SCRS/2011/139 Collect. Vol. Sci. Pap. ICCAT, 68(3): (2012)

Transcription

1 SCRS/2011/139 Collect. Vol. Sci. Pap. ICCAT, 68(3): (2012) A STANDARDIZED CATCH RATE INDEX FOR YELLOWFIN TUNA (THUNNUS ALBACARES) FROM THE U.S. RECREATIONAL FISHERY IN THE WESTERN NORTH ATLANTIC OCEAN, Shannon L. Cass-Calay 1 SUMMARY Catch and effort data from the U. S. Marine Recreational Fisheries Statistical Survey (MFRSS) off the Atlantic coast and Gulf of Mexico (excluding Texas) were used to construct an index of abundance for yellowfin tuna. Standardized catch rates were estimated using a Generalized Linear Mixed modeling approach assuming a delta-lognormal error distribution. The explanatory variables considered for standardization included: year, geographic area, season, and fishing mode (a factor that classifies recreational fishing as charter or private/rental boat). The indices suggest that the catch rates of yellowfin have declined to a historic low in the most recent years ( ). It is not clear whether this is an effect of reduced availability of yellowfin to the U.S. Recreational Fishery or a true reflection of a decrease in stock abundance. A reduction in availability could occur if yellowfin tuna were distributed further offshore than the recreational boats can reach. RÉSUMÉ Les données de prise et d'effort de l'enquête statistique des pêcheries récréatives marines des États-Unis (MFRSS) opérant au large du littoral atlantique et du golfe du Mexique (Texas exclus) ont été utilisées pour obtenir un indice d'abondance de l'albacore. Les taux de capture standardisés ont été estimés en utilisant une approche de modèle linéaire généralisé mixte postulant une distribution d erreur delta lognormale. Les variables explicatives considérées pour la standardisation incluaient : année, zone géographique, saison et mode de pêche (facteur qui classe la pêche récréative selon que l'embarcation est affrétée ou privée/en location). Les indices suggèrent que les taux de capture de l'albacore ont chuté à un faible historique au cours de ces toutes dernières années ( ). On ne sait pas au juste si cela est dû à la disponibilité réduite de l'albacore à la pêcherie récréative des États-Unis ou si cela reflète réellement une baisse de l'abondance du stock. Une réduction de la disponibilité pourrait survenir si l'albacore était distribué plus loin des côtes, hors de portée des bateaux récréatifs. RESUMEN Se utilizaron los datos de captura y esfuerzo de la prospección estadística de las pesquerías de recreo marítimas de Estados Unidos (MFRSS) en aguas de la costa atlántica y del Golfo de México (con la exclusión de Tejas) para obtener el índice de abundancia para el rabil. Se estimaron las tasas de captura estandarizadas mediante modelos lineales mixtos generalizados asumiendo una distribución de error delta-lognormal. Las variables explicativas consideradas para la estandarización incluían: año, zona geográfica, temporada y modo de pesca (un factor que clasifica la pesca de recreo en barco fletado o buque privado/alquilado). Los índices sugieren que las tasas de captura de rabil han descendido hasta un nivel bajo histórico en los años más recientes ( ). No está claro si esto es un efecto de una reducción en la disponibilidad de rabil para la pesquería de recreo estadounidense o si se trata de un reflejo real de una disminución de la abundancia del stock. Podría producirse un descenso en la disponibilidad si los rabiles se distribuyesen en una zona más alejada de la costa a la que no pueden acceder los barcos de recreo. KEY WORDS Catch/effort, abundance, MRFSS, recreational statistics, multivariate analyses 1 U.S. Department of Commerce, NOAA Fisheries, Southeast Fisheries Science Center, Miami Laboratory, 75 Virginia Beach Drive, Miami, Florida U.S.A. Shannon.Calay@noaa.gov 953

2 1. Introduction Data collected and estimated by the Marine Recreational Fisheries Statistical Survey (MRFSS) were used to develop standardized catch per unit effort (CPUE) indices for yellowfin tuna in the Gulf of Mexico and western North Atlantic. The MRFSS survey began in 1979, and its purpose was to establish a reliable database for estimating the impact of marine recreational fishing on marine resources. More detailed information on the methods and protocols of the survey can be found at overview.html. 2. Materials and methods The Marine Recreational Fisheries Statistical Survey (MRFSS) program provides estimates of catch and effort for the U.S. recreational fishery. Data were collected by scientific samplers during dockside interviews. Each record includes the following information: catch (by species) in numbers, and whether the catch was retained, released alive or discarded dead, the number of participating anglers, the number of fishing hours, information on gear used, target species, mode (shore, headboat, charter, or private/rental), area (inshore, ocean < 3 miles, 3 < ocean < 10 miles, ocean > 10 miles), county/state, and date. One potential problem with indices derived from the MRFSS database is the selection of trips/interviews that are relevant to the analysis. The MRFSS program includes information from recreational trips by shore anglers, inshore fishing trips, as well as large charter vessels fishing offshore. The task is then to identify trips that had a significant probability of catching yellowfin tuna. During the interview, anglers are asked which species were targeted during the fishing trip (primary and secondary target). For the yellowfin index, trips were included in the analysis if the primary or secondary target was a member of the group of large pelagic fish (Table 1). However, because a substantial proportion of trips do not report any target, trips were also included if they caught at least one species that occurred on at least 1% of trips that targeted a large pelagic fish (Table 2). As described subsequently, additional criteria were applied to further limit trips. The MRFSS data includes estimates of catch and effort from 1981 through 2007 from the U.S. States of Louisiana through Maine. Because very few trips reported catching yellowfin tuna before 1986, the indices were constructed for the period Nearly all yellowfin (>99%) were landed using hook and line. Therefore, the indices were constructed using only hook and line trips. Additionally, shore and shelf effort were excluded from the analyses because it is unlikely to land a yellowfin tuna from a dock or shoreline, or within the shallow continental shelf. Finally, because headboat sampling is not reported consistently in the dataset in time and space, that fishing mode was also excluded from the analysis. Effort was excluded in certain time/area combinations because fishing did not generally occur there, or was not directed at tropical tunas. In the northeastern and mid Atlantic U.S. (CT, RI, MA, NH, ME, DE, NJ, NY, VA, MD) fishing effort that occurred during the winter and spring (Dec-May) were excluded from the analysis. The following factors were considered as possible influences on the proportion of trips that observed yellowfin tuna (proportion positive), and the catch rates on trips that caught yellowfin tuna. Because of the small number of records for some states, regional areas were defined and used as a spatial factor. Months were aggregated into seasons to account for seasonal fishery distribution through the year. Factor Levels Values YEAR SEASON 4 WIN = (Dec-Feb) SPR = (Mar-May) SUM = (Jun-Aug) AUT = (Sep-Nov) MODE 2 Charter (CB) and Private (PB) REGION 4 NE U.S. (CT, RI, MA, NH, ME, DE, NJ, NY) Mid Atlantic U.S. (VA, MD, NC) Southeast U.S. (FL East Coast, GA, SC) Gulf of Mexico (FL West Coast, AL, MS, LA) Catch per unit effort (CPUE) was defined as the total kept, discarded or released (AB1B2, in number of fish) per 1000 angler hours. 954

3 CPUE = (Number Landed + Discarded Dead + Released Alive) / 1000 Angler Hours A delta-lognormal approach (Lo et al. 1992) was used to develop the standardized catch rate indices. This method combines separate generalized linear modeling (GLM) analyses of the proportion positive trips (trips that caught yellowfin tuna) and the catch rates of successful trips to construct a single standardized index of abundance. Parameterization of each model was accomplished using a GLM procedure (GENMOD; Version 8.02 of the SAS System for Windows SAS Institute Inc. Cary, NC, USA). A forward stepwise regression procedure was used to determine the set of fixed factors and interaction terms that explained a significant portion of the observed variability. For both the binomial and lognormal portions of the delta-lognormal model, deviance tables were constructed to determine the proportion of total variance explained by the addition of each factor or interaction term. In addition, a χ2 analysis was performed to test the significance of the reduction in deviance between each consecutive set of nested models (McCullagh and Nelder 1989). Factors and interaction terms were selected for final analysis if: 1) the relative percent of deviance explained by adding the factor exceeded 5%, 2) the χ2 test was significant and 3) the Type-III test was significant for the specified model. Once a set of fixed factors was identified, the influence of the YEAR*FACTOR interactions were examined. As per the recommendation of the statistics and methods working group of the SCRS (1999), YEAR*FACTOR interaction terms were included in the model as random effects. Selection of the final mixed model was based on the Akaike s Information Criterion (AIC), Schwarz s Bayesian Criterion (BIC), and a chi-square test of the difference between the 2 log likelihood statistics between successive model formulations (Littell et al. 1996). The final delta-lognormal model was fit using the SAS macro GLIMMIX and the SAS procedure PROC MIXED (SAS Institute Inc. 1997) following the procedures described by Lo et al. (1992). 3. Results and discussion The GLM model construction, results and statistics are summarized in Tables 3-4 (binomial component) and Tables 5-6 (lognormal component). The final models selected were as follows. PPT = REGION+MODE+SEASON+YEAR+YEAR*REGION+YEAR*SEASON LOG(CPUE) = YEAR+SEASON+REGION+YEAR*SEASON+YEAR*REGION The analysis dataset included 103,719 trips that either targeted a large pelagic species, or caught at least one associated species. Of these, only 10,646 (10.3%) reported landing, discarding or releasing yellowing tuna. The annual proportion of positive trips (PPT: trips that caught yellowfin) was low, ranging from 3% to 21% (Figure 1, Table 7). PPT was generally less than 10% before 1992, and then increased to 10 to 21% until However, since that time, the proportion positive tips are the lowest on record, 3% to 5%. Nominal CPUE follows a very similar pattern (Figure 2, Table 7). The lowest levels were also observed during 2008 to Diagnostic plots were constructed to examine the fit of the components of the delta-lognormal model. The chisquare residuals, by factor, are shown in Figure 3. The strong effect of a few positive outliers is noted. The frequency distribution of the proportion of positive trips by strata (year, region, fishing mode and season) is shown in Figure 4. It is evident that most strata have low numbers of positive trips. To use the binomial model, it is generally recommended that at least 10-20% of trips observe the species of interest. Therefore, it is possible that the low proportion of positive trips violates the assumptions of this model component. The residuals of the lognormal model, by factor, are shown in Figure 5. In this case, the residuals are more evenly distributed above and below zero, supporting an appropriate fit of the lognormal component. The frequency distribution of nominal catch rates is shown in Figure 6. Ideally, the frequency distribution of log(cpue) should resemble the normal distribution overlaid in red (Figure 6). In this case, some departure from the expectation is noted, but the model fit appears adequate. The QQ-Plot (Figure 7) also indicates the degree of departure from the assumption of a normal distribution (red line). In this case, the QQ-Plot indicates an appropriate fit to the lognormal component of the delta-model. The standardized index and the nominal CPUE are shown in Figure 8 and Table 7. To facilitate comparison, both series were scaled by dividing the annual estimates by the series mean. The standardized index suggests that 955

4 catch rates of yellowfin tuna varied without obvious trend until Since that time a generally decline is noted. The most recent years ( ) are the lowest on record. It is not assured that this standardized index represents a true abundance trend of western Atlantic yellowfin tuna. The recent reduction in catch rates could be the result of a change in the distribution of yellowfin tuna that has altered their availability to the U.S. Recreational Fishery. For example, the fish could be distributed farther offshore than the fleet generally operates. Therefore, the reliability of this index should be carefully considered before inclusion in a quantitative stock assessment context. 4. Acknowledgments I would like to acknowledge the assistance of John F. Walter and Craig A. Brown of NOAA Fisheries (SEFSC) who provided advice regarding analytical techniques. 5. References Littell, R.C., Milliken, G.A., Stroup, W.W. and Wolfinger, R.D. 1996, SAS System for Mixed Models, Cary NC, USA:SAS Institute Inc., pp. Littell, R.C., Henry, P.R. and Ammerman, C.B. 1998, Statistical analysis of repeated measures data using SAS procedures. J. Anim. Sci. 76: Lo, N.C., Jacobson, L.D., and Squire, J.L. 1992, Indices of relative abundance from fish spotter data based on delta-lognormal models. Can. J. Fish. Aquat. Sci. 49: SAS Institute Inc. 1997, SAS/STAT Software: Changes and Enhancements through Release Cary, NC, USA: SAS Institute Inc., pp. Table 1. Members of the large pelagic species group. Genus - species Tuna genus Yellowfin tuna Dolphin Bluefin tuna Wahoo Billfish family Blue marlin White marlin King mackerel Mackerel family Blackfin tuna Bigeye tuna Albacore Sailfish Shortfin mako Skipjack tuna Common name Thunnus spp. Thunnus albacares Coryphaena hippurus Thunnus thynnus Acanthocybium solandri Istiophoridae Makaira nigricans Tetrapturus albidus Scomberomorus cavalla Scombridae Thunnus atlanticus Thunnus obesus Thunnus alalunga Istiophorus platypterus Isurus oxyrinchus Katsuwonus pelamis 956

5 Table 2. Species caught on at least 1% of trips that targeted a large pelagic species Common name Scientific name Percent occurrence on trips that targeted a large pelagic species Yellowfin tuna Thunnus albacares 15.7% Dolphin Coryphaena hippurus 13.0% Bluefin tuna Thunnus thynnus 10.8% Little tunny Euthynnus alletteratus 5.9% King mackerel Scomberomorus cavalla 4.3% Sailfish Istiophorus platypterus 4.2% Wahoo Acanthocybium solandri 3.3% Bluefish Pomatomus saltatrix 3.2% Blackfin tuna Thunnus atlanticus 2.9% Blue shark Prionace glauca 2.2% Great barracuda Sphyraena barracuda 2.1% Skipjack tuna Katsuwonus pelamis 1.6% Atlantic bonito Sarda sarda 1.5% White marlin Tetrapturus albidus 1.3% Ballyhoo Hemiramphus brasiliensis 1.2% Greater amberjack Seriola dumerili 1.1% Table 3. The deviance table for the binomial model on the proportion of positive trips. Factors were assumed to be significant if they explained >5% of the total deviance (shaded cells), and were significant according to a Chi- Square test. GENMOD (FIXED-FACTOR) OUTPUT Binomial Model Factors - Proportion Positive DF DF Residual Deviance Reduction in Deviance % of Total Deviance Log Like Chi Square Null Region <0.001 Region + Mode <0.001 Region + Mode + Season <0.001 Region + Mode + Season + Year <0.001 Region + Mode + Season + Year + Mode*Region <0.001 Region + Mode + Season + Year + Mode*Region + Season*Region <0.001 Region + Mode + Season + Year + Mode*Region + Season*Region + Season*Mode <0.001 Final Model: PPT = Region + Mode + Season + Year P Table 4. Analysis of the mixed model formulations for the binomial component of the delta-model. The likelihood ratio was used to test the difference of 2 REM log likelihood between two nested models. The final model is indicated with gray shading. Proportion Positive -2 REM Log likelihood Akaike's Information Criterion Schwartz's Bayesian Criterion Likelihood Ratio Test Region + Mode + Season + Year Region + Mode + Season + Year*Region < Region + Mode + Season + Year*Region + Year*Season < Region + Mode + Season + Year*Region + Year*Season + Year*Mode P 957

6 Table 5. The deviance table for the lognormal model on catch rates of positive trips. Factors were assumed to be significant if they explained >5% of the total deviance (shaded cells), and were significant according to a Chi- Square test. (NOTE: The model containing the term SEASON*REGION did not converge in GLMMIX to produce lsmeans for YEAR. Therefore, this factor was removed). Lognormal Model Factors - CPUE DF DF Residual Deviance Reduction in Deviance % of Total Deviance Log Like Chi Square Null Year <0.001 Year + Season <0.001 Year + Season + Region <0.001 Year + Season + Region + Mode <0.001 Year + Season + Region + Mode + Season*Region <0.001 Year + Season + Region + Mode + Season*Region + Season*Mode Year + Season + Region + Mode + Season*Region + Season*Mode + Mode*Region Final Model: log(cpue) = Year + Season + Region + Season*Region Table 6. YFT: Analysis of the mixed model formulations for the lognormal component of the delta-model. The likelihood ratio was used to test the difference of 2 REM log likelihood between two nested models. The final model is indicated with gray shading. P Catch Rates on Positive Trips -2 REM Log likelihood Akaike's Information Criterion Schwartz's Bayesian Criterion Likelihood Ratio Test P Year + Season + Region Year + Season + Region + Year*Season < Year + Season + Region + Year*Season + Year*Region < Table 7. Nominal CPUE, number of trips, number of positive trip, proportion positive trips (PPT), standardized index of abundance and index statistics. Year Nom CPUE Trips Pos trips PPT Relative index CV LCI UCI

7 Figure 1. YFT: Proportion of positive trips, by year. Figure 2. YFT: Nominal CPUE (fish per 1000 angler hours), by year. 959

8 A) B) C) D) Figure 3. Chi-square residuals for the fit to the binomial model, by year (A), region (B), fishing mode (C) and season (D). Figure 4. Frequency distribution of proportion positive trips by the strata year, region and fishing mode. 960

9 A) B) C) Figure 5. Residuals of the fit to the lognormal model, by year (A), region (B) and season (C). Figure 6. Frequency distribution of nominal catch rates (fish per 1000 angler hours) on positive trips. 961

10 Figure 7. The cumulative normalized residuals (QQ-Plot) from the lognormal model on the catch rates of positive trips. Figure 8. Nominal CPUE (blue line) and the delta-lognormal index (red line) with 95% confidence intervals (dashed lines). Both series are scaled to a mean of 1.0 to facilitate comparison. 962

11 Appendix 1 Distribution of observations by year, region, season, fishing mode (charter boat, private boat) and season Table A1. Trips that observed YFT by YEAR. Cumulative year Frequency Frequency ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ Table A2. Trips that observed YFT by SEASON. Cumulative SEASON Frequency Frequency ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ AUT SPR SUM WIN Table A3. Trips that observed YFT by REGION. Cumulative Region Frequency Frequency ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ GOM Mid_ATL NE_US SE_US Table A4. Trips that observed YFT by fishing MODE. Cumulative mode Frequency Frequency ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ CB PB

12 Table A5. Trips that observed YFT by fishing YEAR and REGION. Frequency GOM Mid_ATL NE_US SE_US Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total Table A6. Trips that observed YFT by fishing YEAR and fishing MODE. Frequency CB PB. Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total

13 Table A7. Trips that observed YFT by fishing YEAR and SEASON. Frequency AUT SPR SUM WIN Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total Table A8. Trips that observed YFT by fishing SEASON and REGION. Frequency GOM Mid_ATL NE_US SE_US Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ AUT SPR SUM WIN ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total Table A9. Trips that observed YFT by fishing SEASON and REGION. Frequency CB PB Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ AUT SPR SUM WIN ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total Table A10. Trips that observed YFT by fishing REGION and fishing MODE. Frequency CB PB Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ GOM Mid_ATL NE_US SE_US ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total

14 Table A11. Trips that observed YFT by STATE. Frequency LA MS AL FL W FL E GA SC NC Total ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total Continued Frequency VA MD DE NJ NY CT RI MA ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ ƒƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆƒƒƒƒƒƒƒƒˆ Total Total