Comparative analysis of depth distribution for seventeen large pelagic fish species captured in a longline fishery in the central-eastern Pacific Ocean

The objective of this study was to estimate depth distribution of pelagic species captured in a longline fishery and to evaluate the difference in depth distribution among species. We estimated depth distribution for 17 frequently captured species based on a Chinese longline fishing trip targeting bigeye tuna (Thunnus obesus) in the central-eastern Pacific Ocean in February-November 2006. The depth distributions of 13 bycatch species were significantly different from that of bigeye tuna. Although most of the bycatch species were found to be distributed in water depths shallower than bigeye tuna (i.e. increasing hook depths can decrease catch rates of these species), the rates of catch rates declined with increasing hook depths may be different. The depth distributions were found to be not significantly different between genders for 7 species. There was no significant correlation between fish sizes and capture depths. The information derived from this study can play an important role in reducing bycatch in pelagic tuna longline fisheries in the central-eastern Pacific Ocean.


INTRODUCTION
Pelagic longline fishing gear is widely used in the open ocean to target tuna and billfish.The depth at which fish are captured can provide critical informa-tion for understanding the impacts of longline fisheries on targeted and bycatch species (Bigelow et al. 2006).Deploying longline hooks at appropriate depths can greatly improve catch of desired species, such as bigeye tuna (Thunnus obesus; Suzuki et al. 1977, Boggs 1992) and billfishes (Boggs 1992), and reduce bycatch of protected species and untargeted species, such as sea turtles (Gilman et al. 2006, Beverly et al. 2009).
Several approaches can be used to obtain vertical distribution information for pelagic species.Electronic tags (e.g.acoustic, archival, and satellite) equipped with pressure sensors are an appropriate tool for observing vertical movements of pelagic fish in their habitat (Bach et al. 2003, Musyl et al. 2003, Schaefer et al. 2009, Walli et al. 2009, Stevens et al. 2010).Longlines equipped with time-depth recorders (TDRs) and hook timers provide information on the time and depth of capture for many pelagic species (Boggs 1992, Bach et al. 2003, Bigelow et al. 2006).One advantage of TDRs and hook timers, compared with electronic tagging studies, is that a large number of individuals of different sizes and species in different environmental conditions can be sampled (Bach et al. 2003).
Application of large numbers of TDRs on fishing vessels can be time-consuming and the cost prohibitive, and in practice it is not always feasible for commercial longliners (Bigelow et al. 2002).It has long been recognized that longline hook depth can be predicted using catenary algorithms (Yoshihara 1951, 1954, Suzuki et al. 1977).However, the direction and velocity of ocean currents and wind have important influences on catenary shape and hook depth (Ward and Myers 2006) and predicted depth may differ greatly from actual observed depth (e.g. by TDRs; Rice et al. 2007).Therefore, the catenary method has been frequently used for estimating hook depth in pelagic longline fisheries by using empirically derived correction factors (e.g.Hinton and Nakano 1996, Yano and Abe 1998, Bigelow et al. 2006, Ward and Myers 2006).
Much work has been done to investigate the vertical distribution of fish species of economical and/or ecological importance.Most of the work, however, focuses on single species, such as bigeye tuna (Bigelow et al. 2002, Bach et al. 2003), blue marlin (Makaira mazara; Luo et al. 2006), bigeye thresher shark (Alopias superciliosus; Nakano et al. 2003) and blue shark (Prionace glauca; Bigelow and Maunder 2007).Limited studies focus on the depth distributions of multiple species.For example, the depth distributions of 37 pelagic species caught in pelagic longlines in the Pacific Ocean were inferred by generalized linear mixed models (Ward and Myers 2005).To evaluate how vertical distributions of fish species may interact with longlines, a comparative study on differences in depth distributions among species is needed.This important topic has been rarely dealt with in previous studies.
In this study, we estimated hook depths of capture for 17 pelagic species captured during a Chinese longline fishing trip in the central-eastern Pacific Ocean.One hypothesis to be tested was whether there were significant differences in vertical distributions between target and bycatch species.For species with suitable sample sizes, gender-and size-specific differences in depth distributions were also evaluated.We were also interested to test whether the results of these comparisons were robust to the methods used for correcting longline hook depth.The information derived from this study can improve our knowledge of the vertical distribution of pelagic species and can play an important role in developing methods for mitigating bycatch in pelagic tuna fisheries.

Data collection
All data in this study were collected by an onboard fisheries observer during a commercial longline fishing trip in the central-eastern Pacific Ocean in February-November 2006.This trip was conducted by a deep-set longline vessel (overall length 49 m, beam 9 m, depth 3.9 m, 495 GT) equipped to target bigeye tuna and freeze the catch.The longline gear consisted of a 120-130 km mainline, 40-50 m floatlines, and 46 m branchlines that were spaced 45-48 m apart.The number of hooks between floats (HBF) was 17 or 18 and "J" type tuna hooks were used.Setting began between 2:00 and 9:00 am and hauling began between 02:00 and 03:00 PM (local time).On average, 2500-3000 hooks were deployed for each set.Until now, all other China longline vessels and equipped gears operating in the central-eastern Pacific Ocean to target bigeye tuna, are similar to the one used in this study.
Set-specific latitude, longitude and gear configurations, including speed of vessel (8-14 knot) and speed of shooting mainline (6-8 m s -1 , averaged over 3 observations per set at start of gear deployment, mid-point and close to end of deployment), HBF, length of mainline per basket (a "basket" encompassed the hooks between 2 successive floats, see L in Eq. 1), and length of branchline and floatline were recorded.Fishes captured were randomly subsampled for biological measurements (including sex, length and weight) and the number of the hook that caught the fish was recorded.A total of 17  1), where sample sizes were greater than n=30.Species other than the targeted bigeye tuna were all considered as bycatch species in this study.Set locations are shown in Figure 1.

Depth calculation
Hook depth was calculated using the catenary method (Yoshihara 1951(Yoshihara , 1954)), which predicts the depth according to longline configuration using the following equation: where D j is the depth of jth hook, h f and h b are the lengths of floatline and branchline, respectively (Fig. 2), L is the operational length of mainline in unit basket (calculated by the speed of shooting mainline multiplied by the time spent for deploying one unit of basket), n is the number of branchlines in unit basket, and j° is the angle between horizontal line and tangential line of the mainline at connecting points of the mainline and floatline.For each basket, the 2 hooks closest to the floats on the 2 ends are both numbered as the first hook, assuming that the branchlines were hung symmetrically.Because it was difficult to make direct measurements, j° was solved by iteration of the sagging rate using the following formula (Yoshihara 1954): where k is the sagging ratio defined as the length of horizontal line divided by the length of mainline between unit baskets and estimated as the ratio of the speed of the mainline thrower to speed of vessel (Bigelow et al. 2006).In this study, speed of the line thrower and speed of vessel varied slightly during the trip.The range of k was between 0.760 and 0.804; thus, by solving Eq. ( 2), j° ranged from 60.0 to 55.5°.

Within-set correction of hook depth
The catenary method from Eq. ( 1) results in a single depth value for each longline hook.However, actual hook depth may vary both between and within sets (Bigelow et al. 2002).In this study, sagging ratio k for each set was observed, so j° value for each set was calculated using Eq.(2).Between-set variability of hook depth was therefore not subject to further consideration here.
Within-set variability of hook depth in longline gear has been reported in previous studies (Boggs 1992, Yano and Abe 1998, Bigelow et al. 2006, Bach et al. 2009).Yano and Abe (1998) found that for a given set of longline the hooks in deeper waters tended to have a larger variation in vertical distribution than hooks in shallower waters.Following Bigelow et al. (2002), we corrected hook depth D j using the following linear relationship developed by Yano and Abe (1998): where s(D j ) is the standard deviation of hook depth D j and j is the hook number as described above.For each D j calculated with Eq. ( 1), 1000 random samples of hook depths from normal distributions N~(D j , s 2 (D j )) were generated and the mean of these values was regarded as the estimated depth of hook j.

Correction of shoaling influence
Actual hook depth is usually much shallower than that predicted using the catenary equation and is often referred to as shoaling (Bigelow et al. 2006, Bach 73).See Table 1 for full species names.
The number indicates sample size.et al. 2009).It is a common practice to express longline shoaling in terms of a percentage (Bach et al. 2009).This percentage-also called the correction factor-has been empirically used to adjust the hook depths calculated from the catenary method (Suzuki et al. 1977, Hinton andNakano 1996).Correction factors may differ greatly in different oceanic areas due to different oceanographic conditions.Suzuki et al. (1977) estimated a shoaling of 15% (i.e.actual hook depth reaching 85% of predicted hook depth) to correct hook depth in the equatorial Pacific.This factor was adopted by Hinton and Nakano (1996).Bigelow et al. (2006)  We used 3 methods to correct the predicted hook depth after within-set corrections.First, 3 constant correction factors, 15%, 20% and 25%, were assumed to adjust the predicted depths, respectively.Second, setspecific random correction factors generated from a uniform distribution U(0.15, 0.25) were used.The hook depths in the same set were corrected using the same factor value, so 211 random numbers were used here.Third, hook-specific correction factors also generated from a uniform distribution U (0.15, 0.25) were used to adjust the predicted depths.Each predicted individual depth was assigned a single correction value, so 2343 random numbers were used.

Statistical analysis
The interquartile range (IQR) was used to show the difference in depth distribution among species.The IQR tends to be robust for outliers and extreme values, which are commonly observed in fisheries studies.The one-sample Kolmogorov-Smirnov goodness-of-fit test (one-sample K-S test) was used to test the normality of the depth distributions for each species.If they followed normal distributions, the two-sample t test was used to test whether the mean depth of bigeye tuna differed significantly from the mean depth of bycatch species.We then chose the two-sample Wilcoxon rank-sum test (two-sample Wilcoxon test), a non-parametric method which needs no assumption of a certain distributional form, to examine whether the median depth of bigeye tuna differed significantly from the median depth of bycatch species.The two-sample Kolmogorov-Smirnov goodness-of-fit test (two-sample K-S test) was used to examine whether the depth distribution (the distribution function) of bigeye tuna differed from that of bycatch species (Venables and Ripley 1999).Simple Bonferroni adjustment (target P value =0.05/number of pairwise tests) was used to adjust the significance level for pairwise comparisons, which can reduce the risk of Type I error (Holm 1979).Differences in median depth and depth distribution between females and males were also evaluated by two-sample Wilcoxon test and twosample K-S test, respectively.Relationship between capture depth and fish size was examined by Pearson's correlation coefficient with t test.The analysis was conducted for depth estimates derived from both constant and random shoaling correction factors.
After testing differences in depth between species, we employed divisive hierarchical cluster analysis with Euclidean distance to further identify the potential groups of these species.We divided the depth estimates of each of the 17 species into 4 depth ranges, 50-150 m, 151-200 m, 201-250 m and 251-300 m, and calculated the proportion of catch in number in each range.Thus, 4 variables with 17 observations were obtained for the cluster analysis.Depths derived from the constant (only using a shoaling factor of 20%) and random shoaling factors were analyzed.All the statistical analysis was conducted in the S-PLUS program (Release7.0.6 for Windows).

Depth range and distribution
In addition to the targeted bigeye tuna, we estimated depth ranges for 16 bycatch species in the longline fishery, including 3 tuna species, 2 billfish species, 4 shark species and 7 other species (Table 1).Under constant shoaling assumptions, the minimum capture depth was estimated at 92, 98 and 104 m for the shoaling assumption of 25%, 20% and 15%, respectively, and the maximum capture depth was 253, 269 and 286 m for the 3 shoaling factors, respectively (Table 2).Bigeye tuna was captured at the deepest mean depth, and wahoo (Acanthocybium solandri) was captured at the shallowest mean depth.IQR plots for 20% shoaling suggested that depth distributions varied greatly among species (Fig. 3).Some species, such as blue marlin and blue shark, had similar median depths, but their depth ranges differed greatly.IQR plots for 15% and 20% shoaling factors showed the same trend.

Comparison of depth between bycatch species and targeted bigeye tuna
Because the one-sample K-S test indicated that capture depths of 8 of 17 species (BET, YFT, SKJ, BSH, BTH, ALX, LEC, WAH; See Table 1 for full species names) did not follow normal distribution (P<0.05),we did not conduct t tests for comparing the differences in mean capture depth between bigeye tuna and bycatch species.Differences in median depth and depth distribution between bycatch species and bigeye tuna, based on constant and random shoaling correc-tion methods, were tested with P values given in Table 3.The two-sample Wilcoxon test indicated that, except for bigeye thresher and sickle pomfret (Taractichthys steindachneri), the bycatch species had significantly different median depths from bigeye tuna (Table 3).This was consistent with the results derived by evaluating differences in the depth distribution between bycatch species and bigeye tuna.The two-sample K-S test suggested that, except for bigeye thresher, velvet dogfish (Zameus squamulosus) and sickle pomfret, each of the 13 bycatch species had significantly different depth distribution from bigeye tuna (Table 3).Cluster analysis showed dissimilarities among species regarding capture depth and species belonging to the same cluster tended to be captured in the same depth range (Fig. 4).Bigeye thresher and velvet dogfish showed more similarity to bigeye tuna in capture depth than the other 14 species (Fig. 4).A cluster tree based on constant and set-specific random correction factors showed almost the same grouping results.

Differences in depth distribution between sexes and sizes
The difference in capture depth between females and males was tested for 7 species (BET, YFT, SWO,  See Table 1 for full species names. BSH, BTH, PSK, PLS).The two-sample Wilcoxon test indicated that median depths were not significantly different between females and males for all of the 7 species (target P value =0.0071; P>0.0071).Accordingly, the two-sample K-S test indicated that there were no significant differences in depth distribution between females and males (target P value =0.0071; P>0.0071).
The results were not sensitive to the methods used for shoaling correction.
The relationship between capture depth and individual size (length) was examined using Pearson's correlation coefficients for all 17 species.Correlation coefficients were low and no significant correlations were found for any of the 17 species (target P value =0.0029; P>0.0029), indicating that smaller fishes were captured at almost the same depths as larger fishes.The results were also robust to the methods for correcting shoaling influences.

Implication of differences/similarities in vertical distribution
This study indicates that bigeye tuna showed different depth distributions to those of most bycatch species.Differences in capture depth among species are useful information for developing methods of mitigating bycatch by adjusting depths at which hooks are deployed in bigeye tuna fisheries.
Depth estimates and statistical tests indicated that 3 tuna species (YFT, ALB and SKJ), 2 billfish species (SWO and BUM), 2 shark species (BSH and PSK) and 6 other species (PLS, ALX, TAL, LEC, NEN and WAH) were captured at shallower depths than bigeye tuna.It is likely that setting longline hooks deeper can reduce catch rates of these bycatch species.This has been demonstrated for species such as wahoo, skipjack (Katsuwonus pelamis) and blue marlin.For example, eliminating shallow hooks from standard tuna longlines can significantly reduce wahoo catch rates (Beverly et al. 2009) and catch rates of skipjack and blue marlin decreased with the depth of longline hook being deployed (Nakano et al. 1997).Electronic tagging experiments also provided evidence for vertical habitat preferences of these pelagic species.Satellite telemetry tagging showed that blue sharks spent between 52% and 78% of their time at depths <100 m and between 10% and 16% at depths >300 m off eastern Australia (Stevens et al. 2010).
Sickle pomfret and 2 shark species (bigeye thresher and velvet dogfish) were captured at about the same depth as bigeye tuna.Beverly et al. (2009) found that eliminating shallow hooks in the upper 100 m of the water column from standard tuna longlines could significantly increase catch rates of sickle pomfret.It was also reported that catch rate of bigeye thresher shark increased with the depth of hook deployed (Nakano et al. 1997).Satellite telemetry tagging has demonstrated that bigeye thresher shark spent least time in the surface layers and most time at >300 m depth (Stevens et al. 2010).Therefore, adjusting longline gear in certain depth ranges can reduce catch rates of some species but increase catch rates of others.Similarity or dissimilarity among species, derived from cluster analysis, should be considered in using differences in depth distribution between targeted and bycatch species to mitigate bycatch.Although most of the bycatch species were found to be distributed in shallower waters than bigeye tuna, the extent to which the bycatch decreased with increasing hook depths may be different.Catch rate of wahoo will probably decline more quickly than that of other species.Although capture depth distributions of sickle pomfret, bigeye thresher and velvet dogfish were not found to be significantly different from those of bigeye tuna, cluster analysis indicated that they belonged to different groups.
Because of depth limitation of longline gear, capture depths estimated from pelagic longline cannot represent the whole vertical habitat ranges for most species.For example, blue sharks spend between 35% and 58% of their time at depths of less than 50 m (Stevens et al. 2010), whereas most deep-set longline hooks are deployed beyond this depth.Tagging data showed that bigeye tuna can descend to depths beyond 450 m (Musyl et al. 2003, Schaefer et al. 2009) but the deepest hook depth was about 300 m in the Hawaii-based longline tuna sets (Bigelow et al. 2006).Therefore, the vertical ranges of these species are not fully covered by longline gears.In addition, speciesspecific availability and vulnerability to longline gear can also influence the capture depth distribution of target and bycatch species.Additional experimental fishing trials, covering a broader spatial-temporal range and a variety of gear types and deployment strategies, are needed to quantify the depth distributions of target and bycatch species.
Sexual segregation is a widespread behaviour in the animal kingdom and can arise within a species owing to sex differences in body size, activity, behaviour, nutritional requirements and/or habitat selection (Magurran andMacias Garcia 2000, Wearmouth andSims 2008).Although sexual segregation for most pelagic species is of little interest, sexual segregation for sharks has attracted great interest because of the potential impacts on fisheries population dynamics and management (Mucientes et al. 2009).None of the 7 species examined in this study showed sexual segregation in vertical habitat, as there were no significant differences in vertical distributions between genders.This is probably due to the fairly limited spatial coverage of the study area.Highly migratory species may show sexual segregation on larger spatial scales.

Factors influencing vertical distribution
Many factors, including diel vertical movement, hook type and size, bait type and soak time, may influence the pelagic species' availability and vulnerability to capture by longline gear.Diel vertical movement might be the most important factor, and has been observed in the bigeye tuna (Holland et al. 1990a, Musyl et al. 2003), yellowfin tuna (Thunnus albacares; Holland et al. 1990a, Schaefer et al. 2009), skipjack (Yuen 1970, Schaefer et al. 2009), swordfish (Xiphias gladius; Carey and Robison 1981), blue marlin (Holland et al. 1990b), blue shark (Stevens et al. 2010) and bigeye thresher shark (Nakano et al. 2003, Stevens et al. 2010).Diel vertical movement range, however, may differ among species.Bigeye tuna is mainly distributed between 220 and 240 m during day time and between 70 and 90 m at night (Holland et al. 1990a).Yellowfin tuna, however, inhabits shallower water than bigeye tuna, and stays at an average daytime depth of 71.3 m and an average night-time depth of 47.3 m (Holland et al. 1990a).Blue marlin moves closer to the surface at night, which is consistent with the behaviour reported for skipjack (Yuen 1970), swordfish (Carey and Robison 1981) and bigeye thresher sharks (Nakano et al. 2003), but differs from that reported for the striped marlin (Tetrapturus audax; Holland et al. 1990b).However, vertical movement and distribution pattern for majority of pelagic species are still less understood.Collecting depth information covering the whole day time period is essential to improve our understanding of diel movement pattern and to develop appropriate fishing strategies for maintaining catch rates of target species and reducing catch rates of bycatch species at the same time.

Factors influencing longline shoaling
Environmental factors, such as current velocity, shear and wind, are probably the most significant factors accounting for the deviation between predicted hook depths derived using the catenary method and actual depths (Boggs 1992, Bigelow et al. 2006).We used constant and random shoaling factors to adjust predicted depth for pelagic species captured by longlines.The range of shoaling factors was chosen based on the results derived from previous studies (Hinton andNakano 1996, Ward andMyers 2006).
Due to logistic constraints, shoaling factors have usually not been estimated for all hook positions between 2 consecutive floats.For instance, Bigelow et al. (2006) estimated shoaling percentages for deep-set longline gear by placing TDRs in the middle position on the mainline between 2 floats, a technique which was adopted by Bach et al. (2009).Having too few TDRs or not spacing them equally to monitor all hook positions may be a source of error for the estimation of longline shoaling factors.Moreover, a variety of gear configurations and deployment strategies may also cause errors when applying a shoaling factor derived from one gear configuration to another.For example, changing branchline and floatline length, HBF and/ or speed of shooting mainline can cause gear sinking into different water levels where environmental factors may vary.In the Hawaii-based commercial fishery, shoaling reached ~50% for shallow swordfish sets and only ~30% for deeper tuna sets (Bigelow et al. 2006).However, a large discrepancy in the estimates of shoaling between shallow and deep longline sets has been reported.The mean shoaling estimates on a shallow set were found to be ~24% (Hanamoto 1974) and ~11% (Nishi 1990) in the Pacific Ocean and Indian Ocean, respectively.Boggs (1992) estimated a shoaling factor of 46% for a deeper tuna set near the Hawaiian Islands.
Because we did not have TDRs placed on branchlines, hook depths estimated in this study should be considered to be biased to some extent.However, the comparative analysis we conducted with different shoaling factors indicates that the conclusions obtained in this study were robust to the methods used for correcting shoaling influences.This may result from the fact that the relative differences in capture depths among species were so large that they did not change too much after corrections by different shoaling factors.Such differences in capture depths among pelagic species are most likely to be determined by different vertical habitat preferences.
The method of estimating capture depth in this study assumed that all fishes were not caught when the longline was deployed or retrieved.In practice, some individuals can be captured during these 2 periods, but Boggs (1992) showed that most pelagic fish were caught while the longline gear was settled rather than while it was sinking or rising.Species mostly captured by shark hooks on floats were excluded in this study.It should be noted that a shark or large animal captured on these hooks could distort the shape and therefore the depth of longline.It might be better to exclude the baskets (even their adjacent baskets) with at least one big animal captured on shark hooks from the analysis.Unfortunately, we were not able to do this because of lack of this information.

CONCLUSIONS
Differences in depth distributions between targeted bigeye tuna and bycatch species associated with longline fisheries provide important information to mitigate bycatch.Setting deeper longline gear appears to reduce catch rates for most (13) of the 17 bycatch species in this study, including 2 shark species (blue shark and crocodile shark), whereas it increases catch rates of other (3) species, including 2 shark species (bigeye thresher and velvet dogfish).
Trade-offs need to be considered in order to adjust the depth of pelagic longline fishing gear in predicting catch rates among target species, protected species, and other ecologically and/or economically important species.Therefore, it is critical to identify fish species that play key roles in ecosystem dynamics and to investigate their vertical distribution for developing optimal operational depth ranges for pelagic longline fisheries.Investigating biological or ecological mechanisms for vertical habitat preferences by pelagic species can also improve our understanding of their availability and vulnerability to capture.

Fig. 2 .
Fig. 2. -Configuration of unit basket of longline gear similar to the longline used by the Chinese tuna fishery in the central-eastern Pacific Ocean.The number in the illustration indicates the hook number, i.e. there are 17 hooks here in the unit basket.

Fig. 3 .
Fig. 3. -The box-plot of the estimated depth ranges under constant shoaling assumption (20%) for 17 pelagic species captured in the longline observer trip in the central-eastern Pacific Ocean in February-November 2006 (the centre line is the median depth, the edges of the box are the25th and 75th percentiles, respectively, and the whiskers extend to the most extreme data points).See Table1for full species names.

Table 1 .
-Species captured in the longline observer trip in the central-eastern Pacific Ocean in February-November 2006.Species listed here are only those analyzed in this study.

Table 2 .
-Estimated depth ranges (m) under constant shoaling assumptions for 17 pelagic species captured in the longline observer trip in the central-eastern Pacific Ocean in February-November 2006.The Max, Min and Mean represented maximum, minimum and mean depth, respectively.The lower case letter n indicates the sample size of the species.See Table1for full species names.

Table 3 .
-The observed P values for testing the difference in median depth and in depth distribution (estimated from constant and random shoaling assumptions) between bycatch species and bigeye tuna captured in the longline observer trip in the central-eastern Pacific Ocean in February-November 2006.The commonly used significance level of 0.05 was adjusted as 0.05/16=0.0031(SimpleBonferroniadjustment for multiple comparisons) in determining whether the difference was significant.*:significant.See Table1for full species names.