The Effectiveness of Commercially Available Double-Crested Cormorant (Nannopterum auritus) Deterrent Methods in Reducing Loafing Time on Floating Oyster Cages

Abstract

The Wildlife Services—National Wildlife Research Center Mississippi Field Station, working in conjunction with Mississippi State University, conducted a study to evaluate the effectiveness of several non-lethal, commercial, physical bird deterrents. Deterrents were available on the open market to reduce double-crested cormorant (Nannopterum auritus) (DCCO) roosting time on floating oyster cages. We hypothesized that reducing bird loafing time on floating oyster cages would reduce the potential for bird defecation. Double-crested cormorant defecation can increase coliform counts in both the water and oysters, which could result in oyster farm closure. Shellfish farms often differ from one another due to environmental factors and farm locations. Six deterrent devices were selected as options for farmers, and the amount of time DCCOs occupied floats with deterrents was compared to the time they occupied those without deterrents (i.e., on a control pond). All deterrents except the Scarem Kite significantly reduced or halted DCCO use of floats compared to the control. Of note, the Scarem Kite is the only deterrent that requires consistent wind to operate effectively. The research facility is open-air, but we did not have sufficient wind for proper Scarem Kite function. Thus, it is possible that all deterrents would have produced significant reductions in DCCO use. Our results are encouraging and show that minimal deterrent applications to oyster cage floats can help farms achieve their operational plans’ goals and reduce the potential for pathogen transfer by birds. The data presented here should be considered a starting point for developing an efficient deterrent plan that meets state and federal standards.

1. Introduction

As shellfish farming has expanded along the coastal United States, birds have become problematic as a vector spreading human and shellfish pathogens between farm sites and negatively impacting food safety. Farms are increasingly using floating aquaculture gear to facilitate management of shellfish grow-out and to avoid the challenges presented by bottom culture, including reductions in the percentage of survival, daily growth rate, and condition index [1].

The National Shellfish Sanitation Program (NSSP) guide for the control of molluscan shellfish was updated in 2019 [2]. Chapter 6.04 describes the need for an operational plan (OP) for aquaculture that attracts birds or mammals. The guide states, “Each aquaculture site that the Authority determines may attract sufficient birds and/or mammals that their waste presents a human health risk shall have a written operational plan”. This plan must be approved by the Authority prior to being implemented.

When birds congregate on floating cages, they can negatively impact the quality of coastal waters and can influence pathogen levels in or near aquaculture production facilities [3,4]. Wild bird abundance was shown to contribute to coliform counts in shallow waters [4]. These floats provide roosting sites for multiple species of birds, including the double-crested cormorant (DCCO), herring gull (Larus argentatus), and common tern (Sterna hirundo) [5]. In addition, keeping the cages so that they float at the optimum water nutrient level requires exposure to a drying period (usually 24 h) to reduce algae or biofouling. This is performed by flipping the cages. Weekly flipping of floating cages is recommended to minimize biofouling of gear and oysters in Florida growing conditions. Biofouling control during the summer months requires cages to be flipped in the late afternoon and flipped back the following morning to minimize exposure to high air temperatures [6]. The cages become a roosting area, and the birds defecate into the water, increasing coliform counts, which may lead to farm closure until these counts are reduced to levels conducive to reopening [2]. Fecal coliforms are present in the gut and feces of warm-blooded animals and are considered a more accurate indicator of the presence of animal or human waste than total coliforms. Escherichia coli is the major species among the fecal coliforms and is an indicator of fecal contamination and the potential presence of pathogens [7].

Most data on deterrent method efficiency have been generated from observational studies [5,8]. In 2015, a multiyear study performed in the State of New York found that both oyster meat and seawater contained excessive bacterial counts, which led to the emergency closure of the leased area. In 2016, three of four farms sampled were found to have excessive bacterial counts in their oyster meat samples. These three farms used floating gear and were closed on an emergency basis. The fourth farm used submerged gear and did not have excessive bacteria, allowing it to remain open. In 2017, two farms using floating gear were closed due to excessive bacterial concentrations in oyster meat; meanwhile, two farms using submerged gear remained open [9]. Oyster farm closure due to excessive bacteria results in economic losses. This problem seems to be seasonal, with the highest level of coliform counts corresponding to the migratory pattern of many species of birds. In a survey of coastal birds from New Brunswick, Canada, the double-crested cormorant was found to be the most common species observed representing 47.6 percent of all of the counts, followed by Herring Gull (18.7%) and Common Tern (13%) [5]. Surveys were conducted by Rhode Island University, the Department of Natural Resources, and the state of Rhode Island Department of Environmental Management to evaluate the seasonal pattern of the distribution and abundance of waterbirds in relation to shellfish aquaculture in coastal Rhode Island. The surveys determined the seasonal abundance of waterbird species including DCCOs in Rhode Island. Double-crested cormorants begin arriving in the area in March; their numbers continue to increase, peaking in October and decreasing rapidly to 0 by the end of December. The non-breeding period occurs from mid-October through mid-March when birds are wintering in the southern USA [8].

Oysters cultured in floating cages grow faster and have higher survival and better condition, resulting in higher production than from those cultured in bottom cages [1]. Thus, there is a strong impetus to find solutions that deter bird roosting on oyster cage floats. Much research has been conducted on deterrent methods including exclusion techniques, such as overhead wire systems and multiple frightening devices [10]. Audio frightening techniques such as pyrotechnics, propane cannons, and shooting have all been used with various levels of efficiency [11]. The locations of some of the floating oyster cage farms make these methods unacceptable due to nearby housing. We tested the effectiveness of several physical DCCO deterrents available on the open market (Figures S1–S8, Supplemental Materials) with the following three predictions:

(1) We predicted that effective deterrent techniques would reduce the probability and frequency of DCCOs successfully landing on oyster cage floats.

(2) We also predicted that effective deterrent techniques would shorten the duration of time that DCCOs remained on the floats.

(3) We predicted that reducing DCCO loafing time on floating oyster cages would reduce the potential for DCCO feces to increase coliform counts in the water surrounding floating oyster cages and the oysters themselves.

2. Methods, Procedures, and Experimental Design

Fifteen DCCOs were captured in night roosts in Mississippi and Alabama using a customized capture boat, flood lights, and dip nets [12]. Double-crested cormorants were transported to the National Wildlife Research Center Mississippi Field Station Avery (Figure 1) using a closed, climate-controlled trailer (AC 006 Transport of Waterbirds from the field to captive facilities). They were weighed to the nearest 0.01 kg and marked with either a colored, black, or white leg band on alternate legs to identify individual DCCOs from a distance or in photographs. Five DCCOs were released into each of three aviary enclosures containing a 450-square-meter pond (Figure 2) stocked with 500 catfish fingerlings per pond to approximate a stocking rate of 5000 fingerlings per 4047-square-meter pond. Each pond contained a floating oyster cage with a different DCCO deterrent method. Two ponds were treatment ponds, each having a unique deterrent, and the control pond had no deterrent system.

Each pond had 3 motion-activated cameras that recorded DCCO positions and movements. Every seven days, pond treatments were reassigned so that, each week, a different pond was the control pond, and the treatment ponds received a different deterrent method. The deterrent methods that were tested were as follows:

Gullsweep Bird and Seagull Deterrent^® Bird B Gone, Irvine, CA, USA (Figure 3), Bird B Gone Spinning Bird Deterrent^® (4 foot) Bird B Gone, Irvine, CA, USA (Figure 4), Bird Spikes for Bird, Cat, Squirrel, Racoon Animals Repellent^® TANGTEA, source Amazon.com, Seattle, Washington, USA (Figure 5), Zip ties, Hypertough,11 inch black, Walmart, Bentonville, AR, USA(Figure 6), a float-mounted triangle Ketcham Supply, New Bedford, MA, USA (Figure 7), and Scarem Kite^® Flyonte, source Amazon.com, Seattle, Washington, DC, USA (Figure 8). We recorded (1) the number of times an individual DCCO successfully landed on a float, (2) the number of individual DCCOs on a float, (3) the amount of time individual DCCOs spent on a float, and (4) the number of times an individual DCCO unsuccessfully tried to land on a float.

The study test schedule allowed 3 replications of each deterrent method, once on each of the 3 ponds (

Table 1. Study schedule.

Week	Test *	Rest **	Pond 1	Pond 2	Pond 3
	Days	Days	Method	Method	Method
1	0	7	Control	Control	Control
2	5	2	Control	Triangle	Bird B Gone
3	5	2	Scarem Kite	Control	Zip Ties
4	5	2	Gullsweep	Bird Spikes	Control
5	5	2	Control	Bird B Gone	Triangle
6	5	2	Zip Ties	Control	Gullsweep
7	5	2	Bird Spikes	Gullsweep	Control
8	5	2	Control	Zip Ties	Scarem Kite
9	5	2	Bird B Gone	Control	Bird Spikes
10	5	2	Triangle	Scarem Kite	Control

* Test days = deterrent method present; ** rest days = no deterrent method present.

).

3. Statistical Analysis

We tallied the daily number of times an individual DCCO successfully landed on a float (y1) and the daily number of times an individual DCCO unsuccessfully tried to land on a float (y2). We used the formula y1/(y1 + y2) out of the sum of y1 and y2 as a response to calculate the relative success of DCCOs in landing on the float. We used generalized linear mixed models with individual bird identification nested within pond as random effects and deterrent method, treatment sequence (i.e., pond), and trial day as fixed effects to assess the effects of different treatments on the roosting behaviors of DCCOs. If the interaction between treatment and pond was significant (p < 0.05), we conducted multiple comparisons of the marginal or least-squares means between each deterrent method and the control using the Dunnett adjustment of the p value to identify the most effective deterrent method.

We used the weekly duration (min; hereafter, weekly duration on the floats by bird)—the amount of time a bird used the floats during an experiment—as a response variable. The weekly duration was calculated as the sum of all observed durations on the floats during a week for each bird with a visible leg-band identification (ID) number. We also calculated the total number of times an identified DCCO successfully landed on the floats during a week as a response variable to assess the effects of deterrent techniques. We predicted that effective deterrent techniques would reduce the probability and frequency of DCCOs successfully landing on floats. We also predicted that effective deterrent techniques would shorten the duration DCCOs would remain on the floats. We used the square root transformation and the natural logarithmic transformation to normalize the total weekly duration. We added 0.01 min to 0 to handle zeros in the observations for the natural logarithm transformation. In the linear models (LMs) of treatment–week interactions on the total duration, the Akaike information criterion corrected for small sample size (AICc) was 805.04 for the square root transformation but 631.49 for the natural logarithmic transformation [13]. Thus, we used the natural logarithmic transformation in the subsequent analysis.

3.1. Linear Models of the Weekly Total Duration on the Floats for All the Birds in an Enclosure

Reduction in the total use of floats by all birds is a common management goal. We calculated the weekly total duration on the floats over all birds for each pond as the sum of the weekly durations for all birds during a week. We used the natural logarithmic transformation to normalize the weekly total duration for all birds. We tested for the necessity of pond ID as a random effect by comparing the AICc values between the linear mixed models (LMMs) of treatment-and-week interaction with and without the pond ID random effect. To test the fixed effects of deterrent techniques, we built all possible combinations of treatment and week, including their interaction, to assess the effects of treatments on weekly total duration on the floats for all birds. We also used the AICc for model selection, with the best model having the lowest AICc. We calculated the least-squares means of the weekly total duration on the floats by all birds for each treatment. We carried out multiple comparisons of the least-squares means between the control and each treatment using the Dunnett adjustment [14].

3.2. Hurdle Models of the Weekly Total Numbers of Landings on the Floats by an Individual Bird

We used generalized linear mixed models (GLMMs) for the total weekly numbers of successful landings with negative binomial distributions [15,16]). Given the excess number of zeros in the data, we used hurdle models, which included a sub-model that represented the probability of zero landings or failing to land on any float (i.e., complement of successful landing probability) during a week using binominal distributions with a logit link function. The hurdle model represented the non-zero numbers of landings on the floats with negative binomial distributions and a log link function [17]. Since three motion-activated video cameras monitored each float 24 h a day, zero landings by a bird was truly zero, without sampling errors. Therefore, the hurdle models were appropriate for our data. We used the same approach for the selection of the random effects as in the LMMs. When including treatment–week interactions in either the sub-model of the probability of zero or the sub-model for non-zero counts, the hurdle models failed to converge numerically. Thus, we used the additive fixed effects of week and treatment as the most complex fixed-effect models to select the random effect.

To test for the fixed effects of treatment and week on the total weekly number of landings on the floats, we built different models including different combinations of treatments using a two-stage approach. In the first stage, we included the additive effects of treatment and week in the probability of zero landings and built three sub-models of non-zero counts: treatment + week, treatment, and week. We chose the best sub-model of non-zero counts with the lowest AICc. In the second stage, we built three different sub-models: week + treatment, treatment, and week, using the best sub-model of the probability of zero landings selected in the first stage. We selected the model with the lowest AICc as the best hurdle model to assess the effects of deterrent techniques on the mean weekly total number of landings on floats. Because there were no birds successfully landing on any floats in the zip tie treatments (m4) or Gullsweep (m5) during the entire experimental period, the inclusion of zero landings for these two treatments resulted in failures in the numerical convergence of the hurdle models. Therefore, we excluded m4 and m5 from the hurdle models. We used the R package glmmTMB to carry out the analysis with LMMs and GLMMs [17]. We used the ggeffect function of the R package ggeffect to predict the probabilities of not landing on any float by treatment [18]. All statistical analyses were carried out in the R environment v.4.2 [19].

4. Results

The AICc values of LMMs for the weekly total duration on the floats for all birds with treatment–week interactions suggested that the random effect of pond ID was not necessary (AICc: 256.29 with the pond ID random effect vs. 176.33 without the pond ID random effect). Model selection suggested that the LM with the additive effects of treatment and week was the best model with the lowest AICc (

Table 2. Model selection of the linear models of the effects of deterrent techniques on the mean weekly total duration on the floats for all DCCOs in an enclosure.

Model	df	AICc
treatment + week + treatment × week	15	176.33
treatment + week	9	159.55
week	3	172.62
treatment	8	160.90

). However, the LMM of treatment had a greater AIC than the best LMM of the additive effects of treatment and week by 1.35 (<2). Therefore, we chose the simpler LM of treatment only. All deterrent techniques reduced the means of the weekly total duration on the floats (

Table 3. The linear model of the fixed effects of treatment on the weekly total duration on the floats of all DCCOs in an enclosure.

Term	Coefficient	SE	z	Pr (>|z|)
Control	5.07	0.89	5.71	<0.0001
Triangle	−5.61	1.85	−3.03	0.002
Bird B Gone	−7.05	1.85	−3.81	0.0001
Scarem Kite	−3.72	1.85	−2.01	0.04
Zip Ties	−9.68	1.85	−5.23	<0.0001
Gullsweep	−9.68	1.85	−5.23	<0.0001
Bird Spikes	−6.83	1.85	−3.69	0.0002

, p < 0.05). Multiple comparisons with the Dunnett adjustment demonstrated that all deterrents except for the Scarem Kite resulted in a significantly shorter weekly total duration on the floats for all birds (p < 0.05). The 95% CIs of the least-squares means did not overlap between the control and deterrent methods except for the Scarem Kite (Figure 9).

The AICc values of the hurdle models of the additive effects of treatment and week were 389.32 with the bird ID nested within the pond ID random effect, 383.68 with the bird ID random effect, and 385.69 with the pond ID random effect. We chose bird ID as a random effect in the subsequent analysis of hurdle models. The two-stage model selection demonstrated that the hurdle model with the effect of week on non-zero counts and the effect of treatment on the zero-truncated probability (model M4) was the best model with the lowest AICc (

Table 4. Two-stage model selection of hurdle models for the fixed effects on the non-zero number of landings on the float and zero-truncated probability of individual DCCOs.

Stage	Model	Fixed Effect	Zero-Truncated	d.f.	AICc
1	M1	Treatment + Week	Treatment + Week	17	389.32
1	M2	Treatment	Treatment + Week	15	383.68
1	M3	Week	Treatment + Week	15	385.69
2	M4	Week	Treatment	11	381.40
2	M5	Week	Week	14	400.78

). Therefore, deterrent techniques primarily increased the probabilities of individual double-crested cormorants not landing on the float (Table 4). The number of landings significantly declined (slope = −0.28, SE = 0.06, p < 0.0001; Table 4). All four deterrent techniques significantly increased the probability that a bird would not land on the floats during a week (

Table 5. Effects of four deterrent techniques on the probability that DCCOs did would not land on the float during a week.

Term	Estimate	SE	z	Pr (>|z|)
Control	−4.30	1.14	−3.79	0.0002
Triangle	4.40	1.12	3.92	<0.0001
Bird B Gone	5.44	1.46	3.73	0.0002
Scarem Kite	1.42	0.81	1.77	0.08
Bird Spikes	4.59	1.24	3.69	0.0002

Note: Zip ties and Gullsweep resulted in zero landings in all three replicates.

). To illustrate the effects of treatment on the probability of zero-truncation, we predicted the probabilities at week 10, when all treatments had been repeated three times. The 95% CI of the predicted probability that DCCOs would not land on any floats in the control was below those of the triangle, Bird B Gone, and bird spike treatments but overlapped with that of the Scarem Kite (Figure 9). Deterrent methods primarily increased the probability that birds would not land on the floats.

5. Discussion

Our study was designed to evaluate the efficacy of a variety of commercially available bird deterrents for active aquaculture farms. This is important because shellfish farms often differ from one another due to environmental factors and location. We included six deterrent devices as options for farmers and compared the amount of time DCCOs used or did not use floats with deterrents compared to those without (i.e., the control pond).

The tested deterrents worked in two distinct ways: (1) by reducing the probability of DCCO landings on the floating oyster cages and (2) by reducing the loafing duration on the floats. The deterrent methods were rotated between ponds on a weekly basis (Table 1) to reduce the chance of habituation to individual deterrent methods; however, the best LMM detected significant positive interactions between treatment and week, except for the Scarem Kite (p < 0.05), suggesting habituation of DCCOs to the deterrent techniques over time (Table 2). Combinations and rotations of different deterrents may alleviate habituation.

Deterrent methods primarily worked by increasing the probability that birds would not land on the floats, that is, deterring effects. We found that all deterrents except the Scarem Kite significantly reduced or halted DCCO use of floats compared to the control. It is important to note that the Scarem Kite is the only deterrent that requires consistent wind to operate effectively. Although the research facility is in the open air, we did not have the constant wind needed for the Scarem Kite to function properly. However, the efficacy of the Scarem Kite approached significance (p = 0.07) (Table 2), and the device demonstrated significant (p = 0.04) deterrence when considering all DCCOs over time (Table 3). It is possible that all deterrents would have produced significant reductions in DCCO use of floats if constant wind had been present.

In our study, we found that all deterrents were successful at reducing the number of DCCO landings, with varying degrees of efficacy. In terms of cost and overall success, zip ties were the most effective deterrent, with no DCCOs successfully landing at any point in our study. The second most successful deterrent was the Gullsweep Bird and Seagull Deterrent^®, with only one successful DCCO landing. However, the management of floating cages requires them to be occasionally flipped, which could potentially damage a deterrent secured to the top of a float. With this logistical factor in mind, our study suggests that zip ties would be the easiest and most cost-effective deterrent to maintain.

Our results are encouraging and show that minimal deterrent applications to aquaculture floats can help farms reach their OP goals and reduce the potential for pathogen transfer by birds.

6. Management Implications

The data presented here should be considered as a starting point for developing an efficient deterrent plan that meets state and federal standards. Oyster farms/producers have a better understanding of the surrounding environment that influences the species and number of birds that affect their farms. Most of the tested deterrents work very well but may not be practical in every environment or farm. Oyster farms need to determine which individual method or combination of methods considered in this study or which potential new deterrent works best in their situation.