Masking a fish's detection of environmental stimuli: application to improving downstream migration at river infrastructure

According to Signal Detection Theory, the ability to detect a stimulus (discriminability, d') is inversely related to the magnitude of internal and external noise. In this study, downstream moving brown trout Salmo trutta were used to investigate whether external hydrodynamic noise (in this case turbulence) could mask a signal that induced an unwanted response, such as rejecting accelerating velocity gradients commonly encountered at entrances to fish bypass channels. S. trutta behaviour was quantified in the absence (control) or presence of an accelerating velocity gradient created by an unconstricted or constricted channel, respectively, under two levels (low and high) of background turbulent kinetic energy (hydrodynamic noise). Experiments were conducted in an indoor recirculating flume in the dark and a range of passage metrics were quantified. Under the control condition, most (ca. 91%) S. trutta passed, usually oriented downstream (67%), with minimal delay (median 0.13 min). In comparison, fewer S. trutta (ca. 43%) passed under constricted conditions, they tended to orient facing into the flow (ca. 64%) and delay was greater (median > 20 min). When viewed from a coarse‐scale perspective, discriminability of the velocity gradient was lower when turbulence was high suggesting masking of the signal occurred. However, the resulting increase in the percentage of fish that passed, decrease in time to pass and reduction in the distance at which S. trutta reacted (switched orientation) was subtle and non‐significant. Despite the mixed results obtained, the use of masking to manipulate an animal's perception of environmental stimuli as a fisheries management tool is conceptually valid and the results of this experiment present a useful stepping stone for future research.

According to Signal Detection Theory, the ability to detect a stimulus (discriminability, d') is inversely related to the magnitude of internal and external noise. In this study, downstream moving brown trout Salmo trutta were used to investigate whether external hydrodynamic noise (in this case turbulence) could mask a signal that induced an unwanted response, such as rejecting accelerating velocity gradients commonly encountered at entrances to fish bypass channels.
S. trutta behaviour was quantified in the absence (control) or presence of an accelerating velocity gradient created by an unconstricted or constricted channel, respectively, under two levels (low and high) of background turbulent kinetic energy (hydrodynamic noise). Experiments were conducted in an indoor recirculating flume in the dark and a range of passage metrics were quantified. Under the control condition, most (ca. 91%) S. trutta passed, usually oriented downstream (67%), with minimal delay (median 0.13 min). In comparison, fewer S. trutta (ca. 43%) passed under constricted conditions, they tended to orient facing into the flow (ca. 64%) and delay was greater (median > 20 min). When viewed from a coarse-scale perspective, discriminability of the velocity gradient was lower when turbulence was high suggesting masking of the signal occurred. However, the resulting increase in the percentage of fish that passed, decrease in time to pass and reduction in the distance at which S. trutta reacted (switched orientation) was subtle and non-significant. Despite the mixed results obtained, the use of masking to manipulate an animal's perception of environmental stimuli as a fisheries management tool is conceptually valid and the results of this experiment present a useful stepping stone for future research.

K E Y W O R D S
brown trout, fish behaviour, bypass entrance, sensory ecology, signal detection theory, velocity gradient

| INTRODUCTION
For downstream migrating fish, high levels of mortality can occur during passage through dangerous routes at anthropogenic infrastructure (e.g., turbines and spillways; Calles et al., 2010;Stokesbury & Dadswell, 1991;Taylor & Kynard, 1985). Direct causes of mortality include blade strike, shear, cavitation, rapid decompression (barotrauma), gas supersaturation and mechanical damage; indirect causes include stress, disorientation and elevated susceptibility to predation and disease (Kemp, 2016). To mitigate for these adverse effects, physical or behavioural screens are used to guide downstream migrating fish away from dangerous routes to more desirable bypass channels (Larinier, 2008). However, the efficacy of these systems is often low (Calles et al., 2012;Kynard & O'Leary, 1993), with fish avoiding conditions encountered at bypass entrances (Kynard & Buerkett, 1997) resulting in delayed migration (Ovidio et al., 2016) and increased probability of passage through hazardous routes (Johnson et al., 2000).
Accelerating velocity gradients are common at bypass entrances (Haro et al., 1998). Numerous studies have analysed fish behaviour as they encounter velocity gradients (Enders et al., 2009(Enders et al., , 2012Haro et al., 1998;Kemp et al., 2005Kemp et al., , 2006Kemp et al., , 2008Russon & Kemp, 2011;Vowles et al., 2014), but relatively little attention has been directed at reducing rejection rates; among the exceptions are Haro et al. (1998) and Vowles and Kemp (2012). On encountering velocity gradients, downstream moving juvenile salmonids typically switch orientation to face the flow and actively reject the near-field hydrodynamic transition by swimming back upstream; for example, Atlantic salmon smolts Salmo salar L. 1758 (Ovidio et al., 2016); Pacific salmon smolts Oncorhynchus spp. (Kemp et al., 2005); brown trout Salmo trutta L. 1758 . Partial success in reducing the rejection rate at a bypass entrance was achieved by altering the intake design and reducing the rate of acceleration (i.e., the signal strength) that fish experience as they pass (Haro et al., 1998). However, physical alterations that sufficiently reduce the rate of acceleration at bypass intakes may not always be possible. Vowles and Kemp (2012) tested the effect of localized illumination on rejection of a velocity gradient and found that avoidance and delay exhibited by S. trutta increased rather than decreased when multimodal stimuli were present. A better understanding of the sensory ecology and behavioural response of fish in relation to hydraulic stimuli is required to improve passage efficacy through bypass systems if the adverse effects of anthropogenic barriers are to be mitigated.
Psychophysics, a branch of psychology that explores the relationship between physical stimuli and perception, has been successfully utilized for over a century in the fields of sensory ecology and animal behaviour (Akre & Johnsen, 2014;Fechner, 1860;Graham, 1934). Signal detection theory (SDT), a key concept within psychophysics, considers the relationship between the magnitude and perceived intensity of a stimulus (signal) and the ability to discern between the signal and noise (discriminability, d'; Kemp et al., 2012). Kemp et al. (2012) utilized SDT to analyse the behavioural response of fish to different velocity gradients and highlighted its usefulness as a tool for understanding, quantifying and potentially manipulating how fish perceive and respond to sensory stimulus. An important premise within SDT is that d' is inversely related to the magnitude of internal and external noise. This masking effect, defined as a change in the probability of perceiving a signal in the presence of a second stimulus (Gelfand, 2010), has been observed for a diverse range of species in relation to multiple sensory modalities; e.g., sound (Lohr et al., 2003;Wollerman & Wiley, 2002), vibration (Wu & Elias, 2014), vision (Woo et al., 2009) and hydrodynamic signal detection (Bassett et al., 2006;Engelmann et al., 2002;Kröther et al., 2002). However, to the best of our knowledge, masking has never been used to limit the detection of a signal for conservation purposes, such as the protection of fish at river infrastructure.
Building on the results of Kemp et al. (2012), the aim of this study was to assess whether increased turbulence reduced the discriminability of an accelerating velocity gradient by downstream moving S. trutta in an experimental flume in the absence of visual cues. This was experimentally tested by assessing the behaviour of S. trutta in the absence (control) or presence of an accelerating velocity gradient created by an unconstricted and constricted channel, respectively, under high or low turbulent kinetic energy (TKE). The results were analysed in respect to: 1) coarse scale behaviours, the percentage of fish that approached and passed and time to pass; 2) fine scale behaviours, fish orientation, nature of response (orientation switch or rejection) and response location; 3) SDT, d' and a measure of response bias (response criterion), calculated from both a coarse and fine-scale perspective. It was predicted that S. trutta would be less likely to reject the accelerating velocity gradient when background levels of TKE were high, resulting in increased numbers passing through the constriction with reduced delay. nortekgroup.com; sample frequency 50 Hz, sample volume 0.05 cm 3 , record length 90 s) 150 mm above the bed (0.53 d). Hydraulics conditions were measured at a single depth as vertical variation in the flow was assumed to be minimal due to the lateral constriction of the channel and vertically uniform spacing of the mesh grid and flow straightener. Velocity data was collected at 0.53 d rather than 0.60 d (the depth that best approximates average vertical flow velocity; Herschy, 1995), as noise at this depth, probably acoustic reflection from the flume bed, interfered with ADV performance. Fewer hydraulic measurements were recorded in the control treatments as the flow environment was more homogeneous. Mean velocity (U, m s −1 ; Equation 1) and TKE (J m −3 ; Equation 2) were calculated from the filtered (3D cross correlation filter; Cea et al., 2007) ADV data: where u, v and w are the instantaneous velocity values corresponding to the x, y and z spatial coordinates, overbar and prime denote time-averaged and deviation from mean, respectively, and ρ is density  was at least 10% greater than free stream, a proxy for the spatial extent of the velocity gradient (hereafter referred to as the acceleration zone), was similar under both low and high TKE ( Figure 2).
Prior to each trial fish were placed in a porous container in the flume and allowed to acclimatize for at least 1 h. Each trial commenced when a single fish was placed in the release chamber and ended when it had either passed downstream beyond the point of maximum constriction (or equivalent location under control conditions) or after 20 min. Each S. trutta was used only once and weighed (g) and measured (L F , mm) at the end of each trial. After experimentation fish were humanly killed by immersion in an overdose of anaesthetic (2-phenoxyethanol: 1 ml l −1 ) followed by confirmation of death by permanent cessation of the circulatory system.

| Fish behaviour
The results were analysed in respect to: 1) coarse scale behaviours, percent approached, percent passed and time to pass; 2) fine scale behaviours, approach orientation and pass orientation (positive or negative rheotaxis), nature of response (orientation switch or rejection) and distance of initial response (R d , mm) from the constriction; 3) SDT, discriminability (d') and response criterion (c), from both a coarse and fine scale perspective ( Table 2). The metrics d' and c are measures of whether an organism can detect a stimulus and the level at which an internal response results in the exhibition of a behaviour, respectively. The advantage of using SDT to assess animal behaviour is that it accounts for response bias, i.e., the tendency for an animal to react as though a signal is present even if one is not. The two SDT metrics were assessed by calculating the percentage of fish that correctly or incorrectly passed downstream (coarse scale perspective) or displayed a behavioural response (fine-scale perspective) when the flume was either constricted or unconstricted (Table 2). Based on an assumption that fish should avoid an accelerating velocity gradient (signal), under the coarse-scale assessment fish were considered to have responded correctly if they failed to pass downstream under constricted conditions (hit; Figure 5) and incorrectly if they failed to pass downstream under control conditions (miss; Figure 5). When considering the fine scale perspective, fish were considered to have responded correctly if they displayed either an orientation switch or rejection (or both) when the flume was constricted (hit; Figure 5) or incorrectly if they displayed such a behavioural response under control conditions (false alarm; Figure 5).

FISH
The metric d' (Table 2) (Table 2) is a measure, in standard deviation units, of the distance of the response criterion from being unbiased (equal probability of a false alarm or a miss; Stanislaw & Todorov, 1999;Wickens, 2002). If the response criterion is unbiased, c has a value of 0. Negative values of c signify a bias toward responding yes whereas positive values signify a bias toward responding no.
The influence of the following variables, constriction (present or absent), turbulence (high or low) and approach orientation (positive or negative rheotaxis, where applicable) on each metric was assessed using appropriate statistical tests (  Time-to-pass was influenced by constriction (X 2 MC (1) = 31.599, P < 0.001; Figure 7(a)) and approach orientation (X 2 MC (1) = 4.707, P < 0.05; Figure 7(b)). Delay tended to be longer under the constricted treatment (median > 20 min v. median 0.13 min; Figure 7 (Figure 7(a)) but this did not translate to a significant influence of turbulence on time-to-pass.

| Fine-scale behaviours
Approach orientation and pass orientation were influenced by constriction (X 2 (1) = 5.765, P < 0.05 and X 2 (1) = 6.510, P < 0.05, respectively), but not turbulence. Salmo trutta tended to exhibit positive rheotaxis whilst they approached and passed when the flume was constricted (61.9 and 69.2%, respectively), but exhibit negative  The four potential signal-response outcomes that may occur in the presence or absence of a specific environmental stimulus. In this study, the signal was the velocity gradient and the response was downstream passage (coarse-scale assessment) or displaying avoidance behaviour (fine-scale assessment) during a trial rheotaxis when they approached and passed under the control (64.3 and 66.7%, respectively; data combined from both low and high turbulence treatments). Under control conditions, pass orientation was influenced by approach orientation (FET: P < 0.05), with S. trutta more likely to pass downstream facing the same direction as they approached. Pass orientation was not influenced by approach orientation when the flume was constricted.
Within the observation zone the majority of S. trutta (ca. 64%) displayed an orientation switch at least once (Table 3). The percentage of fish that displayed an orientation switch was not influenced by approach orientation, turbulence, or constriction (Table 3). The percentage of S. trutta that showed at least one rejection was influenced by constriction (X 2 (1) = 16.045, P < 0.001), but not by approach orientation or by turbulence, with a higher percentage doing so under the constricted treatment (85.7 v. 44.2%; Table 3).
Orientation switches tended to occur further away from the point of maximum constriction (R D ≈ 345 mm) than rejections (R D ≈ 85 mm; μ diff.: +262 mm, BCa CI: [154,374], t(36) = 4.492, P < 0.001; Figures 8-9). Salmo trutta appeared to exhibit orientation switches closer to the constriction under high (R D = 337 mm) v. low (R D = 360 mm) TKE, but the influence of turbulence on R D was not significant for either orientation switches or rejections (Figures 8-9).  ) and downstream (negative approach rheotaxis, ); aggregated data for constricted and control treatments. X, instances of right-censored data

| DISCUSSION
This study investigated the potential to use turbulence to reduce the discriminability of an accelerating velocity gradient by downstream moving S. trutta in an experimental flume in the absence of visual cues. During experimentation, S. trutta displayed behaviours typical of downstream moving salmonids, facing downstream in an unconstricted channel, but positively oriented towards the flow and displaying frequent rejections when encountering a velocity gradient, (Haro et al., 1998;Kemp et al., 2005;Enders et al., 2009b). Vowles et al. (2014) directly linked avoidance behaviours (e.g., rejections) with increased delay and distance travelled before downstream passage.
Similar links were evident in this study, with a lower proportion of passes and increased delay when the channel was constricted. When viewed from a coarse-scale perspective, signal discriminability was lower when turbulence was high suggesting masking of the signal occurred. However, the resulting increase in the percentage of fish that passed, decrease in time-to-pass and reduction in the distance at which S. trutta reacted (switched orientation) was subtle and non-significant. In addition, when viewed from a fine-scale perspective, signal discriminability was counterintuitively higher when turbulence increased suggesting masking of the signal did not occur. The results of this study support the growing body of evidence that velocity gradients can cause downstream moving fish to be delayed at anthropogenic infrastructure, places where they can suffer elevated risk of predation (Garcia de Leaniz, 2008) and highlight the need for effective mitigation measures to be developed. However, further work is required to assess whether turbulence can be used to effectively reduce discriminability of an accelerating velocity gradient.
Trout exhibited a negative response criterion when assessment was undertaken on fine scale behaviours, indicating that they were biased towards frequently expressing orientation switches and rejections even when the flume was not constricted. There was no clear reason as to why these behaviours were frequently exhibited. The hatchery fish used in this study displayed similar behaviours to those expressed by wild fish as they encounter a velocity gradient, such as orientation switches and rejections (Enders et al., 2009;Haro et al.,  Receiver-operating characteristics plot of hit rate against false-alarm rate for coarse ( , ) and fine ( , ) scale assessment of the behavioural responses of Salmo trutta to a velocity gradient under low ( , ) and high ( , ) turbulent kinetic energy. Reference discriminability (d' = 0, 1, 2, 3; ) and response criterion (c = −1, −0.5, 0, 0.5, 1; − --). Increases in d' represent greater signal discriminability and increases in c represent greater bias towards responding 1998; Kemp et al., 2005Kemp et al., , 2006. However, differences in the velocity gradient response thresholds of Chinook salmon Oncorhynchus tshawytscha (Walbaum 1792) (Enders et al., 2009) v. S. trutta  and in the rejection rates at a bypass entrance of Atlantic salmon smolts v. American shad Alosa sapidissima (Wilson 1811) (Haro et al., 1998), highlight the variability of these behaviours between species. It is possible that there is also variability in how readily these behaviours are exhibited between hatchery and wild fish and this warrants further investigation.
In this study, increased hydrodynamics noise only had a small and non-significant influence on key passage metrics. It is likely that either the signal was overly intense or the noise level was not high enough for effective masking to occur for all S. trutta (i.e. the signal to noise ratio was too high). Further reductions in discriminability, by reducing the abruptness of the velocity gradient or using greater intensities of turbulence, may be needed before biologically relevant shifts in passage efficiency occur. High levels of turbulence occur naturally in river systems (TKE > 20 J m −3 in the wake of pebble clusters in a shallow river; Lacey & Roy, 2007) and there is scope for higher levels to be used to try to mask hydrodynamic signals. However, turbulence rapidly dissipates after generation and inducing sufficiently high levels at an accelerating velocity gradient without physically interfering with fish movement may be difficult. The use of grids with greater bar width and spacing to generate larger scales of turbulence that permeate downstream further (Mohsen & LaRue, 1990) or using boundary roughness prior to and directly at the intake (Raupach et al., 1991) may provide a solution. Alternatively, other stimuli (e.g., sound, electricity or other hydrodynamic noise sources) might be employed to provide masking effects that reduce rejection rates. However, careful consideration of fish sensory systems is required to predict what the uni or multimodal effect of these stimulus sources might be (Vowles et al., 2014).
The mechanosensory lateral-line system has the appropriate anatomical distribution and physiological properties to identify regional differences in flow over different parts of a fish's body (Montgomery et al., 2000) and it probably plays a key role in how fish sense and respond to velocity gradients. The inner ear can also provide information on changes in flow by detecting whole-body accelerations (Pavlov & Tjurjukov, 1995;Sand & Karlsen, 2000) and fish may be able to gain additional information on the presence of a velocity gradient via this sensory system. It is possible that the inner ear plays a greater role in the discrimination of a velocity gradient under high compared with low levels of background hydrodynamic noise, as the relative sensitivity of the lateral line to regional flow differences diminishes (i.e., masking occurs). The contribution of the inner ear or the lateral line to detecting flow variations is likely to depend on species, stage of development and characteristics of the flow stimulus and warrants further investigation.
The findings of this study further highlight the importance of fish perception and behaviour in relation to hydraulic conditions encountered at downstream bypass facilities. Although, at a coarse scale, signal discriminability was lower when turbulence was high, suggesting masking may have occurred, high levels of hydrodynamic noise only resulted in a small and non-significant increase in the percentage of fish that passed, decrease in time to pass and reduction in the distance at which S. trutta reacted (switched orientation) to the velocity gradient. It is possible that the reduction in discriminability was not great enough to significantly improve passage or that alternative sensory information (e.g., from the inner ear) enabled fish to adequately detect the velocity gradient, despite effective masking of the signal to the mechanosensory lateral line. Further reductions in discriminability might be produced if higher levels of background hydrodynamic noise are utilized in future experiments. Despite mixed results, the use of masking to manipulate an animal's perception of stimuli in an anthropogenically modified environment is conceptually valid and the results of this experiment present a stepping stone for future research.