SCRS/2001/107

Col.Vol.Sci.Pap. ICCAT, 54 (3): 791-804. (2002)

STANDARDIZED CATCH RATES FOR SAILFISH (Istiophorus platypterus) FROM THE PELAGIC LONGLINE FISHERY IN THE NORTHWEST ATLANTIC AND THE GULF OF MEXICO. Mauricio Ortiz and Craig A. Brown1

SUMMARY Indices of abundance of sailfish from the United States pelagic longline fishery in the Atlantic are presented for the period 1986-2000. The index of weight (kg) per number of hooks (thousand) was estimated from numbers of sailfish caught and reported in the logbooks submitted by commercial fisherman, and from mean annual weight estimated by scientific observers aboard longline (Pelagic Observer Program) vessels since 1992. The standardization analysis procedure included the following variables; year, area, season, gear characteristics (light sticks, main line length, hook density, etc) and fishing characteristics (bait type, operations procedure, and target species). The Pelagic Observer Program collects more detailed information that allowed evaluating the relationships between billfish catch rates and other fishing (hook type and size, main line material and size, rattlers, gangion size and material, etc) or environmental variables (sea-surface temperature, weather condition, wind) for the US longline fishery. The standardized index was estimated using Generalized Linear Mixed Models under a delta lognormal model approach. RÉSUMÉ Le présent document fait état des indices d=abondance du voilier en provenance de la pêche palangrière pélagique des Etats-Unis dans l=Atlantique pour les années 1986 à 2000. L=indice pondéral (kg) par nombre d=hameçons (milliers) a été estimé d=après le nombre de voiliers capturés et déclarés dans les carnets de pêche remis par les pêcheurs commerciaux, ainsi que le poids annuel moyen estimé par les observateurs scientifiques à bord des palangriers (Pelagic Observer Program) depuis 1992. La méthode analytique de standardisation comprenait les variables suivantes: année, zone, saison, caractéristiques des engins (bâtons lumineux, longueur de la ligne principale, densité des hameçons, etc.) et les caractéristiques de la pêche (type d=appât, processus opérationnel et espèce -cible). Le Pelagic Observer Program rassemble une information plus détaillée qui permet d=évaluer la relation entre le taux de capture des istiophoridés et d=autres variables concernant la pêche (type et dimensions des hameçons, matériau de la ligne principale, signaux acoustiques, dimension et matériau des avançons, etc.) ou l=environnement (température de surface, conditions météo, vent). L=indice standardisé a été estimé au moyen de modèles linéaires généralisés mixtes selon une approche delta-lognormale. RESUMEN Este documento presenta índices de abundancia del pez vela para el periodo 1986-2000 procedentes de la pesquería de palangre pelágico de Estados Unidos en el Atlántico. El índice de peso (kg) por número de anzuelos (miles) se ha estimado a partir del número de peces vela capturados y declarados en los cuadernos de pesca presentados por los pescadores comerciales, y a partir del peso medio anual estimado por los observadores científicos abordo de palangreros (Programa de Observadores Pelágicos) desde 1992. El procedimiento de análisis de la estandarización incluía las siguientes variables: año, área, estación, características del arte (bastones de luz, longitud de la línea madre, densidad de anuelos, etc) y características pesqueras (tipo de cebo, procedimiento de operaciones, y especie objetivo). 1

U.S. Department of Commerce National Marine Fisheries Service Southeast Fisheries Science Center, 75 Virginia Beach Drive, Miami, Florida 33149 U.S.A. Email: [email protected]

El Programa de Observadores Pelágicos recopiló información más detallada que permitió evaluar las relaciones entre las tasas de captura de marlines y otros tipos de pesca (tipo y tamaño de los anzuelos, material y tamaño de la línea madre, dispositivos sonoros, material y tamaño de la brazolada, etc.) o variables medioambientales (temperatura de la superficie del mar, condiciones atmosféricas, viento) para la pesquería de palangre estadounidense. El índice estandarizado se estimó utilizando modelos lineales mixtos generalizados según un enfoque del modelo delta lognormal. KEYWORDS Catch/effort, abundance, longlining, Fish catch statistics, By catch, logbooks, Multivariate analyses

1. INTRODUCTION: Information on the relative abundance of sailfish is necessary to tune stock assessment models. Data collected from the US longline fleet has been used to develop standardized catch per unit of effort (CPUE) indices of abundance for billfish species including blue and white marlin (Ortiz and Scott 2001). This report documents the analytical methods applied to the available US longline fleet data through 2000 and presents correspondent standardized CPUE indices for the western Atlantic sailfish stock unit. Catch in numbers and effort data were obtained from the Pelagic Longline Logbook reports data, while size information was gathered from the Pelagic Observer Program. The US longline fleet operates over a wide geographical range of the western North Atlantic Ocean and although billfish including sailfish are not now targeted nor landed by the US fleet, this bycatch constitutes a component of fishery mortality on these stocks that can be quantified. 2. MATERIALS AND METHODS: Hoey and Bertolino (1988) described the main features of the fleet and numerous authors (Hoey et al. 1989, Scott et al. 1993, Cramer and Bertolino 1998, Ortiz et al. 2000) have reviewed the available catch and effort data from the US Pelagic Longline fishery. Sailfish standardized catch rate indices were previously estimated for the 1994 stock assessment using Generalized Linear Models (GLM) (Farber 1994). The present report updates the catch and effort information through 2000 and includes analyses of variability associated with random factor interactions particularly for interactions that include the Year effect, following the suggestion of the statistics and methods working group of the SCRS in 1999. Logbook records from the US Longline Pelagic fleet have been collected since 1986. From 1986 to 1991, submission of logbooks was voluntary, and thereafter, submission of logbook reports became mandatory. Swordfish, yellowfin, and other tunas are the main target species for the US Pelagic Longline fleet. Marlins are not retained by the U.S. fleet, although catch records of these and other bycatch species are recorded on logbooks. Since 1992, trained observers have recorded detailed information on gear characteristics, fishing operations as well morphometric and biological information from a target sub-sample level of 5% of the US longline Pelagic effort (Lee and Brown 1998). These constitute the Pelagic Observer Program (POP) data, which provide size and weight information on marlins caught by longline operations. The POP data collects substantially more detailed fishing information, which permits evaluation of relationships between sailfish catch rates and additional factors, such as environmental (e.g. sea surface temperature, wind direction and intensity, and general weather conditions), gear configurations and characteristics (main line type and length; gangion type and length; hook type, size, and density per unit of main line; floats number and density; rattlers; light sticks; surface light-bouys; etc), and fishing operations (bait type, condition, and number; depth of set; soaking time; etc.).

The Pelagic Longline Logbook data comprises a total of 209,776 record-sets from 1986 through 2000. Each record contains information of catch by trip/set, including: date and time, geographical location, catch in numbers of targeted and bycatch species, and fishing effort (as number of hooks per set). Of these trips, sailfish were reported as being caught in 10,443 sets (5%). Figures 1 and 2 show the geographical distribution of mean nominal CPUE (numbers of fish per thousand hooks) by 1° latitude-longitude grouping, and the mean total number of hooks reported (i.e. fishing effort) per set for the same strata. Logbooks only record numbers of fish. As per the recommendation of the SCRS Billfish Species Group, indices of abundance should be reported both in weight and numbers of fish, when possible. In order to convert number of fish to weight, size information on blue and white marlin caught by the US longline fleet was retrieved from the POP. The POP covers about 5% of the total annual U.S. Atlantic pelagic longline trips, but POP data are available only since 1992. Figures 3 shows the size frequency distribution for sailfish, respectively from POP data and their respective mean and standard deviation by year. The number of fish measured was considered as being too small to estimate mean size in strata smaller than the year average. Conversion from mean annual size to weight used the current size-weight relationships for combined sex (Prager et al. 1995). For years prior to 1992, the mean size value from 1992 was applied. The longline fishing grounds for the US fleet extends from the Grand Banks in the North Atlantic to latitudes of 5-10° south, off the South America coast, including the Caribbean Sea and the Gulf of Mexico. Eight geographical areas of longline fishing were used for classification (Fig 4). These include: the Caribbean (CAR, area 1), Gulf of Mexico (GOM, area 2), Florida East coast (FEC, area 3), South Atlantic Bight (SAB, area 4), Mid-Atlantic Bight (MAB, area 5), New England coastal (NEC, area 6), Northeast distant waters (NED, or Grand Banks, area 7), the Sargasso Sea and the North central Atlantic (SNA, area 8) and Southern Offshore (OFS, area 10, ranging to 5°N latitude). However, in the case of sailfish, very small numbers of fish where reported from the North and central offshore areas (NED, SNA and OFS), thus these areas where combined into a single one for standardization purposes. Calendar quarters were used to account for seasonal fishery distribution through the year (Jan-Mar, Apr-Jun, Jul-Sep, and Oct-Dec). Other factors included in the analyses of catch rates included; the use of light-sticks and the density of light-sticks, type of bait (alive or dead), and a variable named operations procedure (OP), which is a categorical classification of US longline vessels based on their fishing configuration, type and size of the vessel, and main target species and area of operation(s). This variable has been shown to be significantly im portant as a predictor in the analyses of swordfish and marlins catch rates (Ortiz and Scott 2001, Ortiz et al. 2000). Fishing effort is reported in terms of the total number of hooks per trip and number of sets per trip. As number of hooks per set varies, catch rates were calculated as number of marlin caught per 1000 hooks. The longline fleet targets mainly swordfish and yellowfin tuna, but other tuna species are also targets including bigeye tuna and albacore (to a lesser extent, some of the trips-sets target other pelagic species including sharks, dolphin and small tunas). A target variable was defined based on the proportion of the number of swordfish caught to the total number of fish per set, with four discrete target categories corresponding to the ranges 0-25%, 25-50%, 50-75%, and 75-100%. As marlins are not targeted species by the US longline fleet, this measure of targeting was investigated to allow evaluation of targeting towards swordfish or tunas. As mentioned previously, the Pelagic Observer Program samples about 5% of the US longline fleet trips but collects significantly more detailed information compared with the logbook reports. This information includes specifics of gear configurations such as main line material, size, diameter, total length; hook type, size, and brand; light-sticks number and color; gangion size, material and length; leader material and size; rattlers; number and type of floats; number of hooks between floats; number of surface lights. Also, specifics about fishing configuration such as depth of the float-line, soak time, intended target species, bait type and number/weight of bait per set are recorded. Some general environmental information such as sea surface temperature at the beginning and end of the set and haul retrieval, wind speed and direction, estimated depth of hooks, bottom depth and general

weather condition (calm, storm, rain/snow, etc) are also recorded. The POP data includes 4,586 record-sets from 1992 through 2000, from which 613 sets caught sailfish. An exploratory analysis of the relationship between catch rates for blue and white marlin with several gear, fishing and environmental factors has been perform to identify other potentially significant effects that could account for variability of catch rates for billfish species, not consider in the PLL analysis (Ortiz and Scott 2000). For continuos variables (sea surface temperature, depth, main length, density of lightsticks per hook, density of hooks per unit of main line, gangion length, and distance between gangions) General Additive Models (GAMs) has been used to analyze the relative influence of various factors on catch rates for billfish including blue marlin and white marlin (Ortiz and Scott 2000), as well swordfish (Bigelow et al. 1999, Kleiber and Bartoo 1998), and blue sharks (Bigelow et al. 1999). For the PLL data, relative indices of abundance for sailfish were estimated by a GLM approach assuming a delta-lognormal model distribution. The delta model fits separately the proportion of positive sets assuming a binomial error distribution and the mean catch rate of sets where at least one marlin was caught assuming a lognormal error distribution. The standardized index is the product of these model-estimated components. The log-transformed frequency distribution for sailfish is shown in figure 5. The estimated proportion of successful sets per stratum is assumed to be the result of r positive sets of a total n number of sets, and each one is an independent Bernoulli-type realization. The estimated proportion is a linear function of fixed effects and interactions. The logit function was used as a link between the linear factor component and the binomial error. For sets that caught at least one sailfish (positive observations), estimated CPUE rates were assumed to follow a lognormal error distribution (lnCPUE) of a linear function of fixed factors and random effect interactions, particularly when the Year effect was within the interaction. For the pelagic observer program data, relative indices of abundance for sailfish were also estimated by a GLM approach assuming a delta lognormal distribution. For these data, the following factors were included in the analysis: year, area, OP (operations procedure), target species (as specified by the captain prior to the set), season (quarterly months), light-sticks (0, 0-0.75, and > 0.75 light-sticks per hook), hook density, rattlers, surface lights, main line material (1= nylon, 2 = others), hook manufacture (three categories), hook type (circle hooks, J-type hooks, and unknown), hook size (7/0-10/0, 11/0-16/0, and unknown), weather condition (Clear/cloudy, Rain/snow, Severe, Unknown), distance between gangions (< 180 ft, 180 ft), main line length ( < 30 NM, 30 NM), bait kind (including mackerel, herring, squid, sardine, scad, artificial lures, unknown, and several mixed combination of these types), and bait type (classifying sets as live bait only, dead bait, and mixed). A step-wise regression procedure was used to determine the set of systematic factors and interactions that significantly explained the observed variability. Because, the difference of deviance between two consecutive (nested) models follows a χ2 (Chi-square) distribution, this statistic was used to test for the significance of an additional factor in the model. The number of additional parameters associated with the added factor minus one corresponds to the number of degrees of freedom in the χ2 test (McCullagh and Nelder, 1989 pp 393). Deviance analysis tables are presented for both data series, each table includes the deviance for the proportion of positive observations (i.e. positive trips/total trips), and the deviance for the positive catch rates. Final selection of explanatory factors was conditional to: a) the relative percent of deviance explained by adding the factor in evaluation (normally factors that explained more than 5% were selected), b) the χ2 test of significance, and c) the Type-III test significance within the final specified model. Once a set of fixed factors was specified, possible interactions were evaluated, and in particular interactions between the Year effect and other factors. Selection of the final mixed model was based on the Akaike’s Information Criterion (AIC), Schwarz’s Bayesian Criterion (SBC), and a chi-square test of the difference between the [–2 loglikelihood statistic] between successive model formulations (Littell et al. 1996). Relative indices for the delta model formulation were calculated as the product of the year effect least square means (LSmeans) from the binomial and the lognormal model

components. The LSmeans estimates use a weighted factor of the proportional observed margins in the input data to account for the un-balanced characteristics of the data. LSmeans of lognormal positive trips were bias corrected using Lo et al., (1992) algorithms. Analyses were done using the GLIMMIX and MIXED procedures from the SAS statistical computer software (SAS Institute Inc. 1997). 3. RESULTS AND DISCUSSION As with blue and white marlin, the analyses of sailfish catch and the Pelagic Observer Program data should be considered as an exploratory evaluation of relationships between catch rates of marlins and diverse factors associated to the fishing operations (Ortiz and Scott 2001). A main restriction in this analysis is the low percentage of sets with positive sailfish catch, which is characteristic of incidental catch species in the longline fishery. We opted to use the delta approach, thereby restricting the analyses to positive catch set for blue and white marlin rather than add constant positive values to nominal CPUEs to avoid undefined logarithm transformation of zero CPUEs. Bigelow et al (1999) have used GAMs to examine influence of various factors on catch rates of swordfish (target species) and blue shark (bycatch species) on the U.S. North Pacific longline fishery. They added a constant value to avoid zero CPUEs, but their percentage of zero observations was only 25% for blue shark, compared to the 95% that we have for sailfish in the POP data. The objective of conducting the GAM analyses was primarily to choose from a wide array of fishing conditions, gear specifications and environmental variables; those that were more significantly associated with sailfish catch rates. Factor selection and categorization follow the same analysis as done with blue and white marlins (Ortiz and Scott 2001). Table 1 shows the deviance analysis for sailfish from the Pelagic Observer Program data analysis. In the case of sailfish, the fixed effects of area, season and kind of bait were the major factors that explained the probability of capture of at least one sailfish. For the mean catch rate on positive sets, the fixed effects of area, OP, season, light-sticks, hook type and bait kind, and the interactions year*area and year*season were more significant. Once a set of fixed factors was selected, we evaluated first level random interaction between the year and other effects. Table 2 shows the results from the random test analyses. All three-selection criteria used (AIC, SBC and 2 residual log likelihood) showed agreement for the best model selection. The deviance analyses of the Pelagic Longline Logbook data are show in Table 3. For sailfish the proportion of positive sets was explained by the area, season, target and the interaction of year*area. The mean catch rate for sets with sailfish catch was best explained by the main effects of area, season, OP, and light-sticks plus the interactions year*area, year*OP. All interactions that included the year factor were treated as random interactions; Table 4 shows the results of the mixed model (fixed factors and random interactions) and the information criteria used for evaluation. The comparison of the model results from the Observer Program and the Longline Logbook data show that for sailfish the proportion of positive sets is best explained by the main factors: area, season and target2. The Observer data suggest the use of bait kind as explanatory variable, however it is possible that this factor is confounded with the target2, as the selection of bait is determined by the species targeted. In the case of positive sets the main factors of area, season, OP, and light-sticks were most important in the PLL data; the Observer data suggested that hook type is also correlated with catch rates. Also, the observer data suggest bait kind as explanatory variable. Standardized CPUE series for sailfish are shown in Table 5 and Figure 6. Coefficients of variation for the sailfish analysis of the PLL data range from 15 to 33%. We also plotted the estimated CPUE series from the observer data and compared it with the corresponding pattern from the Logbook data (Fig 7). Overall for sailfish both series agreed, although the confidence intervals for the Observer’s CPUE series are much larger.

For comparison, standardized CPUE were also estimated using number of fish per thousand hooks as dependent variable in the Pelagic Longline Logbook dataset. Model formulations were exactly the same as the final models for the weight analyses in terms of explanatory variables and interactions (Table 5). Overall the trends were similar to the ones observed in the weight CPUE series. In order to have a more valid comparison, both weight and number of fish CPUE series were normalized to a mean zero and one standard deviation (Fig 8). For sailfish, the weight and number CPUE series follow similar trend, the major difference between the series occurs in 1997. LITERATURE CITED BIGELOW, K.A., C.H. Boggs, and X. He. 1999. Environmental effects on swordfish and blue shark catch rates in the US North Pacific longline fishery. Fish. Oceanogr. 8(3): 178-198. CRAMER, J. 1998. Large Pelagic Logbook catch rate indices for Billfish. -Col. Vol. Sci. Pap. ICCAT, XLVII:301-307. CRAMER, J. and A. Bertolino. 1998. Standardized catch rates for swordfish (Xiphias gladius) from the U.S. longline fleet through 1997. Col. Vol. Sci. Pap. ICCAT, XLIX(1):449-456. FARBER . 1994. HOEY, J.J. and A. Bertolino. 1988. Review of the U.S. fishery for swordfish, 1978 to 1986. Col. Vol. Sci. Pap. ICCAT, XXVII:256-266. HOEY, J.J., R. Conser and E. Duffie. 1989. Catch per unit effort information from the U.S. swordfish fishery. Col. Vol. Sci. Pap. ICCAT, XXIX:195-249. KLEIBER, P. and N. Bartoo. 1998. Standardizing swordfish, Xiphias gladius, Longline catch per unit of effort using General Additive Models. Pages 181-193 in NOAA Tech. Rep. NMFS 142: Biology and Fisheries of Swordfish Xiphias gladius. Papers from the International Symposium on Pacific swordfish, Ensenada, Mexico, Dec. 1994. LEE, D.W. and C.J. Brown. 1998. SEFSC Pelagic Observer Program Data Summary for 1992-1996. NOAA Tech. Memo. NMFS-SEFSC-408, 21 pp. LITTELL, R.C., G.A. Milliken, W.W. Stroup, and R.D Wolfinger. 1996. SAS® System for Mixed Models, Cary NC, USA:SAS Institute Inc., 1996. 663 pp. LO, N.C., L.D. Jacobson, and J.L. Squire. 1992. Indices of relative abundance from fish spotter data based on delta-lognormal models. Can. J. Fish. Aquat. Sci. 49: 2515-2526. MCCULLAGH , P. and J.A. Nelder. 1989. Generalized Linear Models 2nd edition. Chapman & Hall. ORTIZ, M. J. Cramer, A. Bertolino and G. P. Scott. 2000. Standardized catch rates by sex and age for swordfish (Xiphias gladius) from the U.S. Longline Fleet 1981-1998. Col. Vol. Sci. Pap. ICCAT, LI:1559-1620. ORTIZ, M and G. P. Scott. 2001. P RAGER, M.H., E. D. Prince and D. W. Lee. 1995. Empirical length and weight conversion equation: for blue marlin, white marlin, and sailf ish from the North Atlantic Ocean. Bull of Mar. Sci. 56(1):201-210. SAS Institute Inc. 1997, SAS/STAT® Software: Changes and Enhancements through Release 6.12. Cary, NC, USA:Sas Institute Inc., 1997. 1167 pp. SCOTT, G. P., V. R. Restrepo and A. R. Bertolino. 1993. Standardized catch rates for swordfish (Xiphias gladius) from the US longline fleet through 1991. Col. Vol. Sci. Pap. ICCAT, XL(1):458-467.

Table 1. Deviance analysis table of explanatory variables in the delta lognormal model for sailfish catch rates from the Observer Pelagic Program data. Percent of total deviance refers to the deviance explained by the full model; p value refers to the 5% Chi-square probability between consecutive models.

SAILFISH OBSERVER PELAGIC PROGRAM DATA Model factors positive catch rates values

d. f.

1 YEAR … + AREA … + OP … + TARGETSP … + SEASON … + LGHTC … + RATLR … + SRFLITE … + MAINMAT … + HKBRAND … + HKTYPE … + HKSIZE … + WEATHERC … + GANGDISC … + BAITKND … + BAIT … + YEAR:SEASON … + YEAR:OP … + YEAR:AREA

Model factors proportion positive/total values

Residual deviance 1 8 6 7 4 3 2 1 1 1

3 2 2 3 1 15 2 22 32 22

d. f.

Residual deviance

1 YEAR … + AREA … + OP … + TARGETSP … + SEASON

1 8 6 7 4 3

… + LGHTC … + RATLR … + SRFLITE … + MAINMAT … + HKBRAND

2 1 1 1 3

… + HKTYPE … + HKSIZE … + WEATHERC … + GANGDISC … + BAITKND … + BAIT … + YEAR:SEASON … + YEAR:OP … + YEAR:AREA

413.72 404.68 374.33 352.06 348.67 330.61 317.20 316.82 313.12 312.35 310.22 299.03 297.71 295.91 291.61 258.30 255.39 227.90 200.93 191.31

2 2 3 1 18 2 23 52 46

3568.40 3510.36 3011.47 2943.50 2921.79 2718.47 2691.08 2680.65 2678.41 2676.61 2654.02 2642.68 2617.18 2615.88 2610.50 2514.10 2513.97 2423.26 2247.67 2156.13

Change in deviance

9.0 30.4 22.3 3.4 18.1 13.4 0.4 3.7 0.8 2.1 11.2 1.3 1.8 4.3 33.3 2.9 27.5 27.0 9.6

Change in deviance

58.0 498.9 68.0 21.7 203.3 27.4 10.4 2.2 1.8 22.6 11.3 25.5 1.3 5.4 96.4 0.1 90.7 175.6 91.5

% of total deviance

p

4.1% 0.339 13.6% < 0.001 10.0% 0.002 1.5% 0.494 8.1% < 0.001 6.0% 0.001 0.2% 0.540 1.7% 0.054 0.3% 0.381 1.0% 0.546 5.0% 0.004 0.6% 0.518 0.8% 0.614 1.9% 0.038 15.0% 0.004 1.3% 0.233 12.4% 0.193 12.1% 0.719 4.3% 0.989

% of total deviance

4.1% 35.3% 4.8% 1.5% 14.4% 1.9% 0.7% 0.2% 0.1% 1.6% 0.8% 1.8% 0.1% 0.4% 6.8% 0.0% 6.4% 12.4% 6.5%

p

< 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 0.001 0.134 0.181 < 0.001 0.003 < 0.001 0.729 0.020 < 0.001 0.933 < 0.001 < 0.001 < 0.001

Table 2. Analyses of delta lognormal mixed model formulations for sailfish catch rates from the Observer Pelagic Program data. Likelihood ratio tests the difference of –2 REM log likelihood between two nested models. * indicates the selected model for each component of the final delta mixed model.

Sailfish

-2 REM Akaike's Schwartz's Log Information Bayesian likelihood Criterion Criterion

Likelihood Ratio Test

Proportion Positives Year Area Season Op Year Area Season Op Year*Area Year Area Season Op Year*Area Year*Op

5120.1 4858.8 4847.9

-2561 -2431.4 -2427

-2563.4 -2433.5 -2430.1

261.3 10.9

0.0000 0.0010

Positive Catch Year Area Op Season LightStick Baitknd Year Area Op Season LightStick Baitknd Year*Area Year Area Op Season LightStick Baitknd Year*Area Year*Season Year Area Op Season LightStick Baitknd Year*Area Year*Season Year*Op

1335.5 1309.8 1298.3 1286.3

-668.7 -656.9 -650.6 -646.1

-670.9 -658.7 -653.3 -648.8

25.7 11.5 12

0.0000 0.0007 0.0005

Table 3. Deviance analysis table of explanatory variables in the delta lognormal model for sailfish catch rates from the Pelagic Longline Logbook data. Percent of total deviance refers to the deviance explained by the full model; p value refers to the 5% Chi-square probability between nested models. Pelagic Longline Logbook data for Sailfish Model factors positive catch rates values

Residual deviance

d. f.

1

Change in % of total deviance deviance

p

1

5982.65

14

5661.54

321.1

14.8%

< 0.001

YEAR AREA

8

4840.86

820.7

37.8%

< 0.001

YEAR AREA SEASON

3

4709.56

131.3

6.0%

< 0.001

YEAR AREA SEASON TARG2

3

4669.4 3

40.1

1.8%

< 0.001

YEAR AREA SEASON TARG2 OP

8

4533.35

136.1

6.3%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY

2

4493.79

39.6

1.8%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC

3

4330.74

163.1

7.5%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA

96

4119.94

210.8

9.7%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON

40

4081.97

38.0

1.7%

0.562

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON YEAR:OP

97

3912.36

169.6

7.8%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON YEAR:OP AREA:SEASON

20

3869.80

42.6

2.0%

0.002

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON YEAR:OP AREA:SEASON AREA:OP

34

3810.64

59.2

2.7%

0.005

YEAR

Model factors proportion positives

1

Residual deviance

d.f.

Change in % of total deviance deviance

P

1

23215.2

14

22955.3

259.9

1.6%

< 0.001

YEAR AREA

8

14807.0

8148.4

51.7%

< 0.001

YEAR AREA SEASON

3

12015.3

2791.6

17.7%

< 0.001

11

11669.8

345.5

2.2%

< 0.001

3

9563.5

2106.3

13.4%

< 0.001

111

8665.9

897.6

5.7%

< 0.001

41

8316.6

349.3

2.2%

< 0.001

118

7683.9

632.7

4.0%

< 0.001

42

7442.4

241.6

1.5%

< 0.001

YEAR

YEAR AREA SEASON OP YEAR AREA SEASON OP TARG2 YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR:SEASON YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR:SEASON YEAR:OP YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR:SEASON YEAR:OP YEAR:TARG2

Table 4. Analyses of delta lognormal mixed model formulations for sailfish catch rates from the Observer Pelagic Program data. Likelihood ratio tests the difference of –2 REM log likelihood between two nested models. * indicates the selected model for each component of the final delta mixed model. -2 REM Log likelihood

Sailfish Model

*

*

Schwartz's Bayesian Criterion

Akaike's Information Criterion

Likelihood Ratio Test

Proportion Positives Year Area Season Targ2 Year Area Season Targ2 Year*Area

41459.5 41591.4

-20730.8 -20797.7

-20734.2 -20800.6

-131.9

Positive Catch Year Area Season OP Ligths Year Area Season OP Ligths Year*Area Year Area Season OP Ligths Year*Area Year*OP

21062.5 20800.1 20639.1

-10532.3 -10402.1 -10322.5

-10535.9 -10404.9 -10326.7

262.4 161

#NUM!

0.0000 0.0000

Table 5. Nominal and standardized (delta lognormal mixed model) CPUE series (kg fish/1000 hooks) for the pelagic longline sailfish catch in the western Atlantic. The index column is the scaled to a maximum of standardized CPUE series. Nominal CPUE

Standard CPUE

1986

0.32

0.60

0.326

0.31

0.60

1.14

0.32

1987

0.55

0.41

0.176

0.11

0.41

0.58

0.29

1988

0.73

0.71

0.157

0.17

0.71

0.97

0.52

1989

0.57

0.68

0.157

0.17

0.68

0.93

0.50

1990

0.70

0.71

0.156

0.17

0.71

0.97

0.52

1991

0.78

0.81

0.156

0.20

0.81

1.11

0.60

1992

0.75

0.88

0.150

0.21

0.88

1.19

0.66

1993

1.00

1.00

0.151

0.23

1.00

1.35

0.74

1994

0.76

0.91

0.152

0.22

0.91

1.23

0.67

1995

0.38

0.47

0.160

0.12

0.47

0.64

0.34

1996

0.51

0.48

0.156

0.12

0.48

0.66

0.35

1997

0.63

0.63

0.157

0.15

0.63

0.86

0.46

1998

0.39

0.46

0.170

0.12

0.46

0.65

0.33

1999

0.63

0.60

0.167

0.16

0.60

0.84

0.43

2000

0.45

0.60

0.167

0.16

0.60

0.84

0.43

Year

Coeff Var

Std Error

Index

Upp CI 95%

Low CI 95%

Table 6. Nominal and standardized (delta lognormal mixed model) CPUE series (Numbers of fish/ 1000 hooks) of sailfish from the US Pelagic longline fishery. The index column is the scaled to a maximum of the standardized CPUE series. Year

Nominal Standardized CPUE CPUE

Coeff Var

Std Error

Index

Upp 9%5 CI

Low 95% CI 0.346

1986

0.352

0.066

32.8%

0.022

0.655

1.242

1987

0.602

0.045

17.8%

0.008

0.444

0.631

0.312

1988

0.800

0.078

15.8%

0.012

0.775

1.062

0.566

1989

0.624

0.075

15.8%

0.012

0.744

1.019

0.543

1990

0.767

0.079

15.7%

0.012

0.779

1.064

0.570

1991

0.850

0.090

15.6%

0.014

0.885

1.208

0.649

1992

0.817

0.098

15.1%

0.015

0.964

1.302

0.713

1993

1.000

0.101

15.1%

0.015

1.000

1.352

0.740

1994

0.755

0.091

15.3%

0.014

0.902

1.223

0.665

1995

0.390

0.048

16.2%

0.008

0.477

0.659

0.346

1996

0.501

0.048

15.8%

0.008

0.472

0.647

0.345

1997

0.500

0.050

16.0%

0.008

0.499

0.686

0.363

1998

0.350

0.042

17.4%

0.007

0.413

0.583

0.293

1999

0.560

0.055

17.0%

0.009

0.539

0.755

0.385

2000

0.466

0.063

16.8%

0.011

0.619

0.864

0.443

Figure 1 Distribution of sailfish CPUE from longline logbook data. Each symbol is proportional to the number of fish per 1000 hooks set in each 1°X 1° square. Only locations with at least 5 sets are included..

Figure 2 . Distribution of longline effort from logbook data. Each symbol is proportional to the number of hooks set in each 1°X 1° square. Only locations with at least 5 sets are included

1992

0.05

n = 27

0.00

1993

n = 273

250

0.05 0.00

1994

0.05

n = 117 1995

200 n = 99

0.05 0.00

1996

0.05

150

n = 182

0.00

Standard length (cm)

0.00

1997

n = 221

0.05 0.00

1998

100

n = 47

0.05 0.00

1999

n = 178

50

0.05 0.00

2000

n = 253

0.05 0.00 60

80

100

120

140

160

180

200

220

240

1992 1993 1994 1995 1996 1997 1998 1999 2000

Standard length (cm)

Figure 3 Size frequency distributions by year of sailfish caught by the US pelagic longline fishery fleet. Data summarize from the Observer Pelagic Program of the NMFS.

50° Northeast Distant Waters Northeast Coastal 40° Mid Atlantic Bight South Atlantic Bight 30° Florida East Coast Gulf of Mexico

Sargasso Sea & North Central Atlantic

20° Caribbean 10° Southern Offshore 0°

-100°

-90°

-80°

-70°

-60°

-50°

-40°

Figure 4 Geographical area classification for the US Pelagic longline fleet.

-30°

-20°

0.8

Sailfish Ln(CPUE)

0.6

0.4

0.2

0.0 2.24

2.75

3.25

3.75

4.25

4.76

5.26

5.76

6.27

6.77

7.27

Figure 5. Frequency distribution of log transformed nominal CPUE (kg/1000 hooks) values for trip/sets that caught sailfish from the US Pelagic Longline fleet from 1986 through 2000.

Sailfish Standardized CPUE (+ SE) Pelagic Longline US Fishery 1.6

CPUE (kg/1000 hooks)

1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Figure 6 Standardized (squares) and nominal (diamonds) CPUE for sailfish from the US Pelagic longline fishery. Error bars represents plus minus one standard error.

Sailfish standardized CPUE series

Scaled CPUE (kg/1000 hooks)

6

5

4

Pelagic Lonline Logbook

3

Pelagic Observer Program

2

1

0 1984

1986

1988

1990

1992

1994

1996

1998

2000

Year

Figure 7. Comparison between the standardized CPUE series from the Pelagic Logbook data and the Observer data for sailfish catch. Dotted lines represent 95% confidence bounds.

Comparison of Weight and Number CPUE index for Sailfish 2.5 2

Normalized Z values

1.5 1

Numb fish 0.5

Std N fish Wgt fish

0

Std Wgt Fish -0.5 -1 -1.5 -2 1985

1987

1989

1991

1993

1995

1997

1999

Year

Figure 8 Comparison of standardized CPUE series based on numbers of fish (solid line) or weight of fish (broken line) for sailfish. Series values were normalized to a mean zero a 1 standard deviation unit. Circles represent the nominal CPUE of numbers of fish per 1000 hooks and triangles represent the nominal CPUE in weight per 1000 hooks.

Col.Vol.Sci.Pap. ICCAT, 54 (3): 791-804. (2002)

STANDARDIZED CATCH RATES FOR SAILFISH (Istiophorus platypterus) FROM THE PELAGIC LONGLINE FISHERY IN THE NORTHWEST ATLANTIC AND THE GULF OF MEXICO. Mauricio Ortiz and Craig A. Brown1

SUMMARY Indices of abundance of sailfish from the United States pelagic longline fishery in the Atlantic are presented for the period 1986-2000. The index of weight (kg) per number of hooks (thousand) was estimated from numbers of sailfish caught and reported in the logbooks submitted by commercial fisherman, and from mean annual weight estimated by scientific observers aboard longline (Pelagic Observer Program) vessels since 1992. The standardization analysis procedure included the following variables; year, area, season, gear characteristics (light sticks, main line length, hook density, etc) and fishing characteristics (bait type, operations procedure, and target species). The Pelagic Observer Program collects more detailed information that allowed evaluating the relationships between billfish catch rates and other fishing (hook type and size, main line material and size, rattlers, gangion size and material, etc) or environmental variables (sea-surface temperature, weather condition, wind) for the US longline fishery. The standardized index was estimated using Generalized Linear Mixed Models under a delta lognormal model approach. RÉSUMÉ Le présent document fait état des indices d=abondance du voilier en provenance de la pêche palangrière pélagique des Etats-Unis dans l=Atlantique pour les années 1986 à 2000. L=indice pondéral (kg) par nombre d=hameçons (milliers) a été estimé d=après le nombre de voiliers capturés et déclarés dans les carnets de pêche remis par les pêcheurs commerciaux, ainsi que le poids annuel moyen estimé par les observateurs scientifiques à bord des palangriers (Pelagic Observer Program) depuis 1992. La méthode analytique de standardisation comprenait les variables suivantes: année, zone, saison, caractéristiques des engins (bâtons lumineux, longueur de la ligne principale, densité des hameçons, etc.) et les caractéristiques de la pêche (type d=appât, processus opérationnel et espèce -cible). Le Pelagic Observer Program rassemble une information plus détaillée qui permet d=évaluer la relation entre le taux de capture des istiophoridés et d=autres variables concernant la pêche (type et dimensions des hameçons, matériau de la ligne principale, signaux acoustiques, dimension et matériau des avançons, etc.) ou l=environnement (température de surface, conditions météo, vent). L=indice standardisé a été estimé au moyen de modèles linéaires généralisés mixtes selon une approche delta-lognormale. RESUMEN Este documento presenta índices de abundancia del pez vela para el periodo 1986-2000 procedentes de la pesquería de palangre pelágico de Estados Unidos en el Atlántico. El índice de peso (kg) por número de anzuelos (miles) se ha estimado a partir del número de peces vela capturados y declarados en los cuadernos de pesca presentados por los pescadores comerciales, y a partir del peso medio anual estimado por los observadores científicos abordo de palangreros (Programa de Observadores Pelágicos) desde 1992. El procedimiento de análisis de la estandarización incluía las siguientes variables: año, área, estación, características del arte (bastones de luz, longitud de la línea madre, densidad de anuelos, etc) y características pesqueras (tipo de cebo, procedimiento de operaciones, y especie objetivo). 1

U.S. Department of Commerce National Marine Fisheries Service Southeast Fisheries Science Center, 75 Virginia Beach Drive, Miami, Florida 33149 U.S.A. Email: [email protected]

El Programa de Observadores Pelágicos recopiló información más detallada que permitió evaluar las relaciones entre las tasas de captura de marlines y otros tipos de pesca (tipo y tamaño de los anzuelos, material y tamaño de la línea madre, dispositivos sonoros, material y tamaño de la brazolada, etc.) o variables medioambientales (temperatura de la superficie del mar, condiciones atmosféricas, viento) para la pesquería de palangre estadounidense. El índice estandarizado se estimó utilizando modelos lineales mixtos generalizados según un enfoque del modelo delta lognormal. KEYWORDS Catch/effort, abundance, longlining, Fish catch statistics, By catch, logbooks, Multivariate analyses

1. INTRODUCTION: Information on the relative abundance of sailfish is necessary to tune stock assessment models. Data collected from the US longline fleet has been used to develop standardized catch per unit of effort (CPUE) indices of abundance for billfish species including blue and white marlin (Ortiz and Scott 2001). This report documents the analytical methods applied to the available US longline fleet data through 2000 and presents correspondent standardized CPUE indices for the western Atlantic sailfish stock unit. Catch in numbers and effort data were obtained from the Pelagic Longline Logbook reports data, while size information was gathered from the Pelagic Observer Program. The US longline fleet operates over a wide geographical range of the western North Atlantic Ocean and although billfish including sailfish are not now targeted nor landed by the US fleet, this bycatch constitutes a component of fishery mortality on these stocks that can be quantified. 2. MATERIALS AND METHODS: Hoey and Bertolino (1988) described the main features of the fleet and numerous authors (Hoey et al. 1989, Scott et al. 1993, Cramer and Bertolino 1998, Ortiz et al. 2000) have reviewed the available catch and effort data from the US Pelagic Longline fishery. Sailfish standardized catch rate indices were previously estimated for the 1994 stock assessment using Generalized Linear Models (GLM) (Farber 1994). The present report updates the catch and effort information through 2000 and includes analyses of variability associated with random factor interactions particularly for interactions that include the Year effect, following the suggestion of the statistics and methods working group of the SCRS in 1999. Logbook records from the US Longline Pelagic fleet have been collected since 1986. From 1986 to 1991, submission of logbooks was voluntary, and thereafter, submission of logbook reports became mandatory. Swordfish, yellowfin, and other tunas are the main target species for the US Pelagic Longline fleet. Marlins are not retained by the U.S. fleet, although catch records of these and other bycatch species are recorded on logbooks. Since 1992, trained observers have recorded detailed information on gear characteristics, fishing operations as well morphometric and biological information from a target sub-sample level of 5% of the US longline Pelagic effort (Lee and Brown 1998). These constitute the Pelagic Observer Program (POP) data, which provide size and weight information on marlins caught by longline operations. The POP data collects substantially more detailed fishing information, which permits evaluation of relationships between sailfish catch rates and additional factors, such as environmental (e.g. sea surface temperature, wind direction and intensity, and general weather conditions), gear configurations and characteristics (main line type and length; gangion type and length; hook type, size, and density per unit of main line; floats number and density; rattlers; light sticks; surface light-bouys; etc), and fishing operations (bait type, condition, and number; depth of set; soaking time; etc.).

The Pelagic Longline Logbook data comprises a total of 209,776 record-sets from 1986 through 2000. Each record contains information of catch by trip/set, including: date and time, geographical location, catch in numbers of targeted and bycatch species, and fishing effort (as number of hooks per set). Of these trips, sailfish were reported as being caught in 10,443 sets (5%). Figures 1 and 2 show the geographical distribution of mean nominal CPUE (numbers of fish per thousand hooks) by 1° latitude-longitude grouping, and the mean total number of hooks reported (i.e. fishing effort) per set for the same strata. Logbooks only record numbers of fish. As per the recommendation of the SCRS Billfish Species Group, indices of abundance should be reported both in weight and numbers of fish, when possible. In order to convert number of fish to weight, size information on blue and white marlin caught by the US longline fleet was retrieved from the POP. The POP covers about 5% of the total annual U.S. Atlantic pelagic longline trips, but POP data are available only since 1992. Figures 3 shows the size frequency distribution for sailfish, respectively from POP data and their respective mean and standard deviation by year. The number of fish measured was considered as being too small to estimate mean size in strata smaller than the year average. Conversion from mean annual size to weight used the current size-weight relationships for combined sex (Prager et al. 1995). For years prior to 1992, the mean size value from 1992 was applied. The longline fishing grounds for the US fleet extends from the Grand Banks in the North Atlantic to latitudes of 5-10° south, off the South America coast, including the Caribbean Sea and the Gulf of Mexico. Eight geographical areas of longline fishing were used for classification (Fig 4). These include: the Caribbean (CAR, area 1), Gulf of Mexico (GOM, area 2), Florida East coast (FEC, area 3), South Atlantic Bight (SAB, area 4), Mid-Atlantic Bight (MAB, area 5), New England coastal (NEC, area 6), Northeast distant waters (NED, or Grand Banks, area 7), the Sargasso Sea and the North central Atlantic (SNA, area 8) and Southern Offshore (OFS, area 10, ranging to 5°N latitude). However, in the case of sailfish, very small numbers of fish where reported from the North and central offshore areas (NED, SNA and OFS), thus these areas where combined into a single one for standardization purposes. Calendar quarters were used to account for seasonal fishery distribution through the year (Jan-Mar, Apr-Jun, Jul-Sep, and Oct-Dec). Other factors included in the analyses of catch rates included; the use of light-sticks and the density of light-sticks, type of bait (alive or dead), and a variable named operations procedure (OP), which is a categorical classification of US longline vessels based on their fishing configuration, type and size of the vessel, and main target species and area of operation(s). This variable has been shown to be significantly im portant as a predictor in the analyses of swordfish and marlins catch rates (Ortiz and Scott 2001, Ortiz et al. 2000). Fishing effort is reported in terms of the total number of hooks per trip and number of sets per trip. As number of hooks per set varies, catch rates were calculated as number of marlin caught per 1000 hooks. The longline fleet targets mainly swordfish and yellowfin tuna, but other tuna species are also targets including bigeye tuna and albacore (to a lesser extent, some of the trips-sets target other pelagic species including sharks, dolphin and small tunas). A target variable was defined based on the proportion of the number of swordfish caught to the total number of fish per set, with four discrete target categories corresponding to the ranges 0-25%, 25-50%, 50-75%, and 75-100%. As marlins are not targeted species by the US longline fleet, this measure of targeting was investigated to allow evaluation of targeting towards swordfish or tunas. As mentioned previously, the Pelagic Observer Program samples about 5% of the US longline fleet trips but collects significantly more detailed information compared with the logbook reports. This information includes specifics of gear configurations such as main line material, size, diameter, total length; hook type, size, and brand; light-sticks number and color; gangion size, material and length; leader material and size; rattlers; number and type of floats; number of hooks between floats; number of surface lights. Also, specifics about fishing configuration such as depth of the float-line, soak time, intended target species, bait type and number/weight of bait per set are recorded. Some general environmental information such as sea surface temperature at the beginning and end of the set and haul retrieval, wind speed and direction, estimated depth of hooks, bottom depth and general

weather condition (calm, storm, rain/snow, etc) are also recorded. The POP data includes 4,586 record-sets from 1992 through 2000, from which 613 sets caught sailfish. An exploratory analysis of the relationship between catch rates for blue and white marlin with several gear, fishing and environmental factors has been perform to identify other potentially significant effects that could account for variability of catch rates for billfish species, not consider in the PLL analysis (Ortiz and Scott 2000). For continuos variables (sea surface temperature, depth, main length, density of lightsticks per hook, density of hooks per unit of main line, gangion length, and distance between gangions) General Additive Models (GAMs) has been used to analyze the relative influence of various factors on catch rates for billfish including blue marlin and white marlin (Ortiz and Scott 2000), as well swordfish (Bigelow et al. 1999, Kleiber and Bartoo 1998), and blue sharks (Bigelow et al. 1999). For the PLL data, relative indices of abundance for sailfish were estimated by a GLM approach assuming a delta-lognormal model distribution. The delta model fits separately the proportion of positive sets assuming a binomial error distribution and the mean catch rate of sets where at least one marlin was caught assuming a lognormal error distribution. The standardized index is the product of these model-estimated components. The log-transformed frequency distribution for sailfish is shown in figure 5. The estimated proportion of successful sets per stratum is assumed to be the result of r positive sets of a total n number of sets, and each one is an independent Bernoulli-type realization. The estimated proportion is a linear function of fixed effects and interactions. The logit function was used as a link between the linear factor component and the binomial error. For sets that caught at least one sailfish (positive observations), estimated CPUE rates were assumed to follow a lognormal error distribution (lnCPUE) of a linear function of fixed factors and random effect interactions, particularly when the Year effect was within the interaction. For the pelagic observer program data, relative indices of abundance for sailfish were also estimated by a GLM approach assuming a delta lognormal distribution. For these data, the following factors were included in the analysis: year, area, OP (operations procedure), target species (as specified by the captain prior to the set), season (quarterly months), light-sticks (0, 0-0.75, and > 0.75 light-sticks per hook), hook density, rattlers, surface lights, main line material (1= nylon, 2 = others), hook manufacture (three categories), hook type (circle hooks, J-type hooks, and unknown), hook size (7/0-10/0, 11/0-16/0, and unknown), weather condition (Clear/cloudy, Rain/snow, Severe, Unknown), distance between gangions (< 180 ft, 180 ft), main line length ( < 30 NM, 30 NM), bait kind (including mackerel, herring, squid, sardine, scad, artificial lures, unknown, and several mixed combination of these types), and bait type (classifying sets as live bait only, dead bait, and mixed). A step-wise regression procedure was used to determine the set of systematic factors and interactions that significantly explained the observed variability. Because, the difference of deviance between two consecutive (nested) models follows a χ2 (Chi-square) distribution, this statistic was used to test for the significance of an additional factor in the model. The number of additional parameters associated with the added factor minus one corresponds to the number of degrees of freedom in the χ2 test (McCullagh and Nelder, 1989 pp 393). Deviance analysis tables are presented for both data series, each table includes the deviance for the proportion of positive observations (i.e. positive trips/total trips), and the deviance for the positive catch rates. Final selection of explanatory factors was conditional to: a) the relative percent of deviance explained by adding the factor in evaluation (normally factors that explained more than 5% were selected), b) the χ2 test of significance, and c) the Type-III test significance within the final specified model. Once a set of fixed factors was specified, possible interactions were evaluated, and in particular interactions between the Year effect and other factors. Selection of the final mixed model was based on the Akaike’s Information Criterion (AIC), Schwarz’s Bayesian Criterion (SBC), and a chi-square test of the difference between the [–2 loglikelihood statistic] between successive model formulations (Littell et al. 1996). Relative indices for the delta model formulation were calculated as the product of the year effect least square means (LSmeans) from the binomial and the lognormal model

components. The LSmeans estimates use a weighted factor of the proportional observed margins in the input data to account for the un-balanced characteristics of the data. LSmeans of lognormal positive trips were bias corrected using Lo et al., (1992) algorithms. Analyses were done using the GLIMMIX and MIXED procedures from the SAS statistical computer software (SAS Institute Inc. 1997). 3. RESULTS AND DISCUSSION As with blue and white marlin, the analyses of sailfish catch and the Pelagic Observer Program data should be considered as an exploratory evaluation of relationships between catch rates of marlins and diverse factors associated to the fishing operations (Ortiz and Scott 2001). A main restriction in this analysis is the low percentage of sets with positive sailfish catch, which is characteristic of incidental catch species in the longline fishery. We opted to use the delta approach, thereby restricting the analyses to positive catch set for blue and white marlin rather than add constant positive values to nominal CPUEs to avoid undefined logarithm transformation of zero CPUEs. Bigelow et al (1999) have used GAMs to examine influence of various factors on catch rates of swordfish (target species) and blue shark (bycatch species) on the U.S. North Pacific longline fishery. They added a constant value to avoid zero CPUEs, but their percentage of zero observations was only 25% for blue shark, compared to the 95% that we have for sailfish in the POP data. The objective of conducting the GAM analyses was primarily to choose from a wide array of fishing conditions, gear specifications and environmental variables; those that were more significantly associated with sailfish catch rates. Factor selection and categorization follow the same analysis as done with blue and white marlins (Ortiz and Scott 2001). Table 1 shows the deviance analysis for sailfish from the Pelagic Observer Program data analysis. In the case of sailfish, the fixed effects of area, season and kind of bait were the major factors that explained the probability of capture of at least one sailfish. For the mean catch rate on positive sets, the fixed effects of area, OP, season, light-sticks, hook type and bait kind, and the interactions year*area and year*season were more significant. Once a set of fixed factors was selected, we evaluated first level random interaction between the year and other effects. Table 2 shows the results from the random test analyses. All three-selection criteria used (AIC, SBC and 2 residual log likelihood) showed agreement for the best model selection. The deviance analyses of the Pelagic Longline Logbook data are show in Table 3. For sailfish the proportion of positive sets was explained by the area, season, target and the interaction of year*area. The mean catch rate for sets with sailfish catch was best explained by the main effects of area, season, OP, and light-sticks plus the interactions year*area, year*OP. All interactions that included the year factor were treated as random interactions; Table 4 shows the results of the mixed model (fixed factors and random interactions) and the information criteria used for evaluation. The comparison of the model results from the Observer Program and the Longline Logbook data show that for sailfish the proportion of positive sets is best explained by the main factors: area, season and target2. The Observer data suggest the use of bait kind as explanatory variable, however it is possible that this factor is confounded with the target2, as the selection of bait is determined by the species targeted. In the case of positive sets the main factors of area, season, OP, and light-sticks were most important in the PLL data; the Observer data suggested that hook type is also correlated with catch rates. Also, the observer data suggest bait kind as explanatory variable. Standardized CPUE series for sailfish are shown in Table 5 and Figure 6. Coefficients of variation for the sailfish analysis of the PLL data range from 15 to 33%. We also plotted the estimated CPUE series from the observer data and compared it with the corresponding pattern from the Logbook data (Fig 7). Overall for sailfish both series agreed, although the confidence intervals for the Observer’s CPUE series are much larger.

For comparison, standardized CPUE were also estimated using number of fish per thousand hooks as dependent variable in the Pelagic Longline Logbook dataset. Model formulations were exactly the same as the final models for the weight analyses in terms of explanatory variables and interactions (Table 5). Overall the trends were similar to the ones observed in the weight CPUE series. In order to have a more valid comparison, both weight and number of fish CPUE series were normalized to a mean zero and one standard deviation (Fig 8). For sailfish, the weight and number CPUE series follow similar trend, the major difference between the series occurs in 1997. LITERATURE CITED BIGELOW, K.A., C.H. Boggs, and X. He. 1999. Environmental effects on swordfish and blue shark catch rates in the US North Pacific longline fishery. Fish. Oceanogr. 8(3): 178-198. CRAMER, J. 1998. Large Pelagic Logbook catch rate indices for Billfish. -Col. Vol. Sci. Pap. ICCAT, XLVII:301-307. CRAMER, J. and A. Bertolino. 1998. Standardized catch rates for swordfish (Xiphias gladius) from the U.S. longline fleet through 1997. Col. Vol. Sci. Pap. ICCAT, XLIX(1):449-456. FARBER . 1994. HOEY, J.J. and A. Bertolino. 1988. Review of the U.S. fishery for swordfish, 1978 to 1986. Col. Vol. Sci. Pap. ICCAT, XXVII:256-266. HOEY, J.J., R. Conser and E. Duffie. 1989. Catch per unit effort information from the U.S. swordfish fishery. Col. Vol. Sci. Pap. ICCAT, XXIX:195-249. KLEIBER, P. and N. Bartoo. 1998. Standardizing swordfish, Xiphias gladius, Longline catch per unit of effort using General Additive Models. Pages 181-193 in NOAA Tech. Rep. NMFS 142: Biology and Fisheries of Swordfish Xiphias gladius. Papers from the International Symposium on Pacific swordfish, Ensenada, Mexico, Dec. 1994. LEE, D.W. and C.J. Brown. 1998. SEFSC Pelagic Observer Program Data Summary for 1992-1996. NOAA Tech. Memo. NMFS-SEFSC-408, 21 pp. LITTELL, R.C., G.A. Milliken, W.W. Stroup, and R.D Wolfinger. 1996. SAS® System for Mixed Models, Cary NC, USA:SAS Institute Inc., 1996. 663 pp. LO, N.C., L.D. Jacobson, and J.L. Squire. 1992. Indices of relative abundance from fish spotter data based on delta-lognormal models. Can. J. Fish. Aquat. Sci. 49: 2515-2526. MCCULLAGH , P. and J.A. Nelder. 1989. Generalized Linear Models 2nd edition. Chapman & Hall. ORTIZ, M. J. Cramer, A. Bertolino and G. P. Scott. 2000. Standardized catch rates by sex and age for swordfish (Xiphias gladius) from the U.S. Longline Fleet 1981-1998. Col. Vol. Sci. Pap. ICCAT, LI:1559-1620. ORTIZ, M and G. P. Scott. 2001. P RAGER, M.H., E. D. Prince and D. W. Lee. 1995. Empirical length and weight conversion equation: for blue marlin, white marlin, and sailf ish from the North Atlantic Ocean. Bull of Mar. Sci. 56(1):201-210. SAS Institute Inc. 1997, SAS/STAT® Software: Changes and Enhancements through Release 6.12. Cary, NC, USA:Sas Institute Inc., 1997. 1167 pp. SCOTT, G. P., V. R. Restrepo and A. R. Bertolino. 1993. Standardized catch rates for swordfish (Xiphias gladius) from the US longline fleet through 1991. Col. Vol. Sci. Pap. ICCAT, XL(1):458-467.

Table 1. Deviance analysis table of explanatory variables in the delta lognormal model for sailfish catch rates from the Observer Pelagic Program data. Percent of total deviance refers to the deviance explained by the full model; p value refers to the 5% Chi-square probability between consecutive models.

SAILFISH OBSERVER PELAGIC PROGRAM DATA Model factors positive catch rates values

d. f.

1 YEAR … + AREA … + OP … + TARGETSP … + SEASON … + LGHTC … + RATLR … + SRFLITE … + MAINMAT … + HKBRAND … + HKTYPE … + HKSIZE … + WEATHERC … + GANGDISC … + BAITKND … + BAIT … + YEAR:SEASON … + YEAR:OP … + YEAR:AREA

Model factors proportion positive/total values

Residual deviance 1 8 6 7 4 3 2 1 1 1

3 2 2 3 1 15 2 22 32 22

d. f.

Residual deviance

1 YEAR … + AREA … + OP … + TARGETSP … + SEASON

1 8 6 7 4 3

… + LGHTC … + RATLR … + SRFLITE … + MAINMAT … + HKBRAND

2 1 1 1 3

… + HKTYPE … + HKSIZE … + WEATHERC … + GANGDISC … + BAITKND … + BAIT … + YEAR:SEASON … + YEAR:OP … + YEAR:AREA

413.72 404.68 374.33 352.06 348.67 330.61 317.20 316.82 313.12 312.35 310.22 299.03 297.71 295.91 291.61 258.30 255.39 227.90 200.93 191.31

2 2 3 1 18 2 23 52 46

3568.40 3510.36 3011.47 2943.50 2921.79 2718.47 2691.08 2680.65 2678.41 2676.61 2654.02 2642.68 2617.18 2615.88 2610.50 2514.10 2513.97 2423.26 2247.67 2156.13

Change in deviance

9.0 30.4 22.3 3.4 18.1 13.4 0.4 3.7 0.8 2.1 11.2 1.3 1.8 4.3 33.3 2.9 27.5 27.0 9.6

Change in deviance

58.0 498.9 68.0 21.7 203.3 27.4 10.4 2.2 1.8 22.6 11.3 25.5 1.3 5.4 96.4 0.1 90.7 175.6 91.5

% of total deviance

p

4.1% 0.339 13.6% < 0.001 10.0% 0.002 1.5% 0.494 8.1% < 0.001 6.0% 0.001 0.2% 0.540 1.7% 0.054 0.3% 0.381 1.0% 0.546 5.0% 0.004 0.6% 0.518 0.8% 0.614 1.9% 0.038 15.0% 0.004 1.3% 0.233 12.4% 0.193 12.1% 0.719 4.3% 0.989

% of total deviance

4.1% 35.3% 4.8% 1.5% 14.4% 1.9% 0.7% 0.2% 0.1% 1.6% 0.8% 1.8% 0.1% 0.4% 6.8% 0.0% 6.4% 12.4% 6.5%

p

< 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 0.001 0.134 0.181 < 0.001 0.003 < 0.001 0.729 0.020 < 0.001 0.933 < 0.001 < 0.001 < 0.001

Table 2. Analyses of delta lognormal mixed model formulations for sailfish catch rates from the Observer Pelagic Program data. Likelihood ratio tests the difference of –2 REM log likelihood between two nested models. * indicates the selected model for each component of the final delta mixed model.

Sailfish

-2 REM Akaike's Schwartz's Log Information Bayesian likelihood Criterion Criterion

Likelihood Ratio Test

Proportion Positives Year Area Season Op Year Area Season Op Year*Area Year Area Season Op Year*Area Year*Op

5120.1 4858.8 4847.9

-2561 -2431.4 -2427

-2563.4 -2433.5 -2430.1

261.3 10.9

0.0000 0.0010

Positive Catch Year Area Op Season LightStick Baitknd Year Area Op Season LightStick Baitknd Year*Area Year Area Op Season LightStick Baitknd Year*Area Year*Season Year Area Op Season LightStick Baitknd Year*Area Year*Season Year*Op

1335.5 1309.8 1298.3 1286.3

-668.7 -656.9 -650.6 -646.1

-670.9 -658.7 -653.3 -648.8

25.7 11.5 12

0.0000 0.0007 0.0005

Table 3. Deviance analysis table of explanatory variables in the delta lognormal model for sailfish catch rates from the Pelagic Longline Logbook data. Percent of total deviance refers to the deviance explained by the full model; p value refers to the 5% Chi-square probability between nested models. Pelagic Longline Logbook data for Sailfish Model factors positive catch rates values

Residual deviance

d. f.

1

Change in % of total deviance deviance

p

1

5982.65

14

5661.54

321.1

14.8%

< 0.001

YEAR AREA

8

4840.86

820.7

37.8%

< 0.001

YEAR AREA SEASON

3

4709.56

131.3

6.0%

< 0.001

YEAR AREA SEASON TARG2

3

4669.4 3

40.1

1.8%

< 0.001

YEAR AREA SEASON TARG2 OP

8

4533.35

136.1

6.3%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY

2

4493.79

39.6

1.8%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC

3

4330.74

163.1

7.5%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA

96

4119.94

210.8

9.7%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON

40

4081.97

38.0

1.7%

0.562

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON YEAR:OP

97

3912.36

169.6

7.8%

< 0.001

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON YEAR:OP AREA:SEASON

20

3869.80

42.6

2.0%

0.002

YEAR AREA SEASON TARG2 OP BAITTY LGHTC YEAR:AREA YEAR:SEASON YEAR:OP AREA:SEASON AREA:OP

34

3810.64

59.2

2.7%

0.005

YEAR

Model factors proportion positives

1

Residual deviance

d.f.

Change in % of total deviance deviance

P

1

23215.2

14

22955.3

259.9

1.6%

< 0.001

YEAR AREA

8

14807.0

8148.4

51.7%

< 0.001

YEAR AREA SEASON

3

12015.3

2791.6

17.7%

< 0.001

11

11669.8

345.5

2.2%

< 0.001

3

9563.5

2106.3

13.4%

< 0.001

111

8665.9

897.6

5.7%

< 0.001

41

8316.6

349.3

2.2%

< 0.001

118

7683.9

632.7

4.0%

< 0.001

42

7442.4

241.6

1.5%

< 0.001

YEAR

YEAR AREA SEASON OP YEAR AREA SEASON OP TARG2 YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR:SEASON YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR:SEASON YEAR:OP YEAR AREA SEASON OP TARG2 YEAR:AREA YEAR:SEASON YEAR:OP YEAR:TARG2

Table 4. Analyses of delta lognormal mixed model formulations for sailfish catch rates from the Observer Pelagic Program data. Likelihood ratio tests the difference of –2 REM log likelihood between two nested models. * indicates the selected model for each component of the final delta mixed model. -2 REM Log likelihood

Sailfish Model

*

*

Schwartz's Bayesian Criterion

Akaike's Information Criterion

Likelihood Ratio Test

Proportion Positives Year Area Season Targ2 Year Area Season Targ2 Year*Area

41459.5 41591.4

-20730.8 -20797.7

-20734.2 -20800.6

-131.9

Positive Catch Year Area Season OP Ligths Year Area Season OP Ligths Year*Area Year Area Season OP Ligths Year*Area Year*OP

21062.5 20800.1 20639.1

-10532.3 -10402.1 -10322.5

-10535.9 -10404.9 -10326.7

262.4 161

#NUM!

0.0000 0.0000

Table 5. Nominal and standardized (delta lognormal mixed model) CPUE series (kg fish/1000 hooks) for the pelagic longline sailfish catch in the western Atlantic. The index column is the scaled to a maximum of standardized CPUE series. Nominal CPUE

Standard CPUE

1986

0.32

0.60

0.326

0.31

0.60

1.14

0.32

1987

0.55

0.41

0.176

0.11

0.41

0.58

0.29

1988

0.73

0.71

0.157

0.17

0.71

0.97

0.52

1989

0.57

0.68

0.157

0.17

0.68

0.93

0.50

1990

0.70

0.71

0.156

0.17

0.71

0.97

0.52

1991

0.78

0.81

0.156

0.20

0.81

1.11

0.60

1992

0.75

0.88

0.150

0.21

0.88

1.19

0.66

1993

1.00

1.00

0.151

0.23

1.00

1.35

0.74

1994

0.76

0.91

0.152

0.22

0.91

1.23

0.67

1995

0.38

0.47

0.160

0.12

0.47

0.64

0.34

1996

0.51

0.48

0.156

0.12

0.48

0.66

0.35

1997

0.63

0.63

0.157

0.15

0.63

0.86

0.46

1998

0.39

0.46

0.170

0.12

0.46

0.65

0.33

1999

0.63

0.60

0.167

0.16

0.60

0.84

0.43

2000

0.45

0.60

0.167

0.16

0.60

0.84

0.43

Year

Coeff Var

Std Error

Index

Upp CI 95%

Low CI 95%

Table 6. Nominal and standardized (delta lognormal mixed model) CPUE series (Numbers of fish/ 1000 hooks) of sailfish from the US Pelagic longline fishery. The index column is the scaled to a maximum of the standardized CPUE series. Year

Nominal Standardized CPUE CPUE

Coeff Var

Std Error

Index

Upp 9%5 CI

Low 95% CI 0.346

1986

0.352

0.066

32.8%

0.022

0.655

1.242

1987

0.602

0.045

17.8%

0.008

0.444

0.631

0.312

1988

0.800

0.078

15.8%

0.012

0.775

1.062

0.566

1989

0.624

0.075

15.8%

0.012

0.744

1.019

0.543

1990

0.767

0.079

15.7%

0.012

0.779

1.064

0.570

1991

0.850

0.090

15.6%

0.014

0.885

1.208

0.649

1992

0.817

0.098

15.1%

0.015

0.964

1.302

0.713

1993

1.000

0.101

15.1%

0.015

1.000

1.352

0.740

1994

0.755

0.091

15.3%

0.014

0.902

1.223

0.665

1995

0.390

0.048

16.2%

0.008

0.477

0.659

0.346

1996

0.501

0.048

15.8%

0.008

0.472

0.647

0.345

1997

0.500

0.050

16.0%

0.008

0.499

0.686

0.363

1998

0.350

0.042

17.4%

0.007

0.413

0.583

0.293

1999

0.560

0.055

17.0%

0.009

0.539

0.755

0.385

2000

0.466

0.063

16.8%

0.011

0.619

0.864

0.443

Figure 1 Distribution of sailfish CPUE from longline logbook data. Each symbol is proportional to the number of fish per 1000 hooks set in each 1°X 1° square. Only locations with at least 5 sets are included..

Figure 2 . Distribution of longline effort from logbook data. Each symbol is proportional to the number of hooks set in each 1°X 1° square. Only locations with at least 5 sets are included

1992

0.05

n = 27

0.00

1993

n = 273

250

0.05 0.00

1994

0.05

n = 117 1995

200 n = 99

0.05 0.00

1996

0.05

150

n = 182

0.00

Standard length (cm)

0.00

1997

n = 221

0.05 0.00

1998

100

n = 47

0.05 0.00

1999

n = 178

50

0.05 0.00

2000

n = 253

0.05 0.00 60

80

100

120

140

160

180

200

220

240

1992 1993 1994 1995 1996 1997 1998 1999 2000

Standard length (cm)

Figure 3 Size frequency distributions by year of sailfish caught by the US pelagic longline fishery fleet. Data summarize from the Observer Pelagic Program of the NMFS.

50° Northeast Distant Waters Northeast Coastal 40° Mid Atlantic Bight South Atlantic Bight 30° Florida East Coast Gulf of Mexico

Sargasso Sea & North Central Atlantic

20° Caribbean 10° Southern Offshore 0°

-100°

-90°

-80°

-70°

-60°

-50°

-40°

Figure 4 Geographical area classification for the US Pelagic longline fleet.

-30°

-20°

0.8

Sailfish Ln(CPUE)

0.6

0.4

0.2

0.0 2.24

2.75

3.25

3.75

4.25

4.76

5.26

5.76

6.27

6.77

7.27

Figure 5. Frequency distribution of log transformed nominal CPUE (kg/1000 hooks) values for trip/sets that caught sailfish from the US Pelagic Longline fleet from 1986 through 2000.

Sailfish Standardized CPUE (+ SE) Pelagic Longline US Fishery 1.6

CPUE (kg/1000 hooks)

1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Figure 6 Standardized (squares) and nominal (diamonds) CPUE for sailfish from the US Pelagic longline fishery. Error bars represents plus minus one standard error.

Sailfish standardized CPUE series

Scaled CPUE (kg/1000 hooks)

6

5

4

Pelagic Lonline Logbook

3

Pelagic Observer Program

2

1

0 1984

1986

1988

1990

1992

1994

1996

1998

2000

Year

Figure 7. Comparison between the standardized CPUE series from the Pelagic Logbook data and the Observer data for sailfish catch. Dotted lines represent 95% confidence bounds.

Comparison of Weight and Number CPUE index for Sailfish 2.5 2

Normalized Z values

1.5 1

Numb fish 0.5

Std N fish Wgt fish

0

Std Wgt Fish -0.5 -1 -1.5 -2 1985

1987

1989

1991

1993

1995

1997

1999

Year

Figure 8 Comparison of standardized CPUE series based on numbers of fish (solid line) or weight of fish (broken line) for sailfish. Series values were normalized to a mean zero a 1 standard deviation unit. Circles represent the nominal CPUE of numbers of fish per 1000 hooks and triangles represent the nominal CPUE in weight per 1000 hooks.