USGS Home
Contact USGS
Search USGS

Abstract:  The incidental take of marine birds was estimated for the following North Pacific driftnet fisheries in 1990: Japanese squid, Japanese large-mesh, Korean squid, and Taiwanese squid and large-mesh combined. The take was estimated by assuming that the data represented a random sample from an unstratified population of all driftnet fisheries in the North Pacific. Estimates for 13 species or species groups are presented, along with some discussion of inadequacies of the design. About 416,000 marine birds were estimated to be taken incidentally during the 1990 season; 80 % of these were in the Japanese squid fishery. Sooty Shearwaters, Short-tailed Shearwaters, and Laysan Albatrosses were the most common species in the bycatch.

Regression models were also developed to explore the relations between bycatch rate of three groups—Black-footed Albatross, Laysan Albatross, and "dark" shearwaters—and various explanatory variables, such as latitude, longitude, month, vessel, sea surface temperature, and net soak time (length of time nets were in the water). This was done for only the Japanese squid fishery, for which the most complete information was available. For modelling purposes, fishing operations for each vessel were grouped into 5-degree blocks of latitude and longitude.

Results of model building indicated that vessel had a significant influence on bycatch rates of all three groups. This finding emphasizes the importance of the sample of vessels being representative of the entire fleet. In addition, bycatch rates of all three groups varied spatially and temporally. Bycatch rates for Laysan Albatrosses tended to decline during the fishing season, hereas those for Black-footed Albatrosses and dark shearwaters tended to increase as the season progressed. Bycatch rates were positively related to net soak time for Laysan Albatrosses and dark shearwaters. Bycatch rates of dark shearwaters were lower for higher sea surface temperatures.

Introduction

Concern about the take of non-target species in driftnet fisheries conducted in the North Pacific led to an international effort to monitor the fisheries. The five fisheries involved are Japanese squid, Japanese large-mesh, Korean squid, Taiwanese squid, and Taiwanese large-mesh. Those fisheries and the monitoring program are described in Yeh and Tung (1991); Fitzgerald et al. (1993); Gong et al. (1993); Nakano et al. (1993); and Yatsu et al. (1993); We examined data on the bycatch of marine birds from those driftnet fisheries; the two Taiwanese fisheries were considered together (Anon. 1991a, 1991b; INPFC. 1991a, 1991b). Information from 1989 was presented in Johnson et al. (1991). Data described herein were from 1990, except that the Japanese large-mesh fishery was conducted during September 1990-April 1991.

We completed two kinds of analyses: estimation of total incidental bycatch and model building for the rate of bycatch. The general approach to estimation involves sample survey, estimating the value of some parameter for a large population from the values associated with a sample from that population. The parameter in the present instance is bycatch, the incidental take of non-target species. Model-building potentially is useful both for estimating values of the parameter of interest in strata that were not sampled, and for attaining an understanding of factors influencing bycatch so that future surveys can be designed more efficiently. Modelling results may also prove helpful for understanding the risks of incidental take and minimizing them.

Because bycatch is expected to vary directly with the fishing activity, interest is in a ratio, bycatch divided by effort, which we term bycatch rate. Bycatch is analyzed by species or species group. Effort is indexed by the number of standardized effective tans set (50 meters of driftnet). Unfortunately, the effort data for the Taiwanese fisheries did not include a measure of tans set, so ratio estimation could not be employed in that case.

Eleven species of marine birds most commonly taken were considered, as well as two composite groups. The species are the Laysan Albatross (Diomedea immutabilis), Black-footed Albatross (D. nigripes), Northern Fulmar (Fulmarus glacialis), Flesh-footed Shearwater (Puffinus carneipes), Buller's Shearwater (P. bulleri), Sooty Shearwater (P. griseus), Short-tailed Shearwater (P. tenuirostris), Leach's Storm-Petrel (Oceanodroma leucorhoa), Fork-tailed Storm-Petrel (O. furcata), Horned Puffin (Fratercula corniculata), and Tufted Puffin (F. cirrhata). Because Sooty and Short-tailed Shearwaters were difficult to distinguish, they were often grouped together with unidentified dark shearwaters and termed dark shearwaters. Although Sooty Shearwaters might be separated from Short-tailed Shearwaters on the basis of known differences in geographic and temporal distributions, we used only the information provided from the monitoring rather than any ancillary information. All remaining birds and birds not identified to species were included under other birds. Bycatch includes the total of birds brought to the deck of the ship, those dropping out of the net, and a few of unknown fate.

Estimation

It is necessary to assume that the data represent a random sample from some population, which would be all driftnet fisheries of a particular nation in the North Pacific during 1990. We did not stratify the fisheries for two reasons. First, from the analysis of bycatch data from the 1989 Japanese squid fishery (Johnson et al. 1991), stratified sampling did not consistently give more precise estimates than unstratified sampling. This was probably a consequence of the stratification being appropriate more for the bycatch of squid or the bycatch of salmonids; such stratifications are unlikely to be useful for estimating bycatch of a broad diversity of seabirds. Second, stratification poses difficulties with strata that were fished (i.e., there were effort data) but not observed (no observation data). Overcoming that problem forces one to use a model-based estimator or to take an ad hoc approach. Models of bycatch have been uniformly poorly fitting, so model-based estimators are of doubtful value.

Estimation is in principle straightforward, and the sampling design does not depend on the response variable (bycatch by species). However, the sampling variances of estimated bycatch (by species) may be large, especially if bycatch rates vary markedly. We used conventional estimators of totals and variances (Cochran 1977). Bootstrapping would probably provide better estimates of precision (Larntz and Garrott 1991), but the greater problem is accuracy. See subsequent section on Quality of the Estimates.

Estimation Methods and Results

Information was available for 74 Japanese vessels in the North Pacific squid driftnet fishery. There were 3,010 operations (involving 2,132,651 standardized effective tans) observed, of a total effort of 23,656 operations (involving 22,769,857 standardized effective tans). To prevent observer fatigue, operations were not observed in their entirety; an average of 80.4 % of observed operations were monitored. The overall sampling fraction, adjusted for the fraction observed, was 10.23 % of operations (and 9.37 % of the standardized effective tans).

The sampling units are operations, each of which possesses two characteristics: the number of standardized effective tans and the incidental bycatch of a particular species. The estimator was a ratio estimator, based on an unstratified random sample (Cochran 1977:150-156).

The observed take of birds in the Japanese squid fishery ranged from 8 Leach's Storm-Petrels to 28,332 dark shearwaters. Table 1 gives values for all taxa, and Table 2 gives observed bycatch rates. Projecting the observed bycatch by taxon to the entire effort gives estimated total bycatches shown in Table 3. Values ranged from 85 (for Leach's Storm-Petrel) to 302,495 (for dark shearwaters). Also given are estimated standard errors.

Japanese Large-mesh Fishery

Information was available for 14 Japanese vessels in the North Pacific large-mesh driftnet fishery. There were 475 operations (involving 205,144 standardized effective tans) observed. We did not have the entire observational data set available; consequently we do not know if the subset we used was representative.

Numbers of birds in the sampled operations ranged from 0 for several species up to 76 Laysan Albatrosses and 82 dark shearwaters, of which at least 72 were Sooty Shearwaters (Table 1). Observed bycatch rates per 1,000,000 standardized effective tans ranged from 0 for several species to 370 for Laysan Albatross and 395 for dark shearwaters (Table 2). Projections for the entire fishery are given in Table 3.

Effort data for the Japanese large-mesh fishery included 3,485 operations involving 4,682,630 standardized effective tans. The overall sampling fractions were 9.45 % of the operations and 4.38 % of the standardized effective tans.

Korean Fishery

The Koreans engaged only in squid fishing. The Korean observer program was new in 1990 and experts in marine bird identification were not available during the training of Korean scientific observers. The low species diversity recorded by Korean scientific observers, 7 identified species and 2 unidentified species categories in 471 operations, compared with that of U.S. scientific observers, 16 identified species and 11 unidentified species categories in 440 operations (Anon. 1991a), led us to believe that some similar-appearing species may have been combined by Korean scientific observers. Since the situation was not clear when we conducted our analysis, we chose to use only U.S. observations in the 1990 data. We had information on 440 operations involving 11 vessels. In total, 1.49 % of the operations were observed by U.S. observers. By assuming that these represent a random sample of operations, we can estimate the bycatch rate of a particular species as described for the Japanese squid fishery.

The recorded take of marine birds ranged from 0 for several species to 312 Sooty Shearwaters and 423 total dark shearwaters (Table 1). We estimate that as many as 40,388 dark shearwaters, of which at least 29,790 were Sooty Shearwaters, 4,774 Laysan Albatrosses, and 4,679 Fork-tailed Storm-Petrels were taken in the Korean squid fishery (Table 3).

Taiwanese Fishery

The effort data for the Taiwanese fishery did not include the number of standardized effective tans. Thus, we could not develop ratio estimators involving tans; we had to treat operations as sample units. Because operations were not observed in their entirety, subsampling was involved, with operations serving as primary sampling units and sections within operations serving as secondary sampling units. That design, coupled with variable sampling intensity, is complicated to analyze. Also, sections were not selected randomly, and the lengths of sections that were not observed were not recorded. We accordingly treated operations as sampling units and scaled up the observed bycatch according to the number of sections observed as a fraction of the total number of sections. For example, if 6 of 8 sections of an operation were observed, and 3 dark shearwaters were identified in the 6 observed sections, we used 3 × (8/6) = 4 as the number for the entire operation. This procedure underestimates the variance by ignoring the variability among secondary sampling units (sections) within primary sampling units (operations). The underestimation is probably minor, however, because a majority of the sections were observed and application of the finite population correction would substantially reduce the variance among secondary sampling units.

The Taiwanese observer program was new in 1990 and experts in marine birds were not available during training of Taiwanese observers. Since none of the birds recorded by Taiwanese observers was identified to species, we chose to use only U.S. observer data in our analysis. We had information on 331 operations observed on 14 vessels, a sampling fraction of 2.33 % of the operations. There were 138 vessels in the Taiwanese fishery in 1990 involved in 11,266 operations. Because effort was not distinguished between squid and large-mesh fisheries, observed data were pooled. This action almost certainly increases the variance of the resulting estimators and can induce bias if the observed data were not representative of the combined squid and large-mesh fisheries.

Observed bycatch ranged from 0 for several species to 35 Laysan Albatrosses and 98 dark shearwaters (Table 1). We estimate the entire fishery took 1,622 Laysan Albatrosses, 4,559 dark shearwaters, and lesser numbers of other species (Table 3).

Quality of the Estimates

The theory of statistical sampling allows one to draw inferences from a sample that are applicable to a larger population (e.g., Cochran 1977). The precision of resulting estimates can also be assessed. The theory, however, rests on the foundation that the sample is in some sense a random subset of the units in the entire population.

The present analysis treats operations as sample units. These cannot realistically be considered a random sample from the entirety of operations, because only operations conducted by vessels with observers could possibly be included in the sample. Further, some vessels in the Korean and Taiwanese fisheries did not have observers aboard at the beginning of the fishing season, thereby biasing downward the probability of including early operations in the sample. A similar problem exists with late operations. The finding that vessel is a very significant variable affecting bycatch rate (Johnson et al. 1991; Larntz and Garrott 1991; this report) supports the contention that operations are not ideal sample units. It might be feasible to treat vessels as sample units, but this approach would be limited by small sample sizes, possibly incomplete coverage of vessels, and other drawbacks.

The 1990 survey should be adequate if the observed fishing effort and activities were representative of the entire fishery. Conscientious attention was paid to obtaining a representative sample in the 1990 fishery, and the spatial and temporal pattern of the sample matches the total fishing effort much more closely than in 1989. Whether or not the bycatch of vessels in the sample is representative of those in the entire fleet is impossible to determine.

Model-building Methods

The general approach to identifying variables affecting bycatch involves building a regression model of bycatch rate for each species. For the problem at hand, we used linear models, so named because they are linear in the parameters to be estimated. This is important because it means that linear models can be used for modelling both linear and certain types of nonlinear relationships.

We restricted our modelling to three of the most commonly caught species or groups: the Black-footed Albatross, the Laysan Albatross, and dark shearwaters. Six explanatory variables were considered in our models: latitude, longitude, month, vessel, sea surface temperature, and net soak time (length of time nets were in the water). Other variables, including net mesh size and nationality of observer, were not considered because their effects were not separable from that of vessel.

Fishing operations were grouped into 5-degree blocks on the basis of latitude and longitude. The response variable, bycatch rate, was calculated for each vessel in each 5-degree block for each month that the vessel fished in that block by summing across fishing operations. Such pooling was necessary to avoid large numbers of zero bycatch rates that occurred during individual fishing operations. Sea surface temperatures and net soak times were averaged across the fishing operations to arrive at values for each vessel in each 5-degree block-by-month combination.

The effects of latitude, longitude, and month were modeled by including linear, quadratic, and cross-product terms involving those three variables in the model. Linear and quadratic terms for sea surface temperature and a linear term for net soak time were also included in the model as covariates.

Fishing vessel identity was included in the model as a random component, based on the assumption that the 74 observed fishing vessels represented a random sample of all vessels in the fleet. The implication of treating fishing vessel as a random component is that we were interested in estimating the vessel-to-vessel variation, but not in comparing the specific vessels that had observers. The total variation in bycatch that is not explained by other variables in the model can then be partitioned into two components—one for vessel and one for inherent variability. For all analyses, vessel was assumed a priori not to interact with other variables; that is, effects were assumed roughly constant for all vessels.

Models were fit using the General Linear Models procedure of SAS (SAS Institute 1989). The general approach to model building was to fit a model that included all variables and then systematically remove those that were not statistically significant. For each group we identified a "best" model, based on a highest coefficient of determination (R²) lowest mean squared error (MSE), and fewest parameters.

Model-building Results

The best model for Laysan Albatrosses involved fishing vessel; linear terms for latitude, longitude, month, and net soak time; a quadratic term for longitude; and cross-product or interaction terms involving latitude, longitude, and month (R² = 0.41, MSE = 6.83 × 10^-7). Because of the interactions, the relationship between the response variable, bycatch rate, and the explanatory variables latitude, longitude, and month was complex and difficult to interpret. Table 4 depicts the estimated spatial and temporal effects on bycatch rate for a typical fishing vessel after adjusting for the effect of net soak time. A general tendency was for bycatch rates, adjusted for differences in soak time, to be higher at the eastern and western edges of the fishery and lower in between. Except at latitude 35°N, adjusted bycatch rates tended to be higher in the west than in the east. Adjusted bycatch rates generally decreased during the season and were higher in the north than in the south. Bycatch rates for Laysan Albatrosses were positively affected by net soak times. A 4-hour increase in net soak time resulted in a 0.0005 increase in bycatch rate. Sea surface temperature did not noticeably influence bycatch rates. The estimated variance component for vessel was 4.4 × 10^-8, 6 % of the total variation. The model for Laysan Albatrosses had little predictive power. This was apparent from the low coefficient of determination and from the fact that the model produced negative estimates for a few 5-degree block-by-month combinations (Table 4).

Black-footed Albatross

The best model for Black-footed Albatrosses involved vessel and linear and cross-product terms for latitude, longitude, and month (R² = 0.21, MSE = 4.88 × 10^-8). In general, bycatch rates increased from west to east. Notable exceptions to this trend occurred in May and June, but only at latitude 40°N (Table 5). Bycatch rates of Black-footed Albatrosses tended to increase during the season. Sea surface temperature and net soak time did not significantly affect bycatch rates of Black-footed Albatrosses. The estimated variance component for vessel was 0.19 × 10^-8, 4 % of the total variation. Negative estimates of bycatch rate occurred for some 5-degree block-by-month combinations in which observed bycatch rates were either zero or very small (Table 5).

Dark Shearwaters

The best model for dark shearwaters involved fishing vessel; linear terms for latitude, longitude, month, sea surface temperature, and net soak time; and cross-product terms involving latitude, longitude, and month (R² = 0.26, MSE = 5.59 × 10^-4). As was the case for the other species, spatial and temporal effects were complex and interrelated (Table 6). For latitudes 35°N and 45°N, adjusted bycatch rates tended to increase from west to east, whereas for latitude 40°N they increased from east to west. The general tendency was for bycatch rates to increase during the season. Bycatch rate was negatively related to sea surface temperature. A 1-degree increase in sea surface temperature resulted in an estimated decrease in bycatch rate of 0.003. Bycatch rate and net soak time were positively related. A 4-hour increase in soak time increased bycatch rate by about 0.005. The estimated variance component for vessel was 2.03 × 10^-5, 3 % of the total variation. As was the case for the two albatross species, the best model for dark shearwaters produced a few negative estimates of bycatch rates (Table 6).

Summary of Model-building

The overriding finding from these analyses is the presence of considerable unexplained variability in bycatch rates. None of the best-fitting models possessed much predictive power. This was evident from the low coefficients of determination and from the fact that some of the models produced negative estimates of bycatch rate in certain time-by-area classes. Despite their low predictive power, the models are nevertheless helpful in identifying variables that affect bycatch of marine birds and that relate to distributions of the birds.

Bycatch rates for all three of the species or groups considered varied both spatially and temporally. The response of bycatch rate to latitude, longitude, and month, however, was often inconsistent and the influence of one variable depended on levels of the other variables.

Net soak time was an important explanatory variable for bycatch rates of both Laysan Albatrosses and dark shearwaters. Average soak times ranged from about 6 hours to nearly 60 hours. The effect of soak time over this range was an increase in bycatch rate of 0.0071 for Laysan Albatrosses and 0.071 for dark shearwaters.

Bycatch rates of dark shearwaters also responded to changes in sea surface temperature. Average sea surface temperatures ranged from 11.2° C to 18.6° C. The estimated response of bycatch rates to increasing temperatures over this range was a decrease of 0.021.

Vessel was found to be a significant determinant of bycatch rates for all three groups, but accounted for only 3 to 6 % of the total variance. This result stands in contrast to the findings of Johnson et al. (1991) in which vessel accounted for 21 and 33 % of the variation in bycatch rates for Black-footed Albatrosses and dark shearwaters, respectively, in the 1989 Japanese squid driftnet fishery. This finding may in part reflect improvements in observer training that were implemented for the 1990 survey.

Some of the variability in bycatch rates that we discuss above is consistent with known patterns of seabird distribution and abundance. The Laysan Albatross is most abundant in the western Pacific while the Black-footed Albatross is most abundant in the eastern Pacific (Fisher and Fisher 1972; Robbins and Rice 1974; Kuroda 1988), and both species tend to be found more frequently at the edges than in the more central area of the North Pacific (Fisher and Fisher 1972; Gould et al. 1982). These patterns could account for the longitudinal trends we find in their bycatch rates. Correlations between bycatch rates and sea surface temperatures may also reflect distributional patterns for Sooty Shearwaters (Kuroda 1991) and Laysan Albatrosses (Fisher and Fisher 1972; Kuroda 1988). Sea surface temperatures increase dramatically from north to south in the fishing area and temperature fronts along this gradient are known to affect the distribution of marine organisms and the animals that prey on them (Gould and Piatt 1993). Similarly, monthly and seasonal fluctuations in bycatch rates are affected by migratory and dispersal patterns of the birds. Laysan and Black-footed Albatrosses, for example, breed from November through June and young and post breeding adults disperse north in July (Fisher and Fisher 1969; Kuroda 1988; Robbins and Rice 1974). Black-footed Albatrosses reach a peak abundance in the Gulf of Alaska in August (Gould et al. 1982) accounting for the trends we find in their bycatch rates. The seasonal decline we see in Laysan Albatross bycatch rates, however, seems to be inconsistent with this explanation.

Rates of bycatch of marine birds in other driftnet fisheries are frequently correlated with the density and behavior of birds in the fishing area (Ainley et al. 1981; Piatt and Nettleship 1987; Jones and DeGange 1988; DeGange and Day 1991). Marine bird density and behavior, in turn, are influenced by complex interactions involving life history features (including breeding cycle, migration, and molt), food availability, and other environmental conditions (Hunt and Schneider 1987; Piatt and Nettleship 1987; Brown 1988; Wahl et al. 1989; Gould and Piatt 1993). In addition, significant correlations with net-soak times and variation among vessels (the latter perhaps related to the targeting of specific areas and water temperatures for fishing activities, or differences in skill and effectiveness among ship crews) further complicates analysis. We suggest that caution should be exercised in interpreting bycatch rates of marine organisms in driftnets.