 1 indicates an observation point at x=7.49. Yes At x=9.74 m, there was an external pole on the flume used as mark for quick orientation in the original study. We’ve removed this information, as it does not affect this manuscript.

7. Figure 1. The figure caption is the first mention of patterns P1 and P5 without defining them. I suggest moving this statement into the main text after P1 and P5 are defined. Yes We’ve inserted into the caption a reference to section 2.2 where the patterns are defined. If the caption was changed per your suggestion, readers could be confused about the meaning of the gray areas and line D.

8. Line 109-119 – The methods and scales at which fish movement was tracked is unclear. Where observers tracking movements in real-time on paper as well as noting location relative to the wall or screen or position in a group? How often was the position recorded? The authors state that observations were verified qualitatively with video records, why not use the video to obtain more detailed tracks. Overall, the methods on how tracking was accomplished needs significant more detail. Yes We’ve added this lacking information to the methods (real-time recording on paper; positions were recorded on every notable change; definitions of changes in the three dimensions).

We’ve also added to methods and appendix S1 the information that x_value and y_value were computed in postprocessing.

A manual video analysis of fish tracks would have been too time consuming and (because cameras were only positioned laterally) would not have been reliable for lateral movements; we also worked on an automated 3D-fishtracking of the videos (along 11 overlapping cameras) but with three fish swimming simultaneously this is demanding and proved to be less efficient than the manual records for this project.

9. How were the set-ups chosen for the laboratory tests and how does this relate to the central hypotheses being tested? Yes (See also response to comment #15 of reviewer #3 and comment #3 of reviewer #1.)

The previous study (now available Schütz et al. 2021, https://doi.org/10.1016/j.ecoleng.2021.106257) was not related to our study. It was designed to investigate the influence of auxiliary water velocity coming from the screen on behavior of five species.

We’ve added the word “unrelated” in section 2.1 for clarification.

We chose two hydraulically different setups (information added) from this study to test the application range of our model/hypotheses. We did not expect the behavior to be this similar (e.g., Kerr 2016 reported low-energy seeking for trout at smaller U=0.40 m/s).

We still found the result interesting, as it expands the common focus of fishways designers on hydraulic stimuli towards vision under certain conditions (possibly low relative flow velocity and visible boundaries).

10. Line 121-147 – How were these patterns chosen and what hypotheses drove these decisions? Their selection seems somewhat random as written. For example, what details informed splitting the channel at a distance of 0.25 m? Does this distance coincide with an observed hydraulic feature or behavior? Yes Added explanation: To capture the most striking spatial behaviors observed (wall and bottom proximity and turns). 0.25 m was chosen arbitrarily to define wall proximity.

11. Line 155 – Omission of the screen bars is not sufficiently detailed in the methods. I understand the reason for not modelling the bars explicitly, but why not model them as a permeable surface to replicate some of the fine-scale turbulence. See Ho et al. (2011).

Ho, J., Coonrod, J., Hanna, L.J., and Mefford, B.W. 2011. Hydrodynamic modelling study of a fish exclusion system for a river diversion. River Res. Applic. 27: 184–192. doi:10.1002/rra.1349. No While it is possible to model the screen with bars as a permeable baffle, we avoided the effort of implementation and calibration, as we judged the flow field to be sufficiently close to ADV measurements with respect to our behavioral data accuracy. See also reference Gisen 2018, p. 33ff., for detailed comparisons.

Regarding fine-scale turbulence, we would expect some kind of damping effect from a permeable baffle, which would require extra attention. 155

12. Line 172 – Again, what underlying hypothesis drove the selection of just these 3 hydraulic variables? Yes See response to comment #1 of reviewer #1.

13. Line 200 – Was distance to the floor included in the wall distance evaluations? No No, it wasn’t, as the vertical behavior was treated as a separate behavioral rule, which relied on the vertical coordinate (~ pressure) as a stimulus.

14. Since the behavioral rules and selection of stimuli are the main contributions of this work, I found the model description provided in the main text to be underwhelming. I generally understand the adherence to the ODD protocol, but this should not sacrifice the completeness of the main text to act as a standalone document. Yes We’ve added information on motivation and fatigue, the basic internal states of the model and added a reference (Castro-Santos 2004) to substantiate the importance of repeated passage attempts in the model description.

Results

15. Table 1 – The general magnitude and relation of patterns appears to be nearly identical between set-ups. I would even doubt there is any statistical difference between values. This would indicate one of two possible failures in the experimental design: 1). The modifications to the laboratory set-up did not achieve a discernable change in behavior; or 2). The patterns do not adequately capture the behavioral changes caused by the modification to the set-up. Either way, since the authors do not provide any rationale as to what the set-up change was indented to do makes interpretation difficult. No (see also response #9)

We agree that our trout behavior as measured by the patterns is very similar between setups. This would be your case (1).

However, the previous, unrelated study (Schütz et al. 2021) did not aim to produce different behavior, but was intentionally limited to a typical maximum slot velocity in a multi-species fishway (1.5 m/s). We’ve added this information in section 2.1.

This limitation coincides with the goal of the present study to provide behavioral rules for fishway design. Hence, the only goal of using two hydraulically different setups was to quantify the (minimum) application range of our model. Finding the maximum application range would be interesting, but less relevant to design of multi-species fishways, which also need to address species with lesser swimming abilities than trout.

16. Figure 6 – This figure was very helpful to understand the model results and laboratory data. Additional versions of this figure to compare the modelled movement against observed movement would be beneficial. Illustrations from both set-ups should be included. No This figure was intended primarily to foster qualitative understanding. It has little explanatory power, as it depicts only two selected observed and two selected model tracks. Showing more tracks like this would not add quantitative information. Also, a figure showing results for setup 2 would add little additional information (as behavior was not much different). Hence, we would prefer to leave the figure unchanged.

Discussion

17. Considering that the observed movement patterns were not largely different between the two set-ups, the conclusion that velocity does not play a role in orientation and navigation is stated too strongly. Based on the available data, it would appear that fish largely exhibited exploratory behavior and followed the walls because the hydraulics did not require any modified behavior. The authors should at least comment that their rule of avoiding high velocities is rather simple. It would not appear to fit the behavioral data either, since passage required movement through an area of elevated velocity in set-up 1. Perhaps a more refined guidance rule surrounding velocity fields is necessary. No By no means we intend to state that velocity does not play are role in orientation, generally. However, we found that in our conditions, which are typical for large multi-species fishways built in Germany, brown trout did not use velocity for orientation (probably because it was not demanding enough). We’ve added explanations to the discussion to clarify this distinction.

Reviewer #3 is right in the point that orientation towards reduced velocity hinders passage through the jet region and slot in setup 1. It is not impossible, however, because the motivation/fatigue driver for movement is independent of the orientation stimulus and there is also a random component to orientation. Further, it cannot be the single cause for the poor performance of the velocity stimulus version in setup 1, as pattern P5 (which measures slot passage) only contributes 20% to the RMSE metric (section 2.5.1).

Still, more refined behavioral rules allowing adaptation to changing flow fields are definitely worthwhile for future work, as discussed in section 4.2.

18. Line 530-532 – The rationale for not running the model with combinations of stimuli is lacking. I can understand the reason for not including this step in this study, but to dismiss their validity is premature. The authors do not explicitly indicate how multiple stimuli could be integrated into a single model, so I do not understand how they can claim the acceleration and velocity versions could not improve upon the wall distance version without explicitly testing this. Yes Our statement was based on comparing the 10th percentile/rank results of the stimulus versions under the assumption of using only one stimulus per setup. It is true that we did not test using multiple stimuli within one setup. However, we argue that velocity and acceleration are not relevant for brown trout orientation in our flume conditions (see comment and response #17), therefore we would expect similar results from a model using both stimuli in the same setup.

We rephrased the relevant lines to clarify this point.

19. Line 589 – The authors repeatedly indicate that wall perception could be accomplished by vision or audition, but it is not entirely clear how or what role audition may play in wall distance perception. I would still expect near field hydrodynamics would override any true acoustical input for most fish. This just strikes me as an odd statement and should be further explained if it remains in the manuscript. Yes We agree that this is a hypothetic stimulus on the scales investigated here and removed the word “acoustic” in the abstract and conclusions to make that clearer.

Appendix S4

20. Page 4, Para. 1 – The authors state that the relationship between fatigue and swimming speed is not well understood. I find this statement to be highly inaccurate. Numerous studies are available that examine brown trout (and similar salmonids) swimming performance, that is their swim speed vs. duration curves. The authors also site work by Castro-Santos et al. (2005 and 2013) that shows how swimming fatigue could be modelled and swimming performance data for brown trout. Yes Agreed, we deleted this statement.

21. Page 8, Table 3 – How was Ug estimated? Castro-Santos et al. (2013) clearly shows laboratory data on brown trout indicated an optimal ground speed of 5.65 BL/s. Yes We estimated the value based on Castro-Santos’ 2005 result for striped bass (2.56 BL/s) as stated. We agree that their 2013 paper is a better reference as it deals directly with brown trout. However, note that their value of 5.65 BL/s is a theoretical value for prolonged swim speed and most individuals chose to swim “well below” it (p. 287). Hence, our estimation seems to be in a realistic order. We changed the reference in Table 3, footnote 2, as well as the rationale in section 7.5 accordingly.

22. Page 10, Section 7.4 – Energy usage is cumulative at the time scale of this study. Once fish expend energy to swim above their sustained swim speed, they are unlikely to recover within 60 mins. However, since the authors assume no fish became fully exhausted from swimming in the flume, their handling of fatigue as a behavioral metric is understandable. This caveat could be clearer in the main text. Yes We did not intend to model fatigue and recovery processes in detail, but just enough to drive model movement as measured by the patterns to enable orientation analyses.

We’ve mentioned the caveat in section 2.4, and included references to Castro-Santos et al. 2013 and Ebel 2014.

23. Page 6, Section 4.7 – The authors should comment on whether the random effects encompass stochasticity in animal behavior and the hydraulic stochasticity (i.e., the real flow field is not static). Yes We’ve added the explanation that our behavioral model’s stochasticity also covers transient flow field effects on the behavior. S4 section 4.7

Submitted filename: Response to reviewers.docx

Click here for additional data file.

22 Oct 2021

Submitted filename: PONE-D-21-08612_R1.pdf

Click here for additional data file.

10 Dec 2021

See tabular document attached (Response to reviewers.docx)

Submitted filename: Response to reviewers.docx

Click here for additional data file.

2 Feb 2022

Development of behavioral rules for upstream orientation of fish in confined space

PONE-D-21-08612R2

Dear Dr. Gisen,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Atsushi Fujimura

Academic Editor

PLOS ONE

4 Feb 2022

PONE-D-21-08612R2

Development of behavioral rules for upstream orientation of fish in confined space

Dear Dr. Gisen:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Atsushi Fujimura

Academic Editor

PLOS ONE