This work is still preliminary and has yet to undergo peer-review. However, I am always open to questions and comment on the research. Thank you for visiting!

Return to Main Webpage

Introduction

Burst swimming, typified by fast movements and short-distance and quick acceleration, is a common fish behavior. It is integral to predation encounters (chasing down prey, evading capture), spawning behavior and fish passage through hydraulic challenges. However, it is challenging to study, often requiring high-speed cameras and abundant time to analyze recordings. This time-burden prevents thorough investigation of the role of environmental characteristics (temperature, salinity etc.) on the physiological performance of burst swimming. We modified the “Fish drag-strip” described in Nelson et al. (2002) using a Raspberry Pi microcomputer. The main advantage of the raspberry pi is its ability to monitor up to 25 laser gates, while previous designs were limited to 8. The increase in the number of gates allowed us to gather more data upon each burst event and to measure new traits in burst physiology. The purpose of this project was to assess the accuracy of our design, determine the most appropriate metrics for calculating burst velocity and assess the relationship between burst performance and other measures of fish swimming physiology (e.g. UCRIT)

Materials

Tunnel Design

We constructed two tunnels, a large and a small. The large tunnel is ~2.0m x 0.2m x 0.2m. while the small tunnel is ~1.0m x 0.1m x 0.15m. Each tunnel is made of clear acrylic. Each tunnel has water pumped up into it via a sump system. The sump is plumbed to a heat pump allowing for precise temperature control over the burst tunnel system.

Small Burst Tunnel used for 2-5g Fish

Lasers and Detectors

Each tunnel is bisected by 25 line lasers. Each laser falls upon an array of light detectors. During an experiment, when a laser is broken, an electrical signal is sent to the Raspberry Pi which records the time when this signal arrives. The 25 lasers are arrayed with greater lasers density near the starting chambers. We also mirrored the arrangement of the lasers across the midpoint of the burst tunnel. This allowed us to burst fish form either end of the tunnel. Allowing fish to burst from either direction allowed us to minimize handling and increase the frequency of burst events.

Burst Tunnel Computer

The burst tunnel is managed by a Raspberry Pi and a custom python script which allows for user control over the burst tunnel. The output of this code is a dataframe reporting the timings for when each laser was broken during a burst event. These timings are then converted into velocity measures using R.

Methodology

Fish Husbandry

This experiment used yearling rainbow trout (Oncorhynchus mykiss) which were acquired from the Nimbus Fish Hatchery. Fish were acclimated at 18C and all experiments were conducted at 18C. Fish were individual marked using PIT (Passive Integrated Transponder) tags to keep track of individuals.

Experimental Timeline

Each individual fish underwent 5 experiments. The first was a UCRIT trial. That was followed by four burst trials. The first and third trials used the Delayed Burst Trial methodology while the second and fifth used the Fatigue Burst Trial methodology (described below). We repeated the two burst trial methodologies to assess how repeatable of a trait burst velocity was.

Experimental Timeline, each fish (n = 24) completed all 5 experimental trials.

UCRIT

We sought to compare our burst velocity measurements with a more accepted form of fish swim physiology, UCRIT, which quantifies the maximum sustained swimming speed of a fish. We conducted our UCRIT measurements using 30L Loligo swim tunnels. We exposed fish to an increasing water velocities (5cm/s increase every 5 minutes). When the fish ceased swimming and rested or impinged upon the rear screen, we would end the experiment and calculate UCRIT using the equation in Brett (1964).

Delayed Burst Trial

One experimenter was needed to complete this trial. Fish were placed into a receiving chamber of the burst tunnel (randomly selected) and then given 30 minutes to acclimate to the tunnel. Fish would then be busted by light caudal pinching and then confined to the receiving tunnel end, or coaxed there by the experimenter. Fish would be given 5 minutes to recover between trials before being bursted again. Fish were bursted 6-7 times.

Each trial required approximately ~60 minutes.

Fatigue Burst Trial

Two experimenters were necessary to complete this trial. One would manage the fish and the other would control the burst tunnel computer. Fish were placed into the receiving chamber of the burst tunnel (randomly selected) and then given 5 minutes to acclimate to the chamber. Fish were then stimulated by caudal pinching. The fish would then burst and be received into the other end of the tunnel. One the fish had completed its burst (1-5 sec) the experimenter managing the computer would advance the program to be ready for another burst event, while the experimenter managing the fish trapped the fish in the starting chamber and coax the fish to face the proper direction. This system allowed for repeated burst events to be started approximately every 20 seconds. Fish were bursted until they elicited 3 refusals to burst or until they had undergone 25 burst events.

An entire trial would take between 10-15 minutes depending on fish performance.

Processing Data

Processing the burst data requires a long sequence of steps to convert the timing data to final metrics for a given fish. The timing data leaves the Raspberry Pi as a .csv. With a column for each laser gate and the time at which that laser gate was broken. Camera data was converted to an identical format by counting the number of frames between when each laser was broken and then converting to timings using the camera’s reported framerate (239.76 fps). Once transformed into a dataframe of laser gate and timings, both datasets (Camera and Laser) can be processed in the same pipeline. I then generate all possible combinations of gates and calculate the timing between them. For instance, I would duration between when laser gate 1 and laser gate 2 were broken, laser gate 2 and laser gate 3, but also laser gate 2 vs. laser gate 4 etc. This ultimately results in 625 combinations of laser gates. I then compute the distance spanned by each laser gate combination and discard all gates which distances below 5cm or above 20cm. Gates distances below 5cm are prone to amplifying the effect of small timing errors and gate distances over 20cm report lower velocities due to averaging over a longer distance. I then calculate the velocity using the equation:

V = Distance / Time

For each remaining 84 pairs of laser gates. Any gates which report a velocity over 20 m/s (72 kph, 45 mph) are discarded as being erroneous. This process culminated in a dataframe of ‘instantaneous burst velocities’ for fish traversing different sets of gates. The resulting data set is then used to calculate several different metrics of burst speed. By selecting different instantaneous burst velocities or averaging groups of instantaneous burst velocities I explored 14 different burst metrics.

Representative example of a analyzed burst event. Each point is an instantaneous burst velocity, the x-coordinate of the point represents the distance from the start of the burst chamber at which that instantaneous burst velocity was measured. The horizontal error bars span the distnace overwhich that burst velocity was calculated, the left error bar extending to the distance where the first gate was broken and the right error bar extending to the distance where the second gate was broken. The dashed horizontal lines indicate values of different calcualted burst metrics.

My attempts at calculating burst metrics sought to fulfill several requirements:

  1. The metric had to be accurate when calculated using the laser measured timings and camera measured timings
  2. The metric had to be repeatable, in other words I sought metrics with high intraclass correlation coefficients. A high ICC supports the idea that burst performance is in part an intrinsic fish trait.
  3. The metric had to be fast, no purpose in using a metric which is severely underestimating a fish’s burst velocity

I calculated ~14 metrics burst metrics which can be broken down into 3 ‘types’ Peak velocities, averages and weighted averages Peak velocities are a single instantaneous burst velocity, one of the many measured segments. I recorded the first, second and third fastest instantaneous burst velocities. I additionally limited my survey to just the first half of the tunnel for reasons mentioned below.

Average velocities simply grouped different suits of instantaneous burst velocities together (e.g. top 3, top 5, top 10% etc.) Weighted velocities averages the same sets of instantaneous burst velocities but weighted those averages by the distance between the triggered gates.

Ultimately, the two methods of averaging reported very similar estimates of burst velocity.

I calculated these 14 metrics for each burst event of each trial for each fish (n = 1005 total). From this data set I could then investigate the fastest burst velocity elicited by a fish, but also their rate of fatigue and the variation in their performance across burst events.

Statistics

Statistical investigations were conducted in R (version 4.1.3). Models investigating repeatability used the R package lme4 to compute linear mixed effect models with a random effect for individual fish. This allowed for calculation of Intraclass Correlation Coefficients (ICC) as a measure of repeatability.

Results

Accuracy

The accuracy of the burst tunnel was determined by comparing the laser-timed velocities with velocities calculated using over-head high speed cameras (240 fps). We found that the burst velocities were quite accurate.

Accuracy of Laser Tunnel vs. High Speed Cameras. Each point represents an instantaneous burst velocity. Multiple points may come from a single fish or single burst event but calculated across different gates. The thick dashed line represents 100% accuracy between the timings generated by the lasers and those based upon the camera data. The dotted lines indicate a 10% error between the two measures. The red line is the modeled association between the two forms of speed measurement. Its slope is .99, r^2 = 0.96, indicating a high level of accuracy.

We did note that accuracy typically declines as the distances over which burst velocity is calculated decrease (i.e., accuracy is greater over a 10cm span than a 2cm span). Likewise faster velocities were more likely to produce disagreement between the camera and the lasers than slow velocities.

Burst Metrics

Burst Velocity

We calculated ~14 metrics of fish burst performance using different methods of averaging and weighting a fish’s set of measured instantaneous burst velocities. These different metrics yielded slightly different estimates for a fish’s burst performance. Using data from the laser tunnel compared against the data generated for the cameras I was able to isolate the metrics of burst velocity which were the most accurate and reported the highest velocities.

Comparisons of burst metric accuracy and speed. A) ar ethe burst metrics calcualted using data from the Delayed Burst Trials. B. are burst metrics using data from the Fatigue Burst Trials. Metrics are arranged in increasing ICC value which represents repeatability. Red metrics did not fulfill one or more of the requirements of a ‘good’ burst metric: velocity greater than 5 bl/s and ICC above 0.55.

Selection of the reported burst metric

Burst metrics varied in their repeatability between trials. The least repeatable metrics were the PEAK_SPEED and those which included the PEAK_SPEED as part of an average. This result is likely due to occasional timing errors which produce a fast, erroneous measurements. Burst metrics which incorporated more data (e.g. averaging all instantaneous velocities in the entire tunnel (AVE_SPEED), or in just the first half AVE_SPEED_FIRST_HALF) were the most repeatably, but also the slowest.

The selected metric for quantifying a fish’s overall burst velocity was PEAK_SPEED_3_FIRST_HALF_SL which translates to the 3rd fastest speed in body lengths per second elicited in the first half of the burst tunnel. We had to limit our observations to the first half of the tunnel due to some user-induced erroneous timings which occurred if fish had to be coaxed into the receiving chamber. Methodology could easily be adjusted to allow sampling across the entire tunnel length.

Overall Repeatability

In both the Delayed Burst and Fatigue Burst trials fish exhibited repeatability in burst performance. In other words, fast fish in one burst trial were likely fast fish in another.

Repeatability of fish burst velocity using a Delayed Burst methodology. Each point is an individual fish.

Repeatability of fish burst velocity using a Fatigue Burst methodology. Each point is an individual fish.

Fatigue Rate

Using data from the Fatigue Burst Trials, I was able to calculate the rate of performance decline over a fish’s repeated burst events. The decline in performance was modeled as 2-order polynomial function of burst event.

Modeling the rate of burst speed decline over repeated bursts for three individual fish. Red represents a fish’s first attempt while white represents its second.

Constructing interpretable models for estimating fish fatigue is a current focus of my work on this project. Still some ways to go.

Comparison with UCRIT

Relationship between Burst Speed and UCRIT. Each color represents a individual fish. The ‘x’ represent 1 of the four burst trials for a given fish. The circle represents the mean of a fish’s burst performance.

We found that UCRIT (maximum sustained swim speed) and burst velocity were not strongly correlated. Fast bursting fish were not necessarily fast sustained swimmers.

Overall Velocity

Swim Trial N Velocity (Lb/Sec) Mass (g) Standard Length (cm)
UCRIT 30 2.53 ± 0.52 234.37 ± 55.86 23.98 ± 1.85
Delayed Burst #1 28 5.29 ± 1.65 218.15 ± 53.20 24.09 ± 1.71
Delayed Burst #2 26 5.63 ± 2.06 235.45 ± 66.35 24.87 ± 1.76
Fatigue Burst #1 26 6.12 ± 1.52 237.96 ± 63.64 24.68 ± 1.76
Fatigue Burst #2 24 6.04 ± 2.91 246.03 ± 68.55 24.99 ± 2.07

Our results indicate that a rainbow trout burst at ~ 6 Lb/sec. This velocity is lower than that reported in experiments using camera-based fast-start methodology (7.3L/s) (Domenici et al. 1997) or burst events elicited in a swim tunnel (6.5 L/s) (Osachoff et al. 2014). Observed burst velocities are greater over shorter distances or when monitored over longer periods of time which may explain some of the difference between our method and fast-start or swim tunnel methods respectively.

Conclusions

Comparing Burst Velocities among Predators and Prey

Dr. Alex McInturf and I conducted an experiment to assess how environmental temperature impacts trophic interactions among juvenile salmon and common predators in the California Central Valley. We found that as temperatures increased, the ability of juvenile salmon to repeatedly burst decreased. Meanwhile predators were more capable of repeated bursts.

Effects of Food Rationing on Burst Velocity

Cassidy Cooper (UC Davis, PhD(c)) and I are conducting an experiment to assess how food availability and energy restriction impact fish thermal physiology. Check out Cassidy’s Talk entitled Influences of Thermal Variation and Feed Restriction on Growth and Thermal Physiology of Early Life-stage Chinook Salmon (Oncorhynchus tshawytscha) on Friday at 8:50 to hear more about this project.

Effect of Pesticides on Burst Velocity

Dr. Anna Steel (UC Davis) is using the burst tunnel to assess how exposure to a Bifenthrin pesticide impacts juvenile green sturgeon mobility.

Potential for Field Deployment

Our burst tunnel design is scalable and could be designed to be deployed in a field setting allowing for physiological assessment of burst performance on wild fish.

Works Cited

  1. Brett, J. R. Journal of the Fisheries Research Board of Canada, 21, no. 5 (1964)

  2. Nelson, J.A, P.S Gotwalt, S.P Reidy, and D.M Webber. Comparative Biochemistry and Physiology Part A: Molecular & Integrative Physiology 133, no. 2 (2002)

  3. Domenici, Paolo, and Robert W Blake. The Journal of Experimental Biology, 200, (1997)

  4. Osachoff, H. L., K. N. Osachoff, A. E. Wickramaratne, E. K. Gunawardane, F. P. Venturini, and C. J. Kennedy. Journal of Fish Biology, 85, no. 2 (2014)

Return to Main Webpage

or

My other ICBF Poster