The hydrodynamic trails of fish belonging to the families Centrarchidae, Tetraodontidae and Cichlidae were investigated. Water movements were measured in six horizontal planes, spaced 10–12 mm apart, for up to 5 min after the passage of a fish, using a computer controlled array of modulated laser diodes. We measured continuously and non-continuously swimming fish. Water velocities decayed rapidly in the leading seconds after the passage of a fish, but could still be measured for a period considerably longer than that. In still water (median water velocity <0.5 mm s–1), the hydrodynamic trails of Lepomis gibbosus lasted for more than 5 min. The trails of Colomesus psittacus and Thysochromis ansorgii could be detected for more than 30 s and more than 3 min, respectively. The water disturbance left behind by these fish was sufficient to be sensed by a piscivorous predator at a distance where vision or hearing frequently fail. Acoustic stimuli estimated from a dipole model in a distance that would be covered by the tested fish in 1 min (4–25 m) were 1.5×10–7 to 3.1×10–10 m s–2, while the hearing threshold of a perch is three orders of magnitude above that. By contrast, the fish wakes after 1 min (except for one Colomesus wake) contained water velocities between 0.95 and 2.05 mm s–1, which are within the detection range of hydrodynamic sensory systems. The three species differed with respect to water velocities, the spatial extent of the fish-generated water disturbances and the structure of the wake.
- wake following
- lateral line
- hydrodynamic reception
- particle image velocimetry (PIV)
Most aquatic animals have hydrodynamic receptor systems (Bleckmann, 1994). They use these systems for rheotaxis (Baker and Montgomery, 2002), the detection of surface waves (Bleckmann et al., 1989), and the detection of midwater hydrodynamic events such as those caused by predators, conspecifics or prey (Bleckmann, 1994). Harbour seals Phoca vitulina can track hydrodynamic trails of moving objects with their vibrissae over a distance where vision and hearing should fail (Dehnhardt et al., 2001). European catfish Silurus glanis can follow the swim paths of their prey, which suggests hydrodynamic or chemical trail-following (Pohlmann et al., 2001). Despite recent advances in the investigation of the behavioural functions of the lateral line and the peripheral and central processing of hydrodynamic sensory information by fish (e.g. Bleckmann et al., 2001), and the sensory abilities of seals (Dehnhardt et al., 2001), data on the information contained in animal-caused water motions, i.e. their frequency content, three-dimensional extension and especially their ageing, are still rare.
In this study, we used scanning digital particle image velocimetry (S-DPIV) to measure the hydrodynamic trails caused by swimming fish of three teleost species. While classical digital particle image velocimetry (DPIV) measures velocities in a single layer of fluid illuminated by a laser light sheet (Adrian, 1991; Westerweel, 1997; Drucker and Lauder, 2001, 2003), S-DPIV measures velocities in multiple layers using one measurement, by scanning the laser light through various planes. The result is an extension of the velocity information from a single layer to a volume.
DPIV has been applied to the water motions caused by moving animals (Stamhuis and Videler, 1995; Müller et al., 1997, 2000; Drucker and Lauder, 2000, 2001). Drucker and Lauder (1999) reconstructed three-dimensional information from successive two-dimensional PIV measurements. Nauen and Lauder (2002) measured three-dimensional velocity information in a water layer behind a swimming trout using the stereoscopic information from two high-speed cameras. Drucker and Lauder (2003) investigated the flow caused by salmonid pectoral fins using DPIV. The above studies focussed on the function of fish fins and body movements, their role in propulsion, steering and braking and the fluid forces. None of these studies attempted to investigate the long-term development of fish-generated wakes.
Hanke et al. (2000) showed that the hydrodynamic wake of a swimming fish can last for up to 5 min. Measurements were confined to a single layer and only one fish species, the goldfish Carassius auratus, to investigate: (1) how general the long duration of fish wakes is across species, (2) whether different fishes have different wake signatures that could possibly be discriminated by predators and (3) if the detection of hydrodynamic trails can be comparable or even superior to the use of other sensory systems operating in the dark, especially the acoustic system.
In this study, we used S-DPIV to measure and compare the long-term development of the wakes caused by three species of fish: the sunfish Lepomis gibbosus, the puffer Colomesus psittacus and the cichlid Thysochromis ansorgii. Lepomis gibbosus lives in the open water and in the reed zones of still temperate waters (Riehl and Baensch, 1991). Swimming manoeuvres include tail-propelled swimming and fast accelerations, although much of the locomotor time budget is accomplished by the pectoral fins. Colomesus psittacus lives in warm still water (Riehl and Baensch, 1991) and shows the typical tetraodontiform (Lindsey, 1978) swimming mode of a puffer, only in some cases the tail fin is used for propulsion. Thysochromis ansorgii is found in the peripheral zone of still and running waters (Riehl and Baensch, 1991). This fish tends to swim calmly with frequent use of the pectoral fins.
Materials and methods
Two fish of each species Lepomis gibbosus L., Colomesus psittacus (Bloch & Schneider) and Thysochromis ansorgii (Boulenger), were obtained from commercial dealers and kept in standard aquaria according to the instructions in Riehl and Baensch (1991).
Experimental set-up and fish training
Two experimental tanks were used. Tank 1 had a floor surface of 100 cm×100 cm to give ample space for a lateral spread of the fish's wake. Tank 2 had a floor surface of 40 cm×100 cm to facilitate the laser illumination of the particles in the field of view and thus improve image quality. Image quality was highly dependent on the number of seeding particles that the laser light had to pass before it reached the area of interest. In both tanks the water level was set to 40±3 cm.
Individual fish were trained to swim on a straight line through the centre of the experimental tank to reach a goal compartment at the opposite wall where they received a food reward. Before a measurement the fish was kept in a small compartment at the start of the swimming route to allow the water in the experimental tank to calm down for at least 5 min. In training sessions the fish were conditioned to swim to a green light (a bright LED flashing at approximately 2 Hz). The light was mounted in the goal compartment at the lower end of a feeding tube, through which the fish was rewarded with a mosquito larva. As soon as the fish reached the goal compartment, it was locked out from the measurement area in the middle of the tank by shutting a sliding door.
To monitor the altitude of a swimming fish in side view, a third camera (camera 3) was positioned beside the experimental tank. The experimental tank was illuminated with green light to aid fish orientation. To avoid impairing the pictures taken by cameras 1 and 2, their object lenses were equipped with red filters. In initial experiments, a dim halogen lamp illuminated the fish while it was in the field of view, allowing to regain its kinematics from camera 1. In later experiments, an additional camera with a green filter (camera 4) was mounted above the experimental tank to improve image quality for kinematic analysis. Camera 4 was also synchronised with the light sheet illumination. Cameras 1, 2 and 4 were type DMK803 (The Imaging Source, Bremen, Germany); camera 3 was a surveillance camera (Conrad Electronic, Hirschau, Germany).
A custom-made PIV device was used to measure the flow around and behind the freely swimming fish. Tracer particles (Vestosint 1101, Degussa AG, Marl, Germany) were seeded into the water and illuminated in six horizontal light sheets generated with modulated laser diodes (wavelength 650 nm, output power 50 mW) (Fig. 1). Vestosint is a synthetic material of density 1.02 g mm–2. Median particle size was 74±5 μm (95%<147±5 μm). The light sheets were spaced equidistantly (vertical distance 10 mm for Colomesus and 12 mm for Lepomis and Thysochromis). The laser diodes were modulated using a micro-controller (Motorola 68HC05, Schaumburg, IL, USA) so that only one of the six planes was illuminated at a time, and the illuminated plane was switched to scan the water volume once in 12 video frames (equivalent to once per 0.48 s). Particle images were taken with one or two CCD cameras (termed cameras 1 and 2) mounted above the tank. Each camera had a chip size 768×582 pixels, frame rate was CCIR standard (50 half frames s–1). The beginning of each half frame taken by the cameras was synchronised with the light sheet illumination. There was no need to adjust the focus of the cameras as light sheets were switched because the distance between the cameras and the light sheets exceeded tenfold the longest distance between light sheets. Pictures were stored on S-VHS video recorders (Philips VR1000, Eindhoven, The Netherlands) together with a synchronisation signal that served to determine the light sheet for each picture. Video frames were digitised using an MJPEG computer board (Miro Video DC30, Pinnacle Systems, Mountain View, CA, USA). The MJPEG board was set to a compression factor of 1:12, which does not affect the PIV results (Freek et al., 1999).
Since the cameras could only be synchronised to the start of a half frame, not a full frame, half frames were, if necessary, rearranged using Delphi 2.0 (Borland, Scotts Valley, CA, USA). Particle images were analysed using custom-made programs in MatLab 5.1 (The Mathworks, Natick, MA, USA). Analysis of particle displacement followed the principles of digital particle image velocimetry (Willert and Gharib, 1991), but was improved using the spurious vector detection technique described by Hart (2000). The interrogation area (the subimage that is subjected to the correlation procedure) was 32×32 pixels with an overlap of 50%, resulting in vector fields of 43×31 vectors. The images of the fish's silhouette from cameras 1 and 3 or cameras 4 and 3, respectively, were analysed manually using Scion Image (Scion Corporation, Frederick, MD, USA).
The analysis of fish-generated water disturbances is complicated by several specific problems. Firstly, water flow may be highly divergent and vortex structures may be small compared to the field of view. This makes the application of standard data post-processing procedures (e.g. Høst-Madsen and McCluskey, 1994; Westerweel, 1997; Hanke et al., 2000) undesirable. Where image quality was sufficient, which was the case in the narrow water tank (tank 2), we discarded post-processing of our velocity data. Velocity vectors were validated using three criteria. (1) The peak position criterion suggested by Hart (2000). Vectors were not used when the position of the first correlation peak in the product of two adjacent correlation planes differed from its position in each of the planes by more than 0.1 times the width of the interrogation area. (2) A peak height criterion. Vectors were assigned a status depending on the height of the correlation peak relative to its surrounding. (3) A velocity criterion. Velocity vectors that corresponded to a particle displacement of more than 0.3 times the width of an interrogation area and were out of the velocity range expected from comparison with similar measurements were discarded.
Secondly, the dynamic velocity range is very broad. The ratio of the highest to the lowest water velocities at a given point in time may easily reach an order of several hundred. This problem was solved by multiple-time-scale processing. Different time scales were not only used for different stages of the ageing of the water disturbances (cf. Hanke and Brücker, 1998), but also for different locations at a given point in time. Analysis started with a time spacing of 12 frames for each interrogation area, followed by a time spacing of 1.0 and 0.5 frames, depending on velocity and vector status information. Time spacing between 12 frames and 1 frame was not possible because of the scanning procedure (see above).
Thirdly, the fish moving through the camera's field of view is a foreign body, which can lead to false velocity information when using correlation techniques. This problem was solved by manually removing false vectors caused by the fish or its shadow using custom-made programs in Delphi 2.0 (Borland). Because our study focused on the long term development of the wake, the fish was in the field of view in less than 2% of the images.
Fourthly, in still water the fish's movements are generally not reproducible from trial to trial. This problem was overcome by the scanning technique, which allowed derivation of three-dimensional information from a single trial.
We compared the width of the trails caused by the three fish species using a U-test (e.g. Sachs, 1997) and the distribution of water velocities in the trails by comparing two trail indices TI1 and TI2, following the principles of discriminant analysis (e.g. Deichsel and Trampisch, 1985).
Here we present four hydrodynamic trails of Lepomis gibbosus, three trails of Colomesus psittacus and three trails of Thysochromis ansorgii measured in the narrow tank (tank 2). In addition we describe two trails of Thysochromis ansorgii measured in the broad tank (tank 1). To evaluate the 12 trails over a time span of 5 min, a total of 63 000 video frames was analysed.
In Fig. 2, the silhouette of a Lepomis is shown as it moved through the field of view. For clarity, the fish silhouettes were shifted laterally; the curved line connecting circles in the centre of the figure indicates the position of the fish's head in successive images.
The water movements caused in this trial are plotted as velocity vector fields and the corresponding divergence in Fig. 3A,B (A, after approximately 10 s; B, after approximately 60 s). The time of each vector field is indicated in its upper right corner; t=0 s is the time when the fish entered the field of view. Times of different vector fields are slightly different because of the scanning procedure. The number of the illuminated layer is indicated in the upper left corner of each vector field, where 1 designates the uppermost, 6 the lowermost layer (spacing between layers was 12 mm). Velocity is indicated by the 10 mm s–1 and 2 mm s–1 scale bars in A and B, respectively, and the spatial extent by the 100 mm scale bars. Divergence, represented by different colours (see colour bar), is defined as (d/dx) vx+(d/dy) vy and indicates the flow out of or into a small section of the plane (x and y are the cartesian coordinates, vx and vy are the x and y components of the velocity). We show the divergence because it gives an impression of the out-of-plane water movements that could not be measured directly in this study, but should be relevant to a predator's sensory system (Bleckmann, 1994).
It is apparent from this example that the trail of Lepomis gibbosus can show a clear vortex structure for at least 60 s (Fig. 3B). Vortices have slightly grown after 60 s compared to those after 10 s (Fig. 3A), but are still in the length scale of the wake generator's body.
Fig. 4A–C shows the development of water flow characteristics over 60 s for this Lepomis trial (Fig. 4A), for a Colomesus trial (Fig. 4B) and a Thysochromis trial (Fig. 4C). Maximum water velocity (top), mean water velocity (middle) and maximum amount of vorticity (bottom) are shown. Vorticity is plotted because it is associated with velocity gradients that fish can detect with their lateral line (Bleckmann, 1994). In each plot, the values for all six laser light sheets are shown in different colours. The fish silhouette with horizontal lines indicate the position of the fish relative to the light sheets.
It is apparent from Fig. 4A–C that the maximum velocity (top) and vorticity (bottom) in all three trials decayed rapidly within the first 10 s, but nevertheless persisted considerably longer than that. The same was true for the mean velocity in Lepomis (4A, centre). Mean velocities in Colomesus (4B, centre) and Thysochromis (4C, centre) decayed even slower (relative to their starting values, which were lower than in Lepomis). Differences between the various layers could be substantial for maximum and mean velocity as well as for maximum vorticity (Fig. 4A–C).
Table 1 shows water velocity values in representative layers and a description of the fish's movements for each of the 12 trials that were evaluated. The amounts of water velocity before the trial, after 5 s and after 60 s (where t=0 s is the time when the fish entered the field of view) are given as median, maximum and upper and lower quartile. tend in Table 1 designates the time when the upper quartile returned to starting conditions; velocity values are given for this time or for 300 s, respectively.
The spatial extent of six fish trails measured in tank 2 in representative layers is shown in Fig. 5. To visualise the development of the trails, the amount of water velocity was averaged over the columns of each vector field and the rows resulting from this procedure were assembled in temporal order. Water velocity is colour coded. Note that the colour scale does not cover the full range of velocities (see Table 1 and Figs 3, 4) in order to resolve the low velocities in the aged trail.
It is apparent from Fig. 5 that the spatial extent and temporal structure of the trails from Lepomis and Thysochromis can be distinguished from the Colomesus trails in their lateral spread. The portions of a fish wake that spread laterally are mainly produced by undulating body movement, lateral tail flicks and pectoral fin movements, while Colomesus mainly used an tetraodontiform swimming style (dorsal and anal fin undulations). Accordingly, the Colomesus trails show essentially one narrow zone of water disturbance, while the trails of the other two species divide in two or more branches. Water velocities caused by the small Colomesus were lower than the water velocities caused by the other two species (note the different velocity scales).
Table 2 adds information on the lateral spread for the five layers that are not shown in Fig. 5 and for the remaining trials. The width of each fish wake at t=10 s and t=20 s in all six layers is given. It is defined as the width of the zone where the water velocity calculated as in Fig. 5 was at least 0.8 mm s–1. If the trail reached the border of the field of view in Lepomis, lower boundaries are given (e.g. width >262 mm), and in Thysochromis, trail width was estimated by extrapolation (marked with *). In trials 4, 11 and 12, trail width was estimated as twice the width on one side of the fish because swim paths were close the border of the field of view. Where no width was calculated, water velocity did not reach 0.8 mm s–1. Comparing the maximal width (replaced by its lower limit if the border of the field of view was reached) in all the trials shows a clear difference between species in these trials (U-test, α=0.05).
The trails of Lepomis, Colomesus and Thysochromis can be distinguished visually by the distribution of velocity over their cross section (Fig. 5). The Lepomis trails appear more sharp-contoured compared to the more diffuse Thysochromis trails. It must be noted that in these trials Lepomis performed exclusively fast swimming manoeuvres (see Table 1 and Discussion).
Table 3 supports this visual impression numerically. We defined two trail indices, TI1 and TI2. To calculate TI1 and TI2 for each trial from tank 2, the vector field at t≈10 s was reduced to a row by averaging the amount of velocity over its columns as in Fig. 5. The result was a curve c(x) that showed the average water velocity as a function of lateral position x at t≈10 s. To make this comparison independent of the velocity differences between species, c(x) was normalized by setting its maximum velocity value to 1, yielding a curve c1(x). The curve c1(x) was then smoothened by a moving average filter (width n=29), resulting in a curve c2(x). TI1 is the number of intersections of c1(x) and c2(x). TI2 is the maximum of c2(x)–c1(x). Both TI1 and TI2 were averaged over all laser light sheets that touched the tail fin of the fish in the centre of the field of view.
The trail indices from Table 3 are plotted in Fig. 6. The trails of different species tend to form clusters, with the Colomesus cluster rather distinct from the other species. Only one Lepomis trail with TI1=3, TI2=0.356 comes close to Colomesus. It must be noted that in this Lepomis trial, the fish did not swim through the centre of the field of view but close to its border, so that a considerable part of the wake was lost for analysis. The wakes of Lepomis and Thysochromis are not intermingled; however, intraspecific distance can be higher than interspecific distance. Discriminant analysis and the estimation of error rates using the leaving-one-out-method (L-method, e.g. Deichsel and Trampisch, 1985) yields an error rate of 20%. This is reduced to 12% if the Lepomis trail with TI1=3, TI2=0.356 is omitted.
Our measurements of water disturbances generated by swimming fish yield insight into the flow patterns caused by three species that differed in swimming style. It was found that, similar to the goldfish trails (Hanke et al., 2000), the water disturbances caused by these species can last in the order of minutes (or at least for about 30 s in the case of Colomesus), even in water with thermal convection currents (see Table 1).
Selection of tank and tank width
Animal-generated flow is often difficult to study because it is usually neither reproducible nor stationary or two-dimensional. To achieve reproducibility, fish were forced to swim in a flow tank (for references, see Drucker and Lauder, 2003). This method is appropriate if one is interested in the structure of the wake measured close to the fish. However, if one is interested in the ageing of fish-generated wakes, flow tanks are usually too short. Furthermore, they add a certain degree of turbulence to the water that can override weak flow structures. The use of S-DPIV in still water copes with these problems.
All of our measurements were therefore done in still water tanks. To improve image quality (see Materials and methods) and thus reduce the number of unreliable vectors, which in our wide tank could be as high as 25% compared to 1% in the narrow tank, most measurements were performed in the narrow tank (tank 2). The disadvantage of the narrow tank is that its walls may have restricted the lateral spread of the fish's wake or changed its structure. In Lepomis, a portion of the wake usually reached the border of the field of view, equivalent to two thirds of the width of tank 2, in a short time (Fig. 5A). However, the clear vortex structures of the fish wakes even after 60 s in Lepomis that can be seen in Fig. 3 and the lateral spread of the wakes of Thysochromis (Fig. 5C, narrow tank) compared with the data from the broad tank (Table 1) led us to believe that the wall effect can be small in some swimming manoeuvres. On the other hand, a lateral spread of a goldfish's wake that exceeded 50 cm has been reported elsewhere (Hanke et al., 2000). Hence tank width should always be adapted to the specific question asked.
Hydrodynamic wake detection
Our data show that even small fish produce wakes that can indicate the presence of the wake generator for at least 0.5–5 min, depending on body length and swimming style. This makes fish-generated wakes a potential source of information for piscivorous fish (cf. Pohlmann et al., 2001) and mammals (cf. Dehnhardt et al., 2001). While wake-following behaviour in the viscous length scale of planktonic organisms has been attributed to chemoreception (Yen et al., 1998), in the swimming fish range of Reynolds number the mechanical component of the wake is an additional cue that fish cannot easily avoid.
Why should the detection of hydrodynamic trails with mechanosensory systems be comparable or even superior to the use of other sensory systems operating in the dark, especially the acoustic system? Consider a prey fish that swims in a manner similar to the Lepomis in our trial 2 (Table 1). With an average speed of 44.6 cm s–1 (5.2 body length s–1; Table 1), it can cover a range of more than 25 m within 1 min. Thehydrodynamic trail after 1 min still contains velocities higher than 1.7 mm s–1 (Table 1), which are above the detection threshold of the lateral line (Görner, 1963). To estimate the acoustic field produced by a small fish 25 m away, we substitute the tail fin by a dipole and use the dipole equations quoted by Kalmijn (1988). This approximation is justified as long as the fish does not change its volume, i.e. the monopole moment is zero; the quadrupole and higher moments fall off with the distance much faster than the dipole moment. Assuming a dipole radius of 10 mm, an oscillation amplitude of 10 mm and an oscillation frequency of 5 Hz, the water velocity produced by the dipole at a distance of 25 m is as small as 1.8×10–11 m s–1, and the corresponding acceleration amplitude is 3.1×10–10 m s–2. The hearing threshold of a typical piscivorous predator, the perch Perca fluviatilis, is in the order of 10–4 m s–2 (Karlsen, 1992). Thus while the hydrodynamic sensory system most probably responds to the hydrodynamic trail, acoustic perception of the prey is impossible in this example. The same calculation with the Thysochromis data from trial 12 (see Table 1, lowest average swimming speed of all trials presented – this may be closer to the fish's routine speed than the Lepomis speeds are) leads to a distance of more than 4 m covered by the fish in 1 min. The dipole in our model causes an acceleration amplitude of 1.5×10–7 m s–2 at a distance of 4 m, which is again well below the hearing threshold of the predator. The hydrodynamic trail contained water velocities of 0.96 mm s–1 and would most probably be sensed.
We do not know how often the high swimming speeds of Lepomis observed in our experiments (average up to 5.2 BL s–1, where BL is body length) occur in nature. Due to experimental constraints Lepomis, like all other fish, was trained to swim to a flashing light where it expected a food reward. In this situation, high swimming speeds represented the spontaneous behaviour of our Lepomis. In nature the upper range of swimming speeds may well be reached in various situations including defending a breeding or feeding territory, reproductive interactions or predator–prey interactions. A 10 cm goldfish Carassius auratus can reach 11.4 BL s–1 for 0.1 min (Tsukamoto et al., 1975). In addition, the results from the fast-swimming Lepomis give a first estimate for the wakes of fast-swimming fish that have not been investigated yet. Prolonged speeds in a 25 cm herring Clupea harengus were 5.5 BL s–1 for 3 min (He and Wardle, 1988; Videler, 1993).
It has not been shown that the hydrodynamic wake of a prey fish unequivocally reveals the species or the swimming style to a predator. However, in the examples presented here, wake patterns were diverse (Figs 3, 5, 6), while vortex structures in the length scale of the flow-generating structures were observed (Fig. 3). Since many hydrodynamic sensory systems measure water flow in multiple points (Bleckmann, 1994), it is likely that a predator can extract information beyond the mere presence of a wake and learn to interpret such flow structures to a certain degree.
We thank Dr G. Dehnhardt for supplying part of the video equipment. The work reported herein was supported by a grant of the Deutsche Forschungsgemeinschaft to H.B. (Bl 242/9-1).
- © The Company of Biologists Limited 2004