Comparison of catch per unit effort among four minnow trap models in the three-spined stickleback (Gasterosteus aculeatus) fishery

Minnow traps are commonly used in the stickleback (Gasterostidae) fishery, but the potential differences in catch per unit effort (CPUE) among different minnow trap models are little studied. We compared the CPUE of four different minnow trap models in field experiments conducted with three-spined sticklebacks (Gasterosteus aculeatus). Marked (up to 26 fold) differences in median CPUE among different trap models were observed. Metallic uncoated traps yielded the largest CPUE (2.8 fish/h), followed by metallic black nylon-coated traps (1.3 fish/h). Collapsible canvas traps yielded substantially lower CPUEs (black: 0.7 fish/h; red: 0.1 fish/h) than the metallic traps. Laboratory trials further revealed significant differences in escape probabilities among the different trap models. While the differences in escape probability can explain at least part of the differences in CPUE among the trap models (e.g. high escape rate and low CPUE in red canvas traps), discrepancies between model-specific CPUEs and escape rates suggests that variation in entrance rate also contributes to the differences in CPUE. In general, and in accordance with earlier data on nine-spined stickleback (Pungitius pungitius) trapping, the results suggest that uncoated metallic (Gee-type) traps are superior to the other commonly used minnow trap models in stickleback fisheries.

differences in CPUE could be (at least partly) explainable by differences in the ability of different trap models to retain sticklebacks. Namely, as for any other passive fishing gear, the CPUE of minnow traps depends not only on the probability that fish will encounter and enter into the trap, but also on the probability that they will be retained within the traps until retrieved 4,20 . In general, the results should prove as useful guidance for choice of minnow trap model in three-spined stickleback fisheries.

Results
Field experiments. A total of 4971 fish (1032 males, 3939 females) were caught during the trapping period.

Figure 1. Boxplots for catch per unit effort (CPUE) for different trap models in (a) Experiment 1 and (b) Experiment 2.
Boxes depict the 25% and 75% quantiles; vertical line within boxes depicts the median. Whiskers depict 10% and 90% quantiles; dots depict actual data points (data for each replicate site over the six different catches). All values refer to data where site differences have been normalized away.
Similar results were obtained when sex was replaced by standard length (z = − 0.26, P = 0.8) or weight (z = 0.35, P = 0.73) in the model. Log-rank tests confirmed that escape rates were lowest and very similar (P = 1.00) for the silver metallic and black canvas traps, and far higher and similar (P = 1.00) for the black metallic and red canvas traps (Table 1; Fig. 2). All other pairwise comparisons of trap models across "low" and "high" escape trap models (cf. Fig. 2) were highly significant, even after Bonferroni correction (Table 1).

Discussion
The most important result of this study was that the four different minnow trap models differed significantly in CPUE, with the silver metallic Gee-type trap having the highest performance of all trap models, by a very large margin. Furthermore, the differences in model-specific CPUEs appeared to depend not only on the probability of sticklebacks to enter the different traps, but also on differences in probability of escaping from them. In the following, the interpretations and implications of these findings are discussed in light of what is previously known about the factors influencing CPUE in stickleback minnow trap fisheries.
The outperformance of the silver metallic Gee-type trap compared to the similarly shaped but differently coated black metallic traps conforms to the results of Merilä et al. 18 , who also discovered this to be the case in a freshwater population of nine-spined sticklebacks. Hence, together these results suggest that the traditional silver metallic Gee-trap might generally be a more efficient minnow trap model for stickleback fisheries in a variety of habitats. Our results further suggest that the reason behind the CPUE difference between these two metallic trap models might be in their ability to retain fish once they have entered the trap: the escape probability was significantly higher for the black than for the silver metallic traps. Although the reason for this difference in escape probability remains unresolved, it is noteworthy that in addition to color (black vs. silver), the two trap models also differ in wire diameter (black metallic: 1.5 mm; silver metallic: 0.5 mm) and pattern of netting (silver metallic: square-shaped netting; black metallic: diamond-shaped netting; see Fig. 1 in ref. 18). One or all of these factors could generate differences in motivation or ease to escape from the black metallic traps, which could explain their lower CPUE. For example, it is possible that the black color and thicker (and diamond shaped) netting make the black metallic traps more "confined", allowing fish inside the traps to detect the trap entrances more easily than in the silver metallic traps with different wire characteristics.  The silver metallic traps did not only outperform the black metallic traps, but also both the black and red canvas traps. This is a significant finding, since the performance of metallic and canvas traps has never been directly compared in the stickleback fishery. Several earlier studies have compared the performance of the two canvas trap models, both in three-spined 16 and nine-spined stickleback fisheries 15,17 , and found that the black traps catch more (adult) sticklebacks than the red traps in both species. Hence, the results from earlier studies would encourage the use of black canvas traps over the red ones. However, although the results of this study conform to the earlier findings of the red canvas traps being inferior, this study also revealed that the difference in CPUE between the two canvas trap models is far smaller than that between the metallic and canvas traps. By directly comparing all four trap models, it is clear that the silver metallic trap is the most likely to yield highest CPUEs in stickleback fisheries.
Differences in CPUE among trap models can be ultimately attributed to differences in the rates at which fish encounter, enter and escape from the traps 3,21 . As the different trap models in this study were deployed side-by-side, at a distance of only 50 cm apart, we can assume that they were encountered at the same rate. Hence, the observed differences in CPUE should reflect differences in either entry or escape rates. Among the metallic trap models, escape rate was low for the silver trap but high for the black one. Hence, the lower CPUE in the black metallic traps -roughly 50% less than that of the silver traps -might be explainable by their high escape rate. Similarly, the high escape rate from the red canvas traps aligns with their low CPUE: they had the highest escape probability and lowest CPUE of all trap models. This likely explains their consistently low efficiency in catching sticklebacks. However, the black canvas traps had a low CPUE in the field, despite the low escape probability in the lab. This would suggest that in this particular trap model, low entry rate might be the likely explanation for the low CPUE. In other words, the finding that trap model-specific CPUEs and escape probability estimates are not perfectly matching suggests that part of the variation in CPUEs must depend on variation in entry rates. Direct estimates of entry rates would be helpful to establish this inference firmly. In the same vein, although there is clear trap model-specific variation in escape probabilities, it should be kept in mind that these estimates were derived in the laboratory using solitary individuals. Individual and other biotic interactions in the wild might influence escape (and entry) probability in a trap model-specific fashion. Nevertheless, whether caused by variation in entry or escape rates, the differences in CPUE among different trap models were clear and consistent with estimates available from earlier studies [15][16][17][18] .
Finally, although perhaps most often used and discussed in a fisheries context, CPUE also has uses in fundamental ecology and conservation biology by providing information about population abundance 3,8,22,23 . However, CPUE is an accurate index of abundance only if catch efficiency over time and space remains constant and unaffected by other factors 2-3 . The finding that the different minnow trap models differ considerably in their efficiency of catching sticklebacks suggests that comparisons of population abundance estimates obtained using different trap models cannot be made reliably -at least not before correcting for differences in catch efficiency among trap models. However, such adjustments may be hard to make if the relative efficiency of the different trap models change in response to variation in biotic and abiotic conditions. For instance, as discussed above, variation in biotic factors influences fish behavior 24 which in turn can translate to (trap model-specific) variation in entry and escape rates 25 . Hence, given these considerations and the evidence that the silver metallic traps yield consistently higher CPUE estimates than other trap models tested in this study, the silver metallic traps should provide a good standard for population abundance estimation.
In conclusion, the results show that different minnow trap models differ significantly in mean CPUE in the three-spined stickleback fishery, and this variation is likely to be explainable by variation in rates of entry and escape from different trap models. As in the case of earlier reports from the nine-spined stickleback fishery, the silver metallic Gee-traps yielded highest CPUEs. Hence, the silver metallic Gee-traps appear to be the trap model of choice irrespectively of whether the aim is to maximize CPUE or provide an estimate of population abundance.

Field-experiments.
Field-experiments were conducted in a brackish-water (average annual salinity 6.03%; 24) bay of Notviken in the southern part of Eckerö, Åland Islands, Finland (ca. 60°11′ 35.66″N, 19°37′06.77″E). Trapping was conducted over five consecutive days (25 th to 29 th May 2015). All four trap models (see below for description) were deployed at five locations approximately 10 m apart and 1-2 m from the shore. Within each location, traps were randomly set approximately 50 cm apart in dense vegetation, at a depth of 30-80 cm. At an additional four locations, the two models of metallic traps (black and silver) were set in pairs, similar to the abovementioned setup. Hence, a total of nine sets were used: five sets had all four trap models, and four sets had only the two metallic traps. All traps were checked daily. The fish from each trap were counted, sexed according to nuptial coloration of males (i.e. blue eyes, red throat; 12) and released at the site of capture. We note that although this method of sexing stickleback in the field is highly reliable in the breeding season 12 , it is possible that some immature males were mis-classified as females. Soak times between each check varied from 8 to 22 h.
Since the primary aim was to compare the CPUE between the two metallic minnow trap models, both the black and silver metallic traps were used in all nine replicate sets. Being of secondary interest, the two different models of collapsible canvas traps were used only in five of the replicate sets. As a consequence, the study design was not fully crossed and balanced. As such, data were analyzed in two parts. First, the CPUE between the two metallic trap models was compared across all nine sets by leaving out the canvas traps from the five sets where all four trap models were deployed (henceforth: "Experiment 1"). In these analyses, a total of 18 traps (nine traps of each model) were fished over six consecutive checks. Second, the comparison of CPUE between all four trap models was done only for the five sets in which all models were deployed (henceforth: "Experiment 2"). In these analyses, a total of 20 traps (four traps per model) were fished over the six consecutive checks.
Laboratory experiments. CPUE depends not only on the rate of fish encountering and entering into the traps, but also on the rate of fish escaping from the traps. Therefore, possible differences in the probability that fish will escape from the different trap models were evaluated. These experiments were conducted in the laboratory utilizing two 327.6L tanks (78 × 140 × 30 cm) in the aquaculture facilities of the Department of Biosciences, University of Helsinki, Helsinki, between 17 July and 13 August 2015. The fish used in these experiments were adults caught between May and June 2015 from four different coastal sites in the Baltic and North Seas (55°01′N, 8°26′E; 56°39′N, 9°59′E; 60°27′N, 26°56′E; 65°05′N, 25°23′E). Since the fish from these different populations originated from and were maintained at different salinities, salinity was gradually (over a period of one week) adjusted to 5 parts per thousand for all populations and experimental tanks before starting the experiments.
To estimate escape probability, individual fish were first acclimated to a trap by placing a 50mL tube (containing the fish) within the trap. Sticks were fixed to each end of the tube such that the tube (and hence, the fish) could be maintained in the middle of the trap. After five minutes, the tube was removed, leaving the stickleback in the trap. Starting from this moment, the location of the test individuals was observed every 30 minutes over a period of three hours. If the fish was seen outside of the trap during this period, the trial was terminated and the escape time was noted. Each fish (N = 92) was tested in all trap models once. The sequence of testing the same fish in different trap models was fully randomized.
Ethics statement. The experiments were approved by the Finnish National Animal Experiment Board under license STH223A, and all used methods were carried out in accordance with approved guidelines. The procedures also adhered to the 'Guidelines for the treatment of animals in behavioural research and teaching' 26 .
Statistical analyses. Since the soak time between consecutive trapping sessions varied (see above), the count data was standardized to CPUE by dividing the number of fish caught by the soak time. Hence, the CPUE estimates refer to fish caught per hour.
Repeated measures analysis of variance was not possible since the CPUE estimates were exponentially distributed. Hence, a series of non-parametric tests were used to investigate the effects of trap model, site and gender on CPUE.
For Experiment 1, CPUE was compared between the two metallic trap models that were used in all nine sets, using a Wilcoxon signed-rank test. In order to control for possible set effects, the CPUE data were first normalized by partitioning out the variance due to differences among the sets.
For Experiment 2, CPUE among all four trap models from the five sets (see above) was compared with Kruskall-Wallis analysis. Again, to control for possible set effects, the CPUE data were first normalized by partitioning out the variance due to differences among the sets. Pairwise comparisons among different trap models were compared with Steel-Dwass tests which correct for multiple testing.
In both experiments, all analyses were performed separately for males and females (including possibly some non-breeding males), as well as for the full data. Temporal variation was not incorporated in any of these analyses as there was no temporal variation in CPUE estimates in either of the two experiments, even after normalizing the data first for trap type (Kruskal-Wallis tests; Experiment 1: χ 2 = 5.55, df = 5, P = 0.35; Experiment 2: χ 2 = 3.86, df = 5, P = 0.57) or site (Kruskal-Wallis tests; Experiment 1: χ 2 = 7.45, df = 5, P = 0.19; Experiment 2: χ 2 = 1.36, df = 5, P = 0.92) effects. Nevertheless, to ensure that our inference is not biased by assumptions about temporal and spatial independence of CPUE estimates, we also fitted generalized linear models for both experiments, where CPUE was modeled as an exponentially distributed response variable (using reciprocal link function), and trap type, set and time interval as fixed effects.
The data on escape probability (reciprocal of probability of fish being trapped) at a given time point was analyzed using Cox-regression 27 , treating trap model, sex, size (standard length), weight, test tank identity and population of origin as factors, and presence or absence of fish within trap as a time dependent binomial response variable. Due to collinearity between sex, size and weight, only one of these three variables at a time was included in the initial model. Model selection was conducted based on Akaike Information Criterion (AIC) and resulted in a model with only one factor: trap model. In this situation, the use of Cox regression is asymptotically equivalent to the use of a log-rank test 28 . Accordingly, log-rank tests 29 were used to compare pairwise (between trap models) probabilities of escaping, and p-values were adjusted for multiple testing using Bonferroni correction.
Statistical analyses were performed using JMP Pro 11 (SAS Institute, Inc., Cary, NC, USA), except escape probabilities, which were compared using the package "survival" in R 30 .