Assessing hydrodynamic space use of brown trout, Salmo trutta, in a complex flow environment: a return to first principles

Understanding how the physical environment influences the distribution and movement of animals is a fundamental theme in ecology (Moorcroft, 2012). To maximise fitness, individuals must utilise space in ways that facilitate energy intake, through the acquisition of food, while minimising costs (Krebs, 1978; Maynard Smith, 1978). Those that most effectively do so can allocate a greater proportion of available resources (time and energy) to other activities, such as the detection and evasion of predators, growth and reproduction (Parker and Maynard Smith, 1990). For stream-dwelling fish adapted to the challenges imposed by a spatially and temporally complex hydrodynamic environment, energetic costs include those associated with maintaining stability (Tritico and Cotel, 2010; Webb and Cotel, 2010), swimming (Enders et al., 2003; Liao, 2004) and capturing food (Chesney, 1989). The use of space that enables fish to minimise energy expenditure under complex flows is of interest to evolutionary biologists, ecologists, fisheries managers and conservationists, and is the focus of this study.

The selection of appropriate hydrodynamic metrics that predict space use by fish has been the subject of recent debate (Lacey et al., 2012) and a cause of controversy. This is for three reasons. First, traditional methods tend to heavily rely on the correlation between single-point time-averaged velocity statistics and space use by fish (e.g. Bovee, 1986; Degraaf and Bain, 1986; Facey and Grossman, 1992; Mäki-Petäys et al., 1997; Jenkins and Keeley, 2010). Such measures fail to capture important flow–fish interaction mechanisms that rely on turbulent flow properties that vary in both time and space. For example, fish use spatial flow patterns around bluff bodies to hold position with reduced energy expenditure by tilting their body at an angle at which hydrodynamic forces cancel out (e.g. entraining; Przybilla et al., 2010). Second, attempts to quantify habitat use by fish in the field typically produce mixed results (e.g. Heggenes et al., 1991; Facey and Grossman, 1992; Jowett and Richardson, 1995; Mäki-Petäys et al., 1997; Cotel et al., 2006; Enders et al., 2009). This is not surprising because space use is influenced by multiple confounding variables impossible to control in situ (e.g. food, predators, competitors and mates). Third, laboratory studies continue to attempt to find statistical links between patterns of fish distribution/movement and one or more of any number of turbulent flow characteristics, such as turbulence intensity (TI), relative turbulence intensity (RTI), turbulent kinetic energy (TKE), turbulent length scale (TLS) or Reynolds shear stresses (τ) (e.g. Smith et al., 2005, 2006; Silva et al., 2011, 2012a,b; Duarte et al., 2012). However, the biophysical interpretation of these statistical links remains obscure. This results in the failure to understand and investigate the explanatory mechanisms that underpin interactions between fish and their hydrodynamic environment. In this study we argue that there is a need to return to first principles and quantify space use by fish in response to complex flows under controlled laboratory conditions. Furthermore, experimental data should be interpreted in-line with the general principle that space use is dictated by energy conservation, or more specifically, strategies to minimise the costs of swimming.

The energetic cost of swimming for motile organisms is intrinsically linked to drag and mass-related gravitational forces (Biewener, 2003). For fish, which are typically of a similar density to the surrounding medium, hydrodynamic drag imposes the largest influence on locomotion because mass-related gravitational forces are negligible (Biewener, 2003). As such, the reduction of hydrodynamic drag plays an important role in improving swimming performance. Fish have evolved numerous passive and active mechanisms to reduce drag and hence energetic costs during locomotion. For example, the streamlined morphology of a fish reduces flow separation and minimises form drag (Vogel, 1996), while epidermal mucus (Daniel, 1981) and riblets (Dean and Bhushan, 2010) reduce skin friction (passive mechanisms). In addition, conventional swimming kinematics are optimised to prevent flow separation during undulation (Anderson et al., 2001), fish can take advantage of reduced drag during intermittent non-undulatory phases of burst–glide locomotion (Weihs, 1974) and, as already mentioned, they can use stable or predictable flow characteristics around bluff bodies to hold position whilst reducing energetic expenditure (Taguchi and Liao, 2011) (active mechanisms). Even when fish exhibit behaviours where other fundamental biological needs are of primary importance, these are undertaken in a way whereby energy expenditure is minimised (e.g. feeding: Fausch, 1984; reproductive migration: McElroy et al., 2012). As such, energy-minimising strategies play a fundamental role in fish ecology and hence are likely to be an important driver in space use. However, calculating the drag force acting on a fish at a specific hydrodynamic location, and hence investigating the link between space use and energy expenditure, is difficult because sources of thrust and drag are not separable for undulating swimmers (Schultz and Webb, 2002). Although advances are being made in assessing the energetic costs and efficiency of swimming (e.g. Maertens et al., 2015), there are still few viable options for accurately calculating likely energetic expenditure at specific hydrodynamic locations.

This study adopted a reductionist approach to investigate hydrodynamic space use by brown trout, Salmo trutta (Linnaeus 1758), under a controlled experimental setting in which key confounding variables (e.g. visual cues, food, predators and conspecifics) were absent. The term ‘hydrodynamic space use’ is used to refer to the distribution of an animal in space in relation to local flow characteristics. To facilitate generalisation of the results, trout from both wild and hatchery origin exhibiting a range of body lengths were used. The position of individual fish was recorded at high spatial resolution in a complex flow field created using an array of different sized vertically oriented cylinders in a large open-channel flume. The flow field offered a highly heterogeneous hydrodynamic environment and, consequently, a wide range of potential positions from which the fish could choose. We hypothesise that under the conditions presented, space use would be governed by the adoption of energy conservation strategies. A hydrodynamic drag function (D) based on single-point time-averaged velocity statistics that incorporates the influence of turbulent fluctuations was used to infer the energetic cost of steady swimming. Further, zones in which there was a potential for fish to exhibit ‘specialised behaviours’ that rely on spatial and/or temporal flow features and enable fish to hold position with reduced energetic cost (e.g. Kármán gaiting, entraining and bow riding) were identified and their occupancy was recorded. Preference curves were constructed to assess the appropriateness of D as a descriptor of space use in comparison to other common hydrodynamic metrics: mean velocity (U), TI, RTI, TKE, horizontal Reynolds shear stress (τ_uv) and TLS. A fundamental assumption and potential source of error of conventional methods of calculating preference (quotient of use and availability) is that an organism has access to, and knowledge of, all space available to them (Beyer et al., 2010). In this study, preference calculations were refined by assessing space use of individual fish in relation to area ‘sampled’ rather than total available. The results of this study, and the methods presented, have important implications for understanding the ecology of fluvial fish that live in hydrodynamically complex environments, and for fisheries management and conservation.

The study was conducted in a large indoor recirculating flume (21.4 m long, 1.4 m wide and 0.6 m deep) at the International Centre for Ecohydraulics Research (ICER) facility, University of Southampton, UK (50°57′42.6″N, 1°25′26.9″W). The experimental area consisted of a centrally located 2.94 m long section of the flume (Fig. 1A,B), screened at the upstream end by a 100 mm thick polycarbonate flow straightener (elongated tubular porosity, 7 mm diameter) to minimise incoming turbulent fluctuations, and downstream by a 10 mm diameter wire mesh, both of which prevented fish leaving. A wire mesh release chamber (0.7 m long, 0.4 m wide, 0.5 m deep) in which the subject fish were held immediately prior to the start of each trial was connected to the experimental area by a centrally located rectangular orifice (0.2 m wide, 0.15 m high) in the downstream screen (Fig. 1A,B). Polyurethane sheeting was used to prevent light entering the experimental area and thus eliminated visual cues. Light intensities were below the threshold at which a human observer could see. Fish behaviour was recorded using low light cameras under infrared illumination.

Fish experienced flow fields created by one of two treatments (A or B) or a control (no cylinders present) during the study (Fig. 1C). In treatment A, cylinder diameters increased incrementally across the flume width so that the smallest and largest cylinders were located adjacent to the channel lateral walls (Fig. 1C). In treatment B, the combination of cylinders was switched so that the largest and smallest occupied locations close to the channel centre (Fig. 1C). The orientation of the cylinder array in each treatment was randomised (by rotating the array 180 deg about the central vertical axis) among trials to control for any hydrodynamic conditions associated with the flume. Data collected under the randomised cylinder array orientations were aggregated for each treatment. Discharge remained constant (0.15 m³ s⁻¹) during each trial independent of treatment. Water depth was roughly uniform (270 mm) throughout the experimental area under the control. Under both treatments, depth was greater upstream (275 mm) than downstream (265 mm) of the array because of head losses generated by the cylinders.

Instantaneous velocity in the longitudinal (u), transverse (v) and vertical (w) directions (Fig. 1) were measured (n=671, 678 and 84 for treatments A, B and control, respectively) using an acoustic Doppler velocimeter (ADV) (Vectrino, Nortek-AS, Norway; velocity sampling frequency 50 Hz, sampling volume 0.05 cm³, record duration 3 min, vertical distance from bed 90 mm). Raw data were filtered using a 3D velocity cross-correlation algorithm (Cea et al., 2007) and the time-averaged (overbar) and fluctuating parts (prime) of each instantaneous velocity component were calculated (e.g. ⁠), along with U, TI, RTI, TKE and τ_uv (see Table 1).

Based on the Reynolds numbers (Table S1), each cylinder wake was expected to be completely turbulent with both sides of the cylinder experiencing boundary layer separation (subcritical wake regime; Sumer and Fredsøe, 1997).

Taylor’s hypothesis of frozen turbulence (Taylor, 1938) (i.e. the hypothesis that turbulent eddies are advected at or near mean flow velocity) is notoriously violated (see Pope, 2000) in the near-wake region of bluff bodies. As such, two-point simultaneous ADV measurements were used to accurately assess the convection speed of the dominant energy-containing eddies (E_s) and infer TLSs. Two to five simultaneous measurements (n=28, sample length 5 min) were taken downstream of the central axis of each cylinder in treatment A using two longitudinally aligned (spacing 180 mm) and synchronised ADVs (first ADV located 270–1590 mm downstream of each cylinder array). E_s was calculated as the ratio between the distance between ADVs and the time lag (T) corresponding to the first peak appearing in the cross-correlation function of the synchronised lateral velocity data (Table 1). Well-defined cross-correlation peaks were observed for the majority of simultaneous measurements, except downstream of the smallest cylinders (10 mm diameter), where eddy coherence was lost. E_s was found to be linearly related (r²=0.687) and similar to the mean velocity ū between the two probes (E_s=0.77ū+0.12) (i.e. convection velocity was similar to flow velocity behind all cylinders). As such, it was deemed valid to use cross-section mean velocity (U_f) as a proxy for convection velocity in all treatments to calculate TLSs from the single-point measurements as E_d=U_fE_t/2, where E_d is the characteristic size (m) and E_t is the turnover time (s) of the dominant energy-containing eddies. E_t was calculated as the time till the first positive peak in the autocorrelation function of the lateral velocity component (Table 1). E_d downstream of each cylinder (11–123 mm for the 10–90 mm diameter cylinders, respectively) was similar (linear trend: y=0.98x+0) and highly correlated (r²=0.975) to the estimated values (λ/2 in Table S1).

Standard deviations of the lateral and vertical velocity components (σ_v and σ_z, respectively) are used as these quantify reasonably well the characteristic magnitude of velocity fluctuations. Although this approximation of drag is new to the field of ecohydraulics, similar approximations are used elsewhere (e.g. canopy drag: Dupont et al., 2008; sedimentology: Nalpanis et al., 1993). We acknowledge that this approximation of drag is static and limited to instances in which fish are station holding (i.e. thrust=drag). Fish actively moving upstream are likely to experience higher levels of relative drag than predicted by this metric. However, the metric represents a good proxy for assessing the likely energetic costs of holding station at different locations and hence is useful when investigating space use in relation to energy conservation strategies.

All the computed hydraulic metrics (U, TI, RTI, TKE, τ_uv, E_d and D) were interpolated (Delaunay triangulation with linear interpolation) at 1 mm resolution to create hydrodynamic maps of the experimental area (Fig. 3 for treatment A). As ADV measurements were not possible within 30 mm of the flume walls, a fitted logarithmic function, based on measurements taken at the walls under the control, was used to estimate boundary conditions. Sensitivity analysis revealed that boundary layer estimation method (several were trialled) had little effect on overall trends. The fitted logarithmic function was used as it was thought to better represent actual conditions.

Three size classes (age cohorts) of hatchery-raised brown trout were obtained so that the influence on space use of fish length and related swimming capability could be assessed. Medium and large fish were obtained from Allenbrook Trout Farm (Wimbourne, 50°53′43.9″N, 1°58′27.4″W) and small fish from Bibury Trout Farm (Bibury, 51°45′37.5″N, 1°50′08.9″W) on 15 March 2011 and 12 March 2012, respectively (Table 2). Wild trout, caught by electric fishing at Tadnoll brook, a tributary of the River Frome (Dorset, 50°41′02.5″N, 2°17′28.4″W), on 14 March 2012 (Table 2), were used to validate the results obtained from the more readily available hatchery fish. The wild trout group was not divided into size classes because of the low number and small size range caught. Instead, comparisons were made against an equivalent size range of hatchery fish where possible. Trout were transported to the ICER facility in aerated water and held in filtered 3000 litre holding tanks (pH 7.5–7.8; ammonia: 0; nitrite: 0; nitrate: <40; 50% weekly water change) at ambient temperature (2011: mean±s.d.=14.07±3.63°C; 2012: 10.23±0.80°C). Fish were fed once daily with commercial trout pellets.

A total of 118 one-hour trials were conducted: 51 between 29 March and 7 April 2011 (medium and large hatchery trout; flume temperature: mean±s.d.=14.80±3.02°C), and 67 between 21 and 30 March 2012 (small hatchery and wild trout; flume temperature: 10.37±0.79°C; Table 2). Individual trout were allowed to acclimatise for a minimum of 1 h in a porous container positioned within the flume before being placed into the release chamber from which they could volitionally enter the experimental area. A trial commenced once a trout entered the experimental area; fish that failed to enter within 20 min, or did so but became impinged on the downstream screen for more than 10 s, were removed and the trial was terminated. Each trial lasted a maximum of 1 h during which each trout was allowed to freely explore and utilise the experimental area. All trials were conducted between 08:00 and 18:00 h (BST). Trout were used once only and were weighed (g) and measured [fork length (FL); mm] at the end of each trial. The research was reviewed and sanctioned by the University of Southampton Ethical Review Board.

For each trial, fish snout positions (x and y spatial coordinates) were obtained every second from the overhead video footage (maximum of 3600 data points per trial) using Logger Pro v3.8.2 (Vernier, Beaverton, OR, USA) and plots of space use and space sampled created. Space sampled was deemed to be that which fell within a fish's mechanosensory field of detection (MFoD) during a trial. The threshold distance for fish to detect hydrodynamic signals using the lateral line (primary mechanosensory organ) varies depending on a number of factors (e.g. signal type, orientation and magnitude) but is approximately 1.5FL (Coombs, 1999), with signal discriminability being considerable higher at closer distances. In this study, a conservative estimate of the spatial extents of a fish's MFoD, deemed to be an area within which it can fully interpret the surrounding flow field, was set as a rectangle extending 0.5FL upstream, 1.5FL downstream and 0.5FL either side of the fish snout position (Fig. S1). Computationally, the space sampled was considered to be all the discrete interpolated hydrodynamic data points (1 mm resolution) that fell with a fish's MFoD during a trial (calculated at 10 Hz after linear interpolation of fish snout position). Sensitivity analysis revealed that the results were relatively unsensitised to the specific size of the assigned MFoD, with a modified MFoD of half and double the size of that outlined above having little effect on overall trends.

Specialised behavioural zones (SBZs) expected to provide opportunities for fish to reduce the energetic costs of station holding through the expression of unique swimming kinematics (e.g. Kármán gaiting, entraining, bow riding) were identified based on information obtained from the literature (e.g. Liao et al., 2003a,b; Liao, 2006; Przybilla et al., 2010), observation of trout during the trials, and clustering evident in the plots of space use. Using data aggregated from both treatments, the proportion of time individual trout spent in each SBZ was calculated and comparisons were made between: (1) observed and expected if fish had been evenly distributed throughout the experimental area (all trout: Wilcoxon signed-rank tests), and (2) wild and hatchery trout, using a sample (n=30) of hatchery fish of equivalent size to those obtained from the wild (Mann–Whitney tests). In addition, the relationship between FL and (3) the proportion of time spent in each SBZ (all trout) and (4) the diameter of the cylinder associated with the SBZ in which fish spent the highest proportion of time were assessed [Pearson's correlation with bootstrapped (n=2000) and bias corrected and accelerated confidence intervals (BCa CI); Efron and Tibshirani, 1993].

where n is the total number of trout in each group. Trends in the P_gh curves were compared with frequency distributions of total available hydrodynamic space (H_t). Trout preference for hydrodynamic space that was rarely available (<0.1% of total area) was disregarded by assessing P_gh curves only over the range of conditions where H_t>0.001. To assess the influence of specialised behaviours, P_gh curves were calculated using both unmodified and modified S_u data. The modified S_u data were constructed by identifying SBZs trout used more frequently than expected if their distribution had been even, and randomly removing the excess proportion of points that occurred within them (for graphical representation of process see Fig. S3). As such, the potential influence of specialised behaviours was removed from the modified preference curves. Differences in trends evident in the P_gh curves constructed using the modified and unmodified S_u data were visually assessed. Final conclusions in relation to hydrodynamic space use by brown trout were drawn from the P_gh curves constructed using the modified S_u data. It is acknowledged that even when confounding variables are absent space use data may be noisy due to erratic behaviour but over a large number of experimental trials it is assumed that underlying patterns evident in the group preference curves will be as a result of hydrodynamic conditions. Data analysis and visualisation was undertaken using MATLAB v7.10.0.499 (MathWorks, Natick, MA, USA), SPSS v20.0.0 (IBM, Armonk, NY, USA) and SigmaPlot v12.5.038 (Systat Software, San Jose, CA, USA).

Space use varied dramatically among individuals and treatments with trout tending to be thigmotactic under the control and influenced by the cylinders under the treatments (Fig. 4). Only 17.6% of trout sampled 100% of the experimental area (median: 86.5%, range: 7.0–100%). The extent of the experimental area sampled was not limited to the downstream section and did not follow a regular pattern (example: Fig. 4B).

The following SBZs were identified: (1) Kármán gaiting zones: the area behind a bluff body where fish can alter their body kinematics to synchronise with the vortices shed to reduce energetic expenditure (e.g. Kármán gait) (2.5d to 7.5d downstream of each cylinder, 1d wide); (2) entraining zones: the area in close proximity to the side of a bluff body where fish can hold position by tilting their body off parallel to the bulk flow to use the resulting lift and wake suction forces to mitigate for drag (e.g. Przybilla et al., 2010) (−d/4 to 1.5d downstream of each cylinder, 2.375d wide); (3) bow riding zones: the low-velocity, high-pressure area in front of a bluff body where fish can hold position with a reduced trailing edge pressure deficit (e.g. Liao et al., 2003a) (1FL to 1FL+d upstream of each cylinder, 1d wide); (4) tail holding zones: where fish were able to hold position facing the flow by placing their tail against the downstream screen and their body along the channel wall (previously undescribed behaviour) (0.6FL to 1.2FL upstream of the downstream screen, flush with the flume wall, 0.25FL wide); and (5) wall holding zones: the area where fish were able to hold position close to the flume wall and slightly downstream of the cylinder array with observed reduced body undulation (previously undescribed behaviour) (inline to 2.5d downstream of each cylinder, flush with flume wall, 2.375d/4 wide) (Fig. 5). Presumably, the wall holding behaviour utilised spatially stable hydraulic conditions created between the cylinder and channel side, enabling fish to hold position with reduced drag. Although wall holding likely involved similar mechanisms to entraining (see Liao, 2006; Przybilla et al., 2010), it was considered a separate behaviour because kinematics and focal position were clearly influenced by the proximity of the channel sides (Fig. 5B).

The period of time spent occupying SBZs varied greatly among individuals and groups (Fig. 6). Tail holding, entraining and wall holding zones were occupied by individual trout for a much higher proportion of a trial than expected (up to 86, 64 and 33%, respectively) if space use had been even throughout the experimental area (Fig. 6). At a group level, tail holding regions by small (z=4.107, P<0.001), medium (z=2.896, P<0.01) and large (z=3.173, P<0.001) hatchery trout, and wall holding regions by large hatchery trout (z=3.363, P<0.001) were consistently used more frequently than expected if the distribution had been even (Fig. 6). There was no difference in the percentage of time wild trout spent in each zone compared with hatchery trout of equivalent size. Smaller hatchery fish spent a higher proportion of time in tail holding (r_S=−0.428, P<0.001, 95% BCa CI=−0.571, −0.250) and a lower proportion in the Kármán gaiting (r_S=0.5451, P<0.001, 95% BCa CI=0.301, 0.588) and bow riding (r_S=0.242, P<0.05, 95% BCa CI=0.072, 0.462) zones than larger trout (Fig. S4). There was no correlation between FL and the proportion of time spent in the entraining or wall holding zones. Smaller trout were more likely to use entraining zones associated with smaller cylinders than larger trout (r_S=0.599, P<0.01, 95% BCa CI=0.244, 0.835; Fig. S5). There was no correlation between FL and cylinder diameter in relation to the Kármán gaiting or bow riding zones which trout used.

The key difference resulting from the two ways of calculating P_gh (with modified versus unmodified S_u data) was the absence of an additional peak in the U and D preference curves constructed using the modified S_u data (e.g. see Fig. 7). The additional peaks in the P_gh curves constructed using the unmodified S_u data represent the influence of fish occupying space in which specialised behaviours could be exhibited.

P_gh curves constructed using modified S_u data for trout group, treatment and hydrodynamic metrics are presented in Fig. 8. Preference for specific areas was highest (up to 25 times) for U and D, with the majority of groups preferentially using areas with lower U and D than was most frequently available under each treatment and the control (exception: large fish in treatment B). Preference for low drag areas was generally higher for wild and small hatchery trout than medium or large fish. In treatments A and B, preference for TI, RTI, TKE and τ_uv followed the H_t histograms, being higher for areas with low levels of each metric. There were a few exceptions, which included a slight preference exhibited by large fish for areas of high TKE (12.3 J m⁻²: 1.1 times) and τ_uv (7.8 N m⁻³: 1.8 times) in treatment A and B, respectively. Under the control, all groups preferentially used areas with higher than most frequently available TI, RTI, TKE and τ_uv. However, the range and magnitude of turbulence available was low. Trout preference for E_d differed for each group under each treatment. Under treatment A, peak preference of wild and small, medium and large hatchery trout was for areas where E_d=ca. 0, 10, 40 and 60 mm, respectively. Medium and large hatchery trout also exhibited an additional slight preference (ca. 1.3 times) for areas where E_d=ca. 125–140 mm. Under treatment B, peak preference of small hatchery trout was for areas with very small E_d (ca. 5 mm), but wild and medium and large hatchery trout showed no clear preference for E_d of any size.

Although it is widely accepted that animal distributions reflect a trade-off between the energetic benefits and costs of the microhabitat selected, for stream-dwelling fish, the identification of the most appropriate hydrodynamic predictors of space use is the focus of much debate (Lacey et al., 2012). In previous studies, a variety of metrics and approaches have been adopted, ranging from simplistic empirical measures of unidirectional velocity in the laboratory (e.g. Baldes and Vincent, 1969) or field (e.g. Conallin et al., 2014) to the outputs of numerical modelling of complex flows (e.g. Crowder and Diplas, 2006). Unfortunately, the variables used tend to lack biological realism because they provide a poor indirect proxy for some more ecologically relevant factor, while the approaches employed frequently fail to account for multiple confounding variables. In this study, we took a reductionist approach to define space use of brown trout under complex non-laminar flows in the laboratory in which density-dependent (conspecifics) and other confounding factors (e.g. visual cues, food and predators) were absent. A simple, robust and biologically relevant descriptor of drag that incorporates both mean and fluctuating velocity components provided a realistic surrogate for the energetic cost of holding station. Maximising the ratio of energetic benefits to costs is commonly surmised as the principle driver for space use (e.g. Bachman, 1984 for brown trout), but it seems to be frequently overlooked. In this study, we demonstrated that, as hypothesised, energy conservation strategies play a key role in space use in an energetically taxing environment, with the majority of trout groups choosing to frequently occupy areas in which specialised behaviours could be adopted or by selecting low-drag regions.

Independent consideration of drag when describing the distribution and movement of fish is not new. For example, McElroy et al. (2012), using a drag-based energetic cost function, found that pallid sturgeon (Scaphirhynchus albus) select less costly migratory routes, and Hughes (2004) proposed the avoidance of ‘wave drag’ as the cause behind discrepancies in the expected versus observed pathways of salmonids during their spawning migration. Further, the importance of turbulence is well recognised, and a variety of different metrics (e.g. TI: Smith et al., 2005; RTI: Cotel et al., 2006; TKE: Smith et al., 2006; and τ: Hayes and Jowett, 1994; Silva et al., 2011, 2012a,b; Duarte et al., 2012) have been used to define suitable habitat or migratory routes, although such statistical links remain difficult to interpret from an ecological perspective. However, attempts to quantify hydrodynamic space use by fish living in lotic environments frequently fail to consider both drag and turbulence together. Previous definitions of drag fail to account for turbulent fluctuations that occur under unsteady flows common in nature, and as a result likely underestimate energetic costs. Unless specific conditions are met, it is generally accepted that turbulence reduces swimming performance (e.g. Lupandin, 2005; Tritico and Cotel, 2010) and increases the cost of swimming (e.g. Enders et al., 2003, 2004, 2005). Hence a combination of both mean velocity and turbulence in a drag metric provides a simple and more biologically relevant hydraulic descriptor than mean velocity and/or a separate measure of turbulence alone. In this study, peaks observed in the drag preference curves were also evident in the mean velocity (U) data. This was expected because under the flow configurations created, U was generally much higher than its variation laterally (σ_v) or vertically (σ_z) (i.e. much higher than velocity variation due to turbulence). However, trends in space use were clearer in the drag than the mean velocity preference curves because the incorporation of the influence of turbulence (lateral and vertical velocity fluctuations) in the drag metric refined its predictive power. Hence, although in this study U provided a good proxy for drag and, concurrently, energetic expenditure, it is likely that in other situations with higher intensities of turbulence (e.g. within fish passes) this may not be the case. In any case, drag, either simply derived (e.g. D∝U²) or after incorporating turbulent fluctuations, is conceptually a more ecologically meaningful metric to describe space use than mean velocity, as it has direct implications for energetic expenditure during swimming. It is interesting to note that for the majority of fish groups, a sharp maximum in the preference curves occurred for D=ca. 0.12. It is likely that this was the lowest available level of drag that fish could effectively exploit, with even lower values occurring at locations in close proximity of solid boundaries that were inaccessible to fish because of their size.

Under certain circumstances fish can exhibit performance-enhancing behaviours (e.g. Kármán gaiting: Liao et al., 2003b; entraining: Przybilla et al., 2010; bow riding: Taguchi and Liao, 2011) that reduce the energetic costs associated with the lotic space selected. Such behaviours are commonly observed in nature, e.g. when trout associate with zones upstream or downstream of boulders and bridge footings or other similar natural or artificial structures. In this study, areas of the experimental arena in which fish were predicted to be able to benefit from specialised performance-enhancing behaviours were identified. In addition to those previously described, trout exhibited two additional behaviours: wall and tail holding. ‘Wall holding’, although similar to entraining as defined by Liao (2006) and Przybilla et al. (2010), was considered distinctly different because focal position was clearly influenced by the proximity of the channel side. It is likely under these conditions that the boundary layer modified local flow conditions and altered the direction of the resulting forces acting on the fish (e.g. lift and drag). Similar ‘wall holding’ clustering was depicted by Przybilla et al. (2010, their Fig. 2) but was not analysed or discussed. During ‘tail holding’, fish utilised a physical object to rest against (the downstream mesh screen), rather than exploiting spatial and/or temporal variations in hydrodynamics to gain an energetic advantage. Out of all the potential specialised behaviours identified, individual fish spent the highest proportion of time in tail holding zones, followed by entraining and then wall holding zones, and at a group level, tail holding and wall holding zones were consistently used. Individual trout utilised tail holding zones for extended periods of a trial (up to 86%) with almost no corrective fin movement or body undulation (J.R.K., personal observations), suggesting that this is the most energetically efficient mode of station holding under the experimental setting described.

Behaviours and distribution relative to the cylinder array varied with body size. As expected (e.g. Beamish, 1978), smaller fish had a lower swimming performance than larger fish, as indicated by shorter swim periods before becoming impinged against the downstream screen. Small fish also spent a higher proportion of time occupying low-drag zones and areas where they could exploit the most energy-efficient tail holding strategy, whilst large hatchery fish were more likely to maintain position in wall holding zones. Indeed, the only fish group not to display a peak preference for space with lower D than that most frequently available was the large fish in treatment B, likely because of their reduced need to conserve energy. There was also a positive relationship between body length and the diameter of cylinders on which trout tended to entrain (Fig. S5), suggesting hydrodynamic characteristics of proportional scale to fish length are beneficial, if not required, for this behaviour. The critical size of eddies required to destabilise a fish has been found to be approximately equivalent to body length (Pavlov et al., 2000; Lupandin, 2005; Tritico and Cotel, 2010). Therefore, small trout may have been destabilised by eddies produced downstream of the larger cylinders, an expectation supported by the observation under one cylinder arrangement (treatment A) in which medium and large hatchery trout tended to associate with areas with large eddy diameters, whereas smaller trout tended to associate with areas with smaller eddy size.

This study represents a step forward in understanding fish behaviour under hydrodynamically complex settings. We have shown that time-averaged single-point hydraulic metrics alone cannot accurately predict space use by fish under complex flows if specialised performance-enhancing behaviours that rely on spatial and temporal variation in flow are not accounted for. However, when specialised behaviours are accounted for, it was shown that a novel definition of drag allows for clear predictions of space use and that it is a more ecologically relevant predictor of space use than other commonly utilised metrics. Our observations provide empirical evidence that energy conservation strategies play a key role in space use by fish. These findings and the methods outlined in this paper have important implications for fish management practices, including the development of effective habitat suitability models, river restoration and fish passage. The results are robust as preference for space use was quantified by taking into account the sampled rather than the available hydrodynamic environment. This represents an improvement to the standard and potentially erroneous quotient-of-use-and-availability preferences curves commonly used (e.g. Mäki-Petäys et al., 1997). Future research should incrementally introduce additional factors (e.g. competitors, predators and food) to advance understanding of fish behaviour under the effects of, and interaction between, multiple variables, which characterize more realistic conditions typical of field settings.

We thank the members of the International Centre for Ecohydraulics Research (ICER) team, who kindly dedicated their time to assist with trials.