Strategies to deter Pacific Herring from aquatic farm infrastructure

ABSTRACT

Objective

Seaweed aquaculture can benefit coastal ecosystems through the provisioning of new habitat. However, not all aquaculture–ecosystem interactions are created equal: Some interactions can result in substantial crop loss and potentially reduce the fitness of wild organisms, as is the case with Pacific Herring Clupea pallasii spawning on kelp farms in Alaska. This study drew upon herring physiology and ecology to identify potential nonlethal methods that aquatic farmers could implement to deter herring from aquaculture infrastructure while also minimizing impacts to marine mammals.

Methods

We exposed Pacific Herring to various stimuli (bubbles, active acoustics, lights, and suspended objects) in a net-pen and tracked fish movement using a Kongsberg-Mesotech Flexview multibeam sonar. The difference in the distance of individual fish from the stimulus when it was on and off was analyzed using linear mixed-effects models, and the effect size of each treatment was determined by calculating Cohen’s d.

Results

Across the methods tested, only bubble curtains elicited a clear behavioral response in Pacific Herring Clupea pallasii based on the effect size of the treatment. Acoustic pingers and strobing lights had no effect on the location of herring in the experimental space, while suspended moving or stationary fishing flashers either had no effect or attracted herring.

Conclusions

Although Pacific Herring demonstrated a behavioral response and avoidance of bubble curtains, deploying this technology at remote farm sites will be a challenge. Further research into developing low-cost strategies to minimize negative interactions between aquaculture and wild organisms is necessary for the sustainable growth of the industry.

INTRODUCTION

Shellfish and macroalgae aquaculture operations are often located in nearshore coastal zones and can create habitat through the introduction of midwater structure and, in the case of macroalgae, can offer a source of food for herbivores and some filter feeders (Theuerkauf et al., 2022). This novel structure can also provide a substrate for fish and invertebrates to complete phases of their life cycles, such as laying eggs or settling out from the planktonic stage. Interactions between aquaculture infrastructure and the nearshore ecosystem in the form of habitat provisioning can have different impacts on both wild and cultivated organisms. The interaction can be positive, resulting in an increase in suitable substrate for the wild organism or a potential increase in biologically available nitrogen for the cultivated species (e.g., Parker & Bricker, 2020); it can be benign, in the form of a change in available substrate or no impact on the cultivated species; or it can be negative, whereby the introduced habitat reduces the fitness of the wild organism or results in a decrease in the quality or quantity of the cultivated species (Barrett et al., 2019). Understanding the result of these ecological interactions is critical for effective regulation of nearshore aquaculture to ensure that it serves as a sustainable food source and contributes to the economic growth of coastal communities.

The Pacific Herring Clupea pallasii is one of many commercially, ecologically, and culturally important species that use nearshore substrate to lay their eggs. From southern California to the Beaufort Sea in Alaska, United States, Pacific Herring spawn in mass aggregations during the spring and deposit their eggs on macroalgae and other substrate in the nearshore and intertidal environment (Scattergood et al., 1959). In Alaska, these mass-spawning events support important traditional and commercial herring roe-on-kelp fisheries and can extend along more than 300 km of coastline in Southeast Alaska (Hebert, 2012). Spawning of Pacific Herring in the Gulf of Alaska occurs between March and May (Hebert, 2019), which overlaps with the kelp aquaculture harvest season.

Growth of the aquaculture industry in Alaska introduces increasing opportunities for interactions between aquatic farms and spawning herring. The leasing and permitting process in the state of Alaska tries to avoid siting prospective farms within known herring spawning sites; however, herring can unexpectedly expand to new areas or revisit areas that had not experienced a spawning event in years, decades, or longer (Hay et al., 2009; Pritchett, 2006). These spawning forays to new locations are thought to occur in response to high population abundance, instances of intense predation, and changes in feeding conditions (Brown & Norcross, 2001; Hay et al., 2009). Although the herring roe-on-kelp fishery is valuable, Alaska permit conditions preclude the harvest of cultivated kelp once herring have laid eggs on the kelp; therefore, a spawning event on an aquatic farm can result in a total loss of product (Permit Conditions, 2021). The fate of herring eggs on farmed kelp is unknown but may be less favorable than in wild kelp habitat because as the kelp senesces, it falls off the suspended lines and sinks into deep water, where the eggs may have a lower probability of survival (Rooper et al., 1999).

To reduce the potential for interactions between aquatic farms and spawning Pacific Herring, there is a great need to identify nonlethal deterrence strategies. We chose deterrents to test based on previous studies of the same or related species as well as known characteristics of the ecology and physiology of Pacific Herring. We also prioritized materials that would be easy for a kelp farmer to acquire and deploy. For any deterrent to be used in practice, it must also comply with the Marine Mammal Protection Act (1973), the Endangered Species Act (1973), and other statutes and regulations governing use of the coastal environment. Consequently, the acoustic deterrent device tested in this study operated below 160 dB to avoid unintentionally harassing marine mammals with intermittent noise (National Oceanic and Atmospheric Administration [NOAA], 2005).

Based on these criteria (herring ecology, ease of acquisition, and regulatory compliance), we identified four major categories of deterrents to test: bubbles, active acoustics, lights, and suspended objects. Bubbles as deterrents take advantage of the ecology of Pacific Herring by mimicking the bubble-­netting behavior of humpback whales Megaptera novaeangliae, which cooperatively create curtains of bubbles to trap herring for easy capture (D’Vincent et al., 1985). Pacific Herring have been shown to respond similarly to artificial bubbles and avoid crossing bubble curtains (Sharpe & Dill, 1997). Extensive work has been done on the response of clupeids to sound (Putland & Mensinger, 2019), including vessel noise (reviewed by Mitson, 1995; Wood, 2011), sonar (Schwarz & Greer, 1984), humpback whale calls (Schwarz & Greer, 1984), odontocete echolocation (Wilson & Dill, 2002), high-frequency sound around hydropower dams (Nestler et al., 1992), and tones at various frequencies and intensities (Handegard et al., 2016). The consensus of these studies is that herring respond more to frequencies below 5 kHz but that habituation to sounds is common. Strobing white light is used as a bycatch reduction strategy in various fisheries (Hannah et al., 2015; Yochum et al., 2022), although evidence is mixed regarding whether herring are attracted to or deterred by light (Dragesund, 1958). Finally, the use of suspended objects as a deterrent was meant to engage the avoidance and startle responses of Pacific Herring, which have been observed in interactions between herring and vessels or purse seines (Misund, 1990), midwater trawls (Suuronen et al., 1997), and simulated predator encounters (Rieucau, De Robertis, et al., 2014).

METHODS

This study was conducted in June 2022 at the National Marine Fisheries Service’s Little Port Walter Biological Field Station, located on the southern tip of Baranof Island along Chatham Strait (56.382678, −134.649883). Approximately 550 adult Pacific Herring were collected in Little Port Walter by using herring jigs and in Mist Cove (56.517065, −134.669428) by using a beach seine. Herring were held in one 216-m³ net-pen (6 × 6 × 6 m) with a 1-cm² mesh size for the duration of the experiment. The net-pen was located in a sheltered cove with no wave energy and high tidal exchange (2.90–5.03 m during the study period; NOAA Station 9451054).

The use of a net-pen was chosen over in situ field manipulations due to both legal and logistical limitations of working with spawning herring. Legally, while we may have been able to deter herring from spawning on suboptimal habitat (e.g., aquatic farms), we were not permitted to deter herring from known spawning areas, and such actions would have the potential to disrupt important activities related to subsistence harvest of the herring roe (Permit Conditions, 2021). Furthermore, herring spawning locations can change unpredictably from year to year, and it is unknown what drives those changes. Therefore, any field testing would require unrealistically large levels of replication, with the associated replication of expensive equipment and trained personnel. Spawning events on kelp farms are unforeseeable due to herring behavioral ecology. They are likewise intentionally limited by permitting requirements, which specify that farms must be located outside of known spawning areas. Therefore, it would be extremely difficult to determine whether herring deterrence from kelp farms was due to the deterrents tested or herring naturally spawning elsewhere. For these reasons, we decided to conduct our study using in-water net-pens.

Pacific Herring behavior was monitored using a Kongsberg-Mesotech Flexview multibeam sonar, which was operated at 950 kHz and mounted 2 m below the water’s surface in a corner of the net-pen, the recommended depth for kelp aquaculture (Flavin et al., 2013). The sonar was set to maximum pulse transmission (about 10 Hz) and resulted in near-video quality images, with pixel resolution representing about 0.4 × 0.4 cm. Acoustic beam dimensions were 140° horizontal × 30° vertical and recorded raw data at a range of 0.2–7.5 m from the sonar. Sonar data were observed in real time, and the dimension of the beam resulted in a full view of the net-pen except for a very small section in the corner behind the sonar (Figure 1).

All experiments were conducted in the net-pen in a paired design consisting of 30 min of exposure to a deterrent that was suspended in the pen (“on”), followed by 30 min of observation of behavior after the deterrent was removed (“off”). Tests of bubbles and light were replicated three times, while tests of each configuration of flashers and sound were replicated twice. Replicate tests of deterrents—and, when applicable, the location of the deterrent within the pen—were randomized throughout the course of the day to account for any effect of tidal cycle, sun angle, or other environmental factors that could influence where herring clustered within the pen. All experiments were conducted during daylight hours. Owing to logistical challenges associated with catching herring, we were limited to one net-pen and used time with no deterrent as our point of comparison to the treatment effects. Based on observations from similar studies, the herring were acting normally (i.e., generally schooling) within the net-pen when not being exposed to experimental stimuli (Handegard et al., 2015; Rieucau, Boswell, et al., 2014).

Bubbles

A PVC tube with 5-mm holes spaced every 2.5 cm was hung diagonally in the net-pen at a depth of 2 m to test the effect of bubbles on herring behavior. Air was forced through the tube using a blower, resulting in bubbles that were approximately 2–8 cm by the time they reached the surface; these sizes are within the typical size range of humpback whale bubbles (Hain et al., 1982). Bubble experiments were replicated three times. The tube remained in the net-pen with the air turned off during all experiments, including those in which other deterrents were tested.

Sound

The effect of sound was tested using a Fishtek Marine whale deterrent pinger (3–20 kHz, with harmonics at 135 dB). This device was chosen because it is available commercially, complies with the Marine Mammal Protection Act, and includes frequencies below 5 kHz, which are within the hearing range (from 100 Hz to 5 kHz) of herring based on auditory brain-stem response (Mann et al., 2005). The pinger was suspended 2 m below the water surface; since the pinger was water-­activated, it was removed to achieve the “off” state. Its location within the net-pen varied over the course of the three replicate experiments.

Light

We tested the effects of lights with a 2 × 2 array of SafetyNet Technologies Pisces lights strobing white light at the brightest setting. The array was suspended 2 m below the surface and occupied an area of 1 m². We did not measure light spectrum attenuation at 2-m depth. As with the pinger, the location of the array varied over the three replicate experiments and the array was removed from the net-pen to achieve the “off” state.

Suspended objects

We used three configurations of 5- × 15-cm reflective flashers, which are commonly used for salmon fishing, to test the effects of suspended objects. Flasher configurations included (1) a single static row of six flashers (1 × 6) spaced 1 m apart and suspended 2 m below the water surface; (2) the same row of six flashers (1 × 6) but moving instead of static; and (3) the same row plus an added row of six flashers (2 × 6), with all flashers moving. For all experiments of suspended objects, the flasher arrays were suspended diagonally across the net-pen and were removed from the pen for the “off” state. Movement was achieved by securing one end of the array and gently tugging on the other end to mimic motion that would result from currents or wave energy in the more oceanographically dynamic setting of a kelp farm.

Data processing and analysis

Sonar data were processed using Echoview version 12.1.64 (Echoview Software Pty Ltd, Hobart, Australia). Raw data from the Flexview were processed using Flexview software to produce beamformed data that were compatible with the Echoview acoustic analysis software. The acoustic image was viewed as a multibeam echogram showing a wedge projecting from the sonar. Static components, such as the net-pen and objects associated with the bubble apparatus within the echogram, were excluded with a multibeam background removal algorithm using the mean of 41 pings with a minimum signal-to-noise threshold of −10 dB. A 3 × 3 median filter was applied to each sample following Boswell et al. (2008) but retained the full resolution of the original data. Multibeam targets were detected using a clustering algorithm for object size (seed threshold = 2.5 cm²; cluster linking distance = 10 cm) and a filter on object intensity (−20 dB). These parameters were selected by iteratively adjusting values for the clustering algorithm and visually checking against detections in the multibeam echogram. The single targets represented instantaneous detections of fish. Sequential targets were accumulated into a fish track using the “alpha-beta” fish track detection algorithm with expanding window and weighting to connect sequential targets in space into fish tracks (Kang, 2011). Further filtering involved selecting fish that had at least seven sequential detections. Fish tracks exported from Echoview containing spatial position (e.g., relative range and beam angle in the multibeam echogram), trajectory, and size were further processed using R (R Core Team, 2023). We first calculated the location of each fish track in Euclidean space and then used this location to determine the distance between the average location of each fish track and the nearest deterrent for every experiment. The “distance from deterrent” behavioral metric was calculated for both the “on” and “off” treatments.

Data were analyzed using linear mixed-effects models with the distance from deterrent for fish tracks as the response variable and treatment state (on or off) and replicate as the predictor variables. Each deterrent was analyzed separately. Tracks from individual fish may have been observed on more than one occasion during the treatment or control period. The first and last 3 min of each 30-min experiment were excluded from the analyses; this was done to ensure that only sustained behavior was being assessed and not temporary startling that may have occurred from introducing or removing a deterrent. The effect size of each treatment was determined by calculating Cohen’s d, which measures the difference between two groups (in this case, the “on” and “off” treatment states) and accounts for both the mean and SD in the data (Cohen, 1988). All statistical analyses were conducted in R (R Core Team, 2023); the Cohen’s d analysis used the lsr package (Navarro, 2015).

RESULTS

The number of fish tracks detected in our Echoview algorithm varied across experiments (Figure 2) from a minimum of 1,479 to a maximum of 23,126. Variation in the number of detected fish tracks was partly influenced by the area of the net-pen that was visible by the sonar. For example, the bubbles restricted the view to only the front half of the net-pen, as the bubbles occluded targets on the opposite side.

Both treatment state and replicate were significant in all linear models, although the coefficient estimates and effect sizes of the treatments varied across models (Table 1). Notably, only the bubble treatment had a treatment slope estimate that was larger than 0.3 and larger than the replicate slope estimate, and the bubble treatment had a very large effect size (Sawilowsky, 2009; Figure 3). Two treatments, the acoustic pinger and the single row of moving flashers, had a significant negative treatment slope, suggesting that the herring moved closer to the deterrent when it was in the “on” state (−0.1 and −0.2, respectively). All treatments other than the bubbles had a small or very small effect size (Sawilowsky, 2009; Table 1).

The average distance of the Pacific Herring to the deterrent did not vary over the course of each 30-min exposure, suggesting that the behavioral response of herring to the stimuli was constant and that the herring did not acclimate or exhibit increasing avoidance during the tested period (Figure S1 [see online Supplementary Material]).

DISCUSSION

The imaging sonar provided a powerful method to measure the position and movement of Pacific Herring in response to the deterrents tested within the net-pen. Water clarity in the net-pen was occluded by algae and detritus, making video observations infeasible. Instead, high-frequency sonar provided a full view of the net-pen at near-video quality resolution, visually interpreted as individual fish swimming around the field of view (Figure 1). However, there were some limitations to this observation method. First, while the horizontal view included the full boundary of the net-pen, the vertical opening of the sonar beam limited the detection of fish to within the middle 1 m of the 4-m-deep net-pen. Second, the strong reflection of the bubble treatment occluded fish targets that were behind the bubbles. Third, the flashers were acoustically reflective objects in the field of view, thus challenging the fish target detector and tracking algorithm. To overcome this, a region in the middle of the pen was set as an exclusion zone for tracking fish and may have biased measurements of distances to the deterrent; however, although we were not able to reliably detect fish among the flashers, fish distance to the region of the flasher deterrents was the closest among the treatments.

Our results suggest that bubble curtains, also known as pneumatic barriers, are the most effective method to deter Pacific Herring from an area. The strong response is likely due, at least in part, to interactions with one of the most important predators of herring in the region: humpback whales (Moran et al., 2018). Humpback whales create walls of bubbles to trap and then consume Pacific Herring in a cooperative behavior known as “bubble netting” (Jurasz & Jurasz, 1979). In both observational and experimental studies, herring have been shown to avoid crossing artificial and whale-derived bubble curtains, potentially due to a combination of the sound created by the bubbles and the physical presence of the bubbles themselves (Leighton et al., 2004; Sharpe & Dill, 1997).

We expected that the bubble curtain would elicit a strong behavioral response, but we were surprised by how ineffective the other methods were in deterring herring. The pinger and lights tested were selected due to their ease of acquisition as commercially available products and for being within the known hearing and visual ranges, respectively, of Pacific Herring (Blaxter & Parrish, 1965; Mann et al., 2005). Pacific Herring have been shown to habituate to sounds (Schwarz & Greer, 1984); however, we observed no behavioral response at all to the pinger. This could have been due to the pinger emitting tones that were outside the best hearing sensitivity of Pacific Herring (200–500 Hz), the pinger emitting some tones (6–20 kHz) that were above the herring hearing range, or the device source level (∼135 dB) being too low to elicit a response; alternatively, the lack of response may simply reflect that the herring were unbothered by the sound despite being able to hear it (Mann et al., 2005). The strobing lights were similarly ineffective despite serving as effective deterrents for other visual-feeding fish species, such as salmon (Johnson et al., 2005). Our experiments were all conducted during daylight hours, so it is possible that the strobe lights were not bright enough to overcome the ambient light or that Pacific Herring simply do not respond behaviorally to strobing light.

The use of salmon fishing flashers was inspired by the terrestrial farming practice of installing reflective ribbon in fields to deter avian pests (Dolbeer et al., 1986). Much like flocking birds, schooling herring are known to startle at sudden movements as an anti-predation adaptation (Rieucau, Boswell, et al., 2014; Rieucau, De Robertis, et al., 2014). Fishing flashers are designed to spin in moving water, allowing them to catch light to attract salmon. They are inexpensive and readily available at any fishing or marine supply store, and many kelp farmers in Alaska would likely already own a large supply of flashers due to the high proportion of kelp farmers who are also commercial or recreational fishers (J. A. Hollarsmith, personal observation). As kelp farms are often located in exposed areas with swell and high water flow, flashers installed around a farm would likely be highly dynamic in the swell and current. Our experiments were conducted in a sheltered location with no swell and a twice-daily tidal exchange, so we mimicked movement that would likely be achieved by suspending strings of flashers from surface buoys. However, contrary to the intended effect, the low-density static design appeared to have no effect, while the low- and high-­density dynamic designs appeared to attract the fish. Herring have very shiny scales that glint as they move about, so it is possible that the reflected light from the flashers mimicked herring rather than appearing as a threat; therefore, it could be worth testing a similar design using non-reflective materials.

Installing a pneumatic barrier around a kelp farm would likely come with other benefits and drawbacks to consider. Of the benefits, a large-scale bubble curtain could increase turbulence around or within a farm, which can reduce the thickness of boundary layers around kelp fronds and increase nutrient and carbon transport to the algae (Stephens & Hepburn, 2014). Tangentially, bubble curtains are used by other aquatic operations (e.g., construction, ports of call) to circulate water so that it does not freeze at the surface. Operators that farm species near the surface (i.e., bull kelp and oysters) might consider using bubbles to protect their crops from freezing events. These benefits would all serve to improve the quality of the crop while also protecting the crop from herring spawn. Among the primary drawbacks are likely the cost and logistical difficulty of installing and operating such a system on a remote aquatic farm, which would likely involve installing a floating platform; generating electricity using renewable energy sources (e.g., solar) or a generator (which must be supplied with fuel) to operate a blower; and maintaining the integrity of the floating and submerged components on multi-hectare sites during adverse weather conditions for up to a month when herring could potentially spawn. Other potential drawbacks could be damage to the crop if turbulence is too high or the ecological impacts of deterring nontarget species that might otherwise occur. Of all the deterrents tested, both the benefits and drawbacks are most pronounced with the bubble curtain.

We note that there are some key differences between the conditions of this study and those of a functioning aquatic farm that could influence the potential efficacy of the deterrents. These experiments were conducted approximately 2 months after the spawning season in the sheltered environment of a floating net-pen during daylight hours. Pacific Herring response to these deterrents may be markedly different under the environmental conditions and behavioral characteristics of spawning. For many herring stocks in the Gulf of Alaska, herring aggregate near their spawning locations one to several months prior to spawning and then move to the specific spawning site in the hydrographically complex environment of the nearshore days to weeks before spawning (Haegele & Schweigert, 1985). Temperature, lunar, and tidal cues initiate spawning in the subtidal zone. In the Gulf of Alaska, herring frequently spawn during neap tides, when the difference between high tide and low tide is the least (Hay, 1990). Spawning may occur during the day or at night, depending on when tides are optimal. In addition to a distinct suite of temperature, light, and tidal current conditions, herring behavior is unique during spawning. Swimming activity before and during spawning is increased, and fish dart along the shoreline, during which time they are particularly sensitive to noise and disturbance (Gauvreau et al., 2017; Thornton et al., 2010; J. Vollenweider, personal observation). Consequently, the sensitivity of herring to deterrents may be greater during spawning than what we observed in this experiment. Alternatively, the efficacy of the bubble deterrents could be reduced in a high-flow environment, where the bubbles could more rapidly dissipate.

Despite these limitations in the experimental design, the present work represents a crucial step forward in determining effective methods of Pacific Herring deterrence. As mentioned previously, field testing of deterrents is logistically challenging and resource intensive. Results would be hard to interpret, as herring spawning locations vary annually; therefore, very high spatial replication would be necessary to conclude whether an absence of herring was due to deterrents or due to stochastic movement of the herring. By conducting our experiments in a controlled environment, we were able to test a greater variety of deterrents than would be possible in a field setting. Based on our results, we recommend that future in situ experiments focus on bubble curtains, including determining the optimal spatial and temporal deployment to maximize effect and minimize cost.

Our experiments did not allow for assessment of the potential for acclimation to the deterrents beyond the 30 min of exposure during each experimental replicate. Although there was some movement of individual fish toward or away from the deterrents during that time, the average location of all fish remained quite constant throughout the experimental time block, suggesting that the herring did not rapidly acclimate to the tested deterrents. Previous experiments on bubble curtains and herring behavior indicated a potential acclimation and motivation for herring to pass through the curtain to join conspecifics on the other side (Sharpe & Dill, 1997). In applications on a kelp farm, the bubble curtain would be established before herring arrival and would be intended to exclude herring from the farm entirely, minimizing the potential for herring to be on the kelp farm side.

Future research should focus on identifying the minimum duration and spatial extent necessary to deter herring such that installation and operational costs to the farmer will be minimized. Given the strong response in the net-pen environment and the high availability of suitable habitat in regions where Pacific Herring spawn in Alaska, we expect that temporally and spatially intermittent use of bubbles would likely be effective. However, further assessments of the currents and environmental conditions at individual kelp farms will be required to determine impacts from high flow on maintaining bubble curtains. For kelp farms that are in close proximity to natural kelp beds, the intent would be to deter and redirect the herring to other nearby kelp to spawn. This study has identified a potential deterrent, bubble nets, and suggestions for further research with the aim of reducing economic impacts from herring spawning on farmed kelp beds.

SUPPLEMENTARY MATERIAL

Supplementary material is available at North American Journal of Aquaculture online.

DATA AVAILABILITY

The script and data used in analyses are available at https://github.com/afscmariculture/Herring_deterrent_publication.

ETHICS STATEMENT

Fish capture was conducted under a permit with the Alaska Department of Fish and Game (Permit Number CF-22-020). Use of acoustical transmitting devices for observing fish behavior were above frequency or power thresholds as indicated by NOAA Ocean Service Programmatic Environmental Assessment.

FUNDING

Funding for this work was provided by the NOAA National Marine Fisheries Service Office of Aquaculture.

ACKNOWLEDGMENTS

Bryan Cormack, Ben Williams, Schuyler Mace, and Anthony Prater assisted in catching fish. Charlie Waters, Heather Fulton-Bennett, Brad Weinlaeder, Costanza Diaz, and Jeremy Rusin provided critical assistance with field logistics. Arora Martinez and Cindy Sweitzer provided assistance with purchasing. Benjamin Binder and Darcie Neff loaned research equipment. Kyle Neumann and Owen Hollarmann provided field support. Three anonymous reviewers and Reginal Harrell provided feedback that improved the quality of the manuscript.

REFERENCES

Author notes