SCRS/2008/079 STANDARDIZED CATCH RATE OF SAILFISH (Istiophorus platypterus) CAUGHT BY BRAZILIAN LONGLINERS IN THE ATLANTIC OCEAN (1986-2006) Catarina Wor 12 ; Bruno L. Mourato 1,3 ; Humberto G. Hazin 1 ; Fábio H. V. Hazin 1 ; Paulo Travassos 1 SUMMARY In the present study, catch and effort data from xx,xxx sets done by the Brazilian tuna longline fleet (national and chartered vessels), from 1986 to 2007, in the equatorial and southwestern Atlantic Ocean, were used to generate standardized CPUE series, by GLM, assuming 3 different error distributions: Negative Binomial, Poisson and Delta-lognormal. Four explanatory factors were considered in all three approaches: Year, Distance from coast, area, month and also the interactions between Year and Month and Year and Area. The sailfish catch, from 1986 to 2006, accounted for only 1.7% of the total catch, varying between 0.4%, in 1990, and 4.7%, in 2005. The mean percentage of fishing sets with no sailfish catch was very high (78%) ranging from 56%, in 2004, to 93%, in 1994. The Poisson model explained 18.6% of the total deviance, while the Negative Binomial model accounted for 16.9%. The Deltalognormal model, in turn, explained 16.8% of the total deviance, for the positive sets, and 58.5% for the proportion of positives. The sailfish CPUE seemed to fluctuate around 0.4 fish per thousand hooks, until 2001 (mean of 0.39, ranging from 0.23 to 0.72). During 2004 and 2005, however, an abrupt increase in CPUE was observed, with values more than doubling. In 2006, the CPUE again dropped to 0.55, going back to the range of the previous period. Although the reasons for the sudden increase of sailfish CPUE in 2004/ 2005 are not clear, it is possible that the introduction of new fishing logsheets, in October 2003, as well as the prohibition of sales of white and blue marlin, from July 2002 on, may have contributed to introduce an unquantifiable bias in CPUE values during these years. 1 Universidade Federal Rural de Pernambuco/Departamento de Pesca e Aqüicultura, Recife- PE, Brasil. 2 Corresponding author: catarinawor@gmail.com 3 PhD student / Departamento de Oceanografia-Universidade Federal de Pernambuco-UFPE
1. Introduction The sailfish (Istiophorus platypterus) is one of the most common billfish species in the Atlantic Ocean, being usually associated to warm tropical waters and to coastal areas or oceanic islands surroundings (Nakamura, 1985). It has a high value for recreational fisheries, as well as for small-scale and artisanal fisheries, particularly in developing countries, where it represents an important source of food and income. The species is also often caught, as by-catch, by the tuna longline fishery. Their catches, however, are largely under-reported due to their relatively low commercial value, when compared to tunas and swordfish (Ortiz and Arocha, 2004). Although sailfish catches represent a small fraction of the total catch in longline fisheries, their fishing mortality resulting from it represents, by far, the main impact on their stocks in the Atlantic Ocean (Restrepo et. al., 2003). The most recent attempt to assess the sailfish stocks in the Atlantic Ocean was conducted in 2006. However, due to the lack of data, it was not possible to achieve any conclusive results, although some indications of early decreases in the biomass of both stocks considered (eastern and western Atlantic) were noticed (ICCAT, 2006). In the present paper, a standardized CPUE series for sailfish caught in the southwestern Atlantic by the Brazilian tuna longline fleet, including both national and chartered vessels, from 1986 to 2006, was generated using a generalized linear model (GLM). A major difficulty was the very large proportion of sets with zero catches (close to 80%) of sailfish, stemming from its non-target nature. In order to address this problem, three different error distribution structures were assumed in the analysis: Poisson, Negative Binomial and Delta-lognormal. The reliability of the estimates and their potential usefulness for future stock assessment were then discussed. 2. Material and Methods In the present study, catch and effort data, from 40,332 individual longline sets, carried out by the Brazilian tuna longline fleet, including both national and chartered vessels, from 1986 to 2006, except for the years of 1990, 1993 and 2003, were used to calculate Catch per Unit of Effort (CPUE), as the number of fish caught per 1,000 hooks. The years of 1990, 1993 and 2003 were excluded from the analysis due to the very low effort in relation to other years (less than 10% of the mean effort)(table 1), as well as to a very high variance of sailfish catches. All years considered in the analysis had an effort over 20% of the mean value for the whole period (Table 1). The longline sets were widely distributed in the western Equatorial and South Atlantic Ocean, ranging from 015ºW to 050ºW and from 08ºN to 52ºS (Figure 1). The resolution of 1º latitude x 1º longitude, taken from the position in the ending of each longline set, was used to characterize the geographical distribution of the fishing effort. Due to the very large proportion of sets with zero catches (close to 80%) of sailfish, stemming from its bycatch nature in this fishery, three standardized CPUE series were generated, assuming different error structures: Poisson, Negative binomial and Delta-lognormal. The factors considered as explanatory variables were Year, Month, Area and distance from the coast ( Dis ). The interactions between Year and Month and Year and Area were also included in the GLM. Deviance tables were constructed for each error distribution aiming to compare the three models. The 20 S latitude was chosen as the boundary line between the two areas considered in the Area factor. The distance from the coast was calculated as the minimum distance, in degrees, between each fishing set and the coast line of the continent or the nearest oceanic Island. The GLM models used can be expressed by the following equations: For the Negative Binomial and Poisson models: CPUE= Year + Dis + Area + Month + interactions + ε For the positive sets in the Delta model: Log(CPUE)= Year + Dis + Area + Month + interactions + ε
In order to evaluate the statistical significance of the factors included in the model, p-values, based on the χ 2 test statistic (α = 0.05), were used. The fits were done in S-Plus and the predicted values of standardized CPUE were obtained for every Year, by fixing the level of remaining factors at the level with the highest number of observations. 3. Results and discussion The sailfish catch, from 1986 to 2007, including the years of 1990, 1993 and 2003, accounted for only 1.7% of the total catch, varying between 0.4%, in 1990, and 4.7%, in 2005 (Figure 2). As expected, the mean percentage of fishing sets with no sailfish catch was very high (78%) ranging from 56%, in 2004, to 93%, in 1994 (Figure 2). Poisson and Negative binomial models presented exactly the same sequence of contributions by the different factors, with the main ones being the interaction between Year and Month, Year and Area, which explained, respectively, 38.2%, 37.5%, and 12.0 % of the total explained deviance, for the Poisson model, and 37.5%, 37.5%, and 10.8%, for the Negative binomial model (Table 2). Regarding the Delta-lognormal model, the positive sets presented a slightly different pattern. The two most important variables were again represented by the interaction between Year and Month and Year, accounting for 57.1% and 30.0% of the total explained variance, respectively, but the third most important factor was Month (5.4%), and not Area (0.3%). For the proportion of positive sets the most important factors were again Year (48.4%), Year : Month interaction (26.5%0, and Area (10%). The Poisson model explained 18.6% of the total deviance, while the negative binomial model accounted for 16.9%. The Delta-lognormal model, in turn, explained 16.8% of the total deviance, for the positive sets, and 58.5% for the proportion of positives (Table 2). Since the explained deviance is a function of the error distribution assumed, the results cannot be compared, being only indicators of goodness of fit between the observed data and the model (Ortiz and Arocha 2004). The Poisson and the delta-lognormal (positive sets) models presented better residual distribution than the negative binomial (Figure 3). Plots of residuals versus quartiles of standard normal showed a much higher deviation for Poisson and Negative Binomial models, which were highly biased towards positive values, than for the delta-lognormal model (Figure 4). The residuals partial plots, however, showed reasonably good distribution in all models, except for the 4 th and 9 th levels of the factor Month (Figures 5-8). Despite the high variance commonly associated to CPUE series generated without data aggregation, the confidence intervals, determined for Poisson and Negative Binomial models, were relatively narrow during the analyzed period (Table 3 and Figure 9). Although there were differences between the three models, due to the distinct error distributions assumed, all standardized CPUE series presented similar patterns, which were also not so different from the nominal CPUE (Figure 10). The sailfish CPUE seemed to fluctuate around 0.4 fish per thousand hooks, until 2001 (mean of 0.39, ranging from 0.23 to 0.72), agreeing with the results obtained by Ortiz and Arocha (2004) and Yokawa and Takeuchi (2001), whose study areas partially overlap with the present work. During 2004 and 2005, however, an abrupt increase in nominal CPUE was observed, with values more than doubling, staying around 1.0. Interestingly, all standardized CPUEs showed a strong increase already in 2002, differently from the nominal value, which remained very close to the mean for the whole period (0.41). In 2006, the nominal CPUE again dropped to 0.55, going back to the range of the previous period. Although the reasons for the sudden increase of sailfish CPUE in 2004/2005 are not clear, it is possible that the introduction of new fishing logsheets, in October 2003, resulted in a better reporting of catches. Another possible influence might have been the prohibition of sales of white and blue marlin, from July 2002 on. In order to circumvent the commercialization ban, it is possible, particularly in the
case of frozen specimens, that some of the white marlins caught have been reported as sailfish. Unfortunately, it is not possible to verify neither to quantify none of these possible sources of bias. 4 Acknowledgements The authors acknowledge the Secretaria Especial de Aqüicultura e Pesca da Presidência da Republica do Brasil (SEAP), which made this study possible. 5 References ICCAT. 2006. International Commission for the Conservation of Atlantic Tunas: Report for biennial period, 2006-07. Parte I, vol. 2, 244p. NAKAMURA, I. 1985. FAO species catalogue. Vo1.5.Billfishes of the World. An annotated and illustrated catalogue of marlins, sailfishes, spearfishes and swordfishes known to date. FAO Fish.Synop., (125)Vo1. 5:65 p. ORTIZ, M. and F. Arocha. 2004. Alternative error distribution models for standardization of catch rates of nontarget species from a pelagic longline fishery: billfish species in the Venezuelan tuna longline fishery. Fisheries Research, 70: 275 297. RESTREPO, V.; Prince, E. D.; Scott, G. P.; Uozumi, Y. 2003. ICCAT stock assessments of Atlantic billfish. Marine and Freshwater Research 54: 361-367 YOKAWA, K and Y. Takeuchi. 2001 Standardization of CPUE for sailfish caught by Japanese longline in the Atlantic Ocean. Col. Vol. Sci. Pap. ICCAT Table 1- Distribution of annual effort, in number of hooks, from 1986 t0 2007, and percentage of the yearly effort in relation to the mean value for the whole period (4,292,077 hooks), for the Brazilian tuna longline fishery, operating in the Atlantic Ocean.
Year Effort % 1986 1,298,317 30.2 1987 888,121 20.7 1988 1,165,798 27.2 1989 939,636 21.9 1990 384,784 9.0 1991 1,726,479 40.2 1992 3,219,458 75.0 1993 395,842 9.2 1994 1,834,953 42.8 1995 4,205,448 98.0 1996 1,619,601 37.7 1997 3,132,105 73.0 1998 5,725,299 133.4 1999 10,612,030 247.2 2000 14,206,350 331.0 2001 15,086,852 351.5 2002 8,857,219 206.4 2003 265,041 6.2 2004 7,138,918 166.3 2005 7,298,942 170.1 2006 2,994,087 69.8
Table 2- Deviance analysis table of explanatory variables for Poisson, Negative Binomial and Delta-lognormal models fitted to Sailfish CPUE from Brazilian pelagic longline fishery. Model Structure Poisson Negative Binomial d.f. Δ p-value Δ Dev. d.f. p-value Res. d.f. Res. Dev. Total Exp. Dev. Res. d.f. Res. Dev. Dev. Total Exp. De Null 40332 81945.4 40332 27713.9 Year 17 40315 76228.1 0.0000 37.5% 7.0% 17 40315 25961.5 0.0000 37.5% 6 Dis 6 40309 75604.8 0.0000 4.1% 7.7% 6 40309 25795.4 0.0000 3.6% 6 Area 1 40308 73773.8 0.0000 12.0% 10.0% 1 40308 25291.8 0.0000 10.8% 8 Month 11 40297 73468.5 0.0000 2.0% 10.3% 11 40297 25200.2 0.0000 2.0% 9 Year:Month 186 40111 67645 0.0000 38.2% 17.5% 186 40111 23447 0.0000 37.5% 15 Year:Area 17 40094 66687.5 0.0000 6.3% 18.6% 17 40094 23043.3 0.0000 8.6% 16 Delta model lognormal positive sets binomial probability of a non-zero catch d.f. Δ p-value Δ Dev. d.f. p-value Res. d.f. Res. Dev. Total Exp. Dev. Res. d.f. Res. Dev. Dev. Total Exp. D Null 8832 6755.21 1539 9921.03 Year 17 8815 6415.02 0.0000 30.0% 5.0% 17 1522 7109.63 0.0000 48.4% 28 Dis 6 8809 6377.84 0.0000 3.3% 5.6% 6 1516 6861.15 0.0000 4.3% 30 Area 1 8808 6374.47 0.0664 0.3% 5.6% 1 1515 6282.89 0.0000 10.0% 36 Month 11 8797 6313.65 0.0000 5.4% 6.5% 11 1504 6186.67 0.0000 1.7% 37 Year:Month 177 8620 5664.99 0.0000 57.1% 16.1% 186 1318 4647.89 0.0000 26.5% 53 Year:Area 17 8603 5619.56 0.0002 4.0% 16.8% 17 1301 4115.17 0.0000 9.2% 58
Table 3- Nominal and standardized (not scaled) CPUE values. The standard errors (SE) and coefficient of variation (CV) are provided for Poisson and negative binomial models. Poisson negative binomial Delta-lognormal Year Nominal CPUE (not CPUE CPUE (not scaled) SE CV CPUE (not scaled) SE CV scaled) 1986 0.230 0.366 0.144 39% 0.466 0.156 33% 0.284 1987 0.545 0.384 0.150 39% 0.418 0.154 37% 0.312 1988 0.299 0.343 0.141 41% 0.207 0.085 41% 0.233 1989 0.724 1.154 0.260 23% 1.232 0.354 29% 0.941 1990 1991 0.273 0.252 0.090 36% 0.227 0.069 30% 0.219 1992 0.355 0.447 0.084 19% 0.344 0.065 19% 0.340 1993 1994 0.144 0.222 0.082 37% 0.176 0.056 32% 0.178 1995 0.309 0.203 0.065 32% 0.175 0.046 26% 0.151 1996 0.528 0.726 0.144 20% 1.393 0.288 21% 0.358 1997 0.656 0.561 0.129 23% 0.578 0.135 23% 0.430 1998 0.500 0.655 0.095 15% 0.649 0.100 15% 0.467 1999 0.280 0.304 0.041 14% 0.272 0.032 12% 0.219 2000 0.334 0.406 0.053 13% 0.356 0.043 12% 0.284 2001 0.491 0.497 0.047 9% 0.458 0.042 9% 0.350 2002 0.410 1.508 0.261 17% 3.506 0.677 19% 1.424 2003 2004 0.950 2.170 0.173 8% 2.590 0.348 13% 1.590 2005 1.063 1.603 0.167 10% 2.759 0.413 15% 1.183 2006 0.554 1.700 0.295 17% 1.627 0.441 27% 1.159
Figure 1- Distribution of sum of fishing effort done by the Brazilian tuna longline fishery, in the Atlantic Ocean, from 1986 to 2007. Figure 2- Annual proportion of sets with and without sailfish catches between 1986 and 2006 (gray areas) and percentage of sailfish catches The years of 1993 and 2003 were excluded because of their low fishing effort.
Figure 3- Distribution of residuals for the GLM analysis for the Poisson, Negative binomial and Deltalognormal (positive catches only). Poisson negative binomial Delta-log (positive sets) Delta-log (proportion of positives) Figure 4- Residual vs. quartiles of standard normal for Poisson, negative binomial, and Delta (positive sets and proportion of positive) models.
Figure 5- Partial residuals for each variable included in the Poisson model. Figure 6- Partial residuals for each variable included in the negative binomial model.
Figure 7- Partial residuals for each variable included in the Delta-lognormal model (positive sets). Figure 8- Partial residuals for each variable included in the Delta-lognormal model (proportion of positive).
Figure 9- Nominal and Standardized catch rates for Sailfish for the Poisson and negative binomial approaches between 1986 and 2007. Diamonds are nominal and lines are Standardized CPUE series. Vertical bars are 95%confidence intervals for Standardized series. Figure 10- Scaled nominal (diamonds) and standardized (lines) sailfish CPUE series for three error distribution assumptions.