SCRS/2007/090 Collect. Vol. Sci. Pap. ICCAT, 62(5): 15601572 (2008)
STANDARDIZATION OF CPUE SERIES OF PRIONACE GLAUCA AND ISUSRUS OXYRINCHUS CAUGHT BY BRAZILIAN LONGLINERS IN THE WESTERN
SOUTH ATLANTIC OCEAN, FROM 1978 TO 2006
Fabio Hazin, Humberto Hazin, Felipe Carvalho, Catarina Wor and Paulo Travassos
SUMMARY
In the present study, catch and effort data from 67,335 sets done by the Brazilian tuna longline fleet (national and chartered), in the southwestern Atlantic Ocean, from 1978 to 2006 (29 years), were analyzed. The CPUE of both species was standardized by a GLM, by two different approaches: in the first one, a negative binomial error structure (log link) was assumed, while in the second one, a more traditional deltalognormal model, assuming a binomial error distribution for the proportion of positive sets and a Gaussian error distribution for the positive blue and mako shark catches was applied. Both models were based on the following factors: quarter, year, area, cluster, and quarter*year. The mean SE for the negative binomial models were slightly smaller than those for the deltalognormal, indicating they might be a better option for the standardization of CPUE for species that have a high proportion of zeros, such as the mako and the blue sharks, particularly because of its higher simplicity. In spite of some year to year oscillation, the CPUE of both species seemed to be pretty stable for the entire period.
RÉSUMÉ
La présente étude analyse les données de prise et d’effort de 67.335 opérations réalisées par la flottille palangrière thonière du Brésil (nationale et affrétée), opérant dans le SudOuest de l’océan Atlantique de 1978 à 2006 (29 ans). La CPUE des deux espèces a été standardisée par un GLM selon deux approches différentes : la première approche a postulé une structure d’erreur négative binomiale (lien logarithme), tandis que la deuxième approche a appliqué un modèle deltalognormal plus traditionnel, postulant une distribution d’erreur binomiale pour la proportion des jeux positifs et une distribution d’erreur gaussienne pour les captures positives de requins peaux bleues et de requins taupes bleues. Les deux modèles se basaient sur les facteurs suivants : trimestre, année, zone, grappe et trimestre*année. L’erreur standard de la moyenne pour les modèles binomiaux négatifs s’est avérée légèrement inférieure à celle pour le modèle deltalognormal, ce qui indique qu’il pourrait s’agir d’une meilleure option pour la standardisation de la CPUE pour les espèces qui ont une forte proportion de zéros, telles que le requin taupe bleue et le requin peau bleue, surtout en raison de sa plus grande simplicité. Malgré une certaine oscillation d’année en année, la CPUE des deux espèces a semblé assez stable pour l’ensemble de la période.
RESUMEN
En el presente estudio se analizaron los datos de captura y esfuerzo de 67.335 lances de la flota palangrera atunera de Brasil (nacional y fletada) en el Atlántico sudoccidental desde 1978 a 2006 (29 años). La CPUE de ambas especies fue estandarizada mediante un GLM con dos enfoques diferentes: en el primero, se asumió una estructura de error binomial negativa (vínculo logarítmico), y en el segundo, un modelo delta lognormal más tradicional, se asumió una distribución de error binomial para la proporción de lances positivos y se aplicó una distribución de error Gaussiana para las capturas positivas de tintorera y marrajo dientuso. Ambos modelos se basaban en los siguientes factores: trimestre, año, área, conglomerado y trimestre*año. La SE media para los modelos binomiales negativos era ligeramente inferior que las del modelo delta lognormal, lo que indica que podrían ser una mejor opción para la estandarización de la CPUE para las especies que tengan una mayor proporción de ceros, como el marrajo dientuso y la tintorera, debido principalmente a su mayor sencillez. A pesar de alguna oscilación de año en año, la CPUE de ambas especies parecía ser bastante estable para todo el periodo.
KEYWORDS
CPUE, Prionace glauca, Isusrus oxyrinchus, longline
1. Introduction
The blue shark, Prionace glauca, is a pelagic oceanic species, with a worldwide distribution in temperate and tropical waters of the world oceans, being one of the most abundant elasmobranch in longline catches (Hazin, 2007). Along the Brazilian coast, blue shark are frequently caught by the longline fishery targeting tunas and swordfish (Hazin, 1990). Although much less abundant than the blue shark, the shortfin mako, Isurus oxyrinchus, is also a common epipelagic species found in tropical and warmtemperate seas (Compagno 1984). In spite of its relatively low catches, though, because of its high commercial value, together with the blue shark, it is one of the best recorded shark species in commercial operations (Clarke et al., 2006).
Since 1956, when the tuna longline fishery began in the South Atlantic, several changes in both gear design and structure, as well as in fishing operation and targeting strategies, have been observed, with a strong influence on catch composition (Amorim and Arfelli, 1984; Arfelli, 1996; Hazin, 1993; Hazin and Hazin, 1999; Menezes de Lima et al., 2000). Such changes, together with seasonal and environmental factors, may lead to strong variations in catchability, which, in turn, can introduce serious errors in the estimation of abundance indices (Fréon and Misund, 1999).
One way to overcome this bias is by standardizing the CPUE series by a Generalized Linear Model (GLM), incorporating the factors that are known to influence catchability (Gulland, 1983). Catch and effort databases, however, often include high proportions of records in which the catch is zero, even though effort is recorded to be nonzero. This is particularly the case for less abundant species and for bycatch species (Maunder and Punt, 2004), like the mako and blue sharks. In such cases, in order to standardize the CPUE by GLM, traditionally, a deltalognormal model is used, assuming different error distributions for the positive catches and for the proportion of positives. Another, less common, method is to assume a negative binomial distribution, using the CPUE as a discrete variable, rounded to integer values. In the present paper both approaches were followed and their results compared.
In 2004, ICCAT carried out a stock assessment of the blue and the shortfin mako sharks, for the first time. The main problem encountered then was the lack of accurate data, particularly of standardized CPUE series from the main longline fisheries, which account for the vast majority of pelagic shark catches in the Atlantic Ocean. During its 2006 meeting, ICCAT has recommended that a new stock assessment for the two species be carried out in 2008, emphasizing, therefore, the acute need for more detailed data, particularly on fishing effort and catches from the main fisheries targeting the species in the Atlantic Ocean. The main objective of the present paper, therefore, was to contribute information on the methods used to standardize CPUE series of pelagic sharks caught by the longline fishery in the Atlantic Ocean, as well as to provide a standardized CPUE series of Brazilian longliners that can be incorporated in the next stock assessment of these important shark species.
2. Material and methods
In the present study, catch data from 67,335 longline sets done by the Brazilian tuna longline fleet, including both national and chartered vessels, from 1978 to 2006 (29 years) were analysed. The longline sets were distributed in a wide area of the Equatorial and South Atlantic Ocean, ranging from 0^{o} to 060^{o}W of longitude, and from 07^{o}N to 50^{o}S of latitude (Figure 1). Data on fishing effort, number of fish caught, by species, and the geographic position at the beginning of each set (latitude and longitude), were obtained from the logbooks, filled in by the skippers of the vessels. The resolution of 1º latitude x 1º longitude, per fishing day, was used for the analysis of the geographical distribution of catches.
The factors used as explanatory variables were: year (29), quarter (4), area (2) (<15^{o}S and >15^{o}S) and the target species (6), as inferred from a cluster analysis, using the Kmeans method (FASTCLUS, Johnson e Wichern, 1988; SAS Institute Inc, 1989), to identify the number of ideal clusters.
Zero catches amounted to 88.0% for the mako and 58.0% for the blue shark. In order to address this high proportion of zeros, the CPUE of both species was standardized by a GLM, by 2 different approaches: in the first one, a negative binomial error structure (log link) was assumed, while in the second one, a more traditional deltalognormal model, assuming a binomial error distribution for the proportion of positive sets and a Gaussian error distribution for the positive blue and mako shark catches, was applied.
The negative binomial error structure is a discrete probability distribution which indicates the number of trials that are necessary to obtain k successes of equal probability θ at the ending of n fishing sets, being given by:
As the negative binomial (NB) distribution requires integer values, the CPUE was transformed to a discrete variable. Since the effort variance was less than 10%, the CPUE was obtained based on the number of fish caught by the mean effort (1,929 hooks per fishing set), rounded to the nearest integer. The general NB formulation used in the present study was expressed by the following form: CPUE= Year+Quarter+Area+Target+Year*Quarter. The general delta log formulation, in turn, was expressed by the following form: Log(CPUE)= Year+Quarter+Area+Target+Year*Quarter. Deviance analysis tables for both models were provided.
3. Results
Similarly to a previous analysis (Hazin et al., 2006), six clusters (or target species) were identified: C1: swordfish/ blue shark; C2: albacore; C3: yellowfin tuna; C4: bigeye tuna; C5: other fish; and C6: Carcharhinus spp (Table 1). The swordfish/ blue shark cluster was the most frequent one, accounting for more than one third (36.2%) of the 67,335 longline sets. For both species, the mean frequency in the cluster C1 was about 5 times higher than in the other five, indicating they are strongly associated in the longline catches. This fact greatly facilitates the switch of the target species between them, a behavior that is becoming increasingly common among several fleets, particularly with the growing value of blue shark fins and meat in the international market. The C1 cluster was also the one with the largest mako frequency.
The negative binomial model explained 32.0% and 23.3% of the variance, for blue shark and mako shark, respectively. Similarly to previous works, the target species (cluster) was the most important factor, explaining 57.2 % and 29.3% of the deviance in blue and mako shark CPUE, respectively. Year (20.5% to blue and 18.3% to mako sharks) and area (13.1% to blue and 18.0% to mako sharks) were the next most important variables, while quarter (1.7% to blue shark and 6.8% to mako sharks) played a minor role (Tables 2 and 3).
For the blue shark deltalognormal distribution, the model explained 24% of the variance for the positive catches and about 68% of the proportion of positives. The distribution of residuals showed a reasonably good fit (Figure 2). The main factor explaining the variance in both cases was the year, accounting for 38% and 60%, respectively (Table 4). Quarter and cluster were also important for the positive catches (26.6% and 21.2%, respectively), whilst for the proportion of positives, besides the year, only cluster had a significant influence (23.4%).
For the mako shark deltalognormal distribution, the model explained about 22% of the variance for the positive catches and about 50% of the proportion of positives, performing a little poorer than for the blue shark, certainly due to the much higher variability and proportion of zero catches observed for mako. The distribution of residuals is shown in Figure 3. The main factor explaining the variance in mako positive catches was year (56.6%), followed by the year:quarter interaction (22.2%). The cluster was the third most important factor, accounting for about 13% of the variance (Table 5). The year was also the most important factor for the proportion of positives (30.5%), followed by cluster and area (both around 25%).
The mean SE for the negative binomial models (0.385, for blue shark, and 0.043 for makos) were slightly smaller than the ones for the deltalognormal (0.425, for blue shark, and 0.052 for makos) (Tables 6 to 9), indicating they might be a better option for the standardization of CPUE for species which have a high proportion of zeros, such as the mako and the blue sharks, particularly because its higher simplicity. That is particularly true if the fishing effort per set does not show a too high variability, such as in the present case. The trends of standardized CPUE for both methods were pretty close, though, for both species, although the negative binomial showed a smaller variance.
In spite of year to year oscillation, with a few spikes and drops, blue shark CPUE by Brazilian longliners from 1978 to 2006 seemed to be pretty stable, particularly in the past 6 years. The sharp increase in the nominal CPUE observed from 2002 on, which probably resulted from a shift of the target species towards swordfish, was not reflected in the standardized CPUE, indicating that the inclusion of the cluster as a factor in the GLM might have satisfactorily compensated for that change in the targeting strategy.
Like for the blue shark, the mako shark CPUE standardized by both methods also showed a relative stability along the entire period, with a slight upward trend, in spite of a much stronger variance, certainly linked to its rarer occurrence in catches.
4. Acknowledgements
This work was made possible by the Secretariat of Fisheries and Aquaculture (SEAP) of the Brazilian Government and by Fundação de Amparo à Ciência e Tecnologia do Estado de Pernambuco FACEPE. We are also grateful to the Tropical Conservation and Development Program (TCD) and the Fisheries and Aquatic Science Department, of University of Florida USA, for the Fellowship/Assistantship provided.
5. Literature cited
AMORIM, A.F., C.A. Arfelli. 1984. Estudo biológicopesqueiro do espadarte, Xiphias gladius Linnaeus, 1958, no sudeste e sul do Brasil (1971 a 1981). Bol. Inst. Pesca 11, 3562.
ARFELLI, C.A. 1996. Estudo da pesca e aspectos da dinâmica populacional de espadarte Xiphias gladius L.1758 no Atlântico sul. Universidade Estadual Paulista, Rio Claro, São Paulo, 175 p.
COMPAGNO, L.J.V. 1984. Sharks of the world. An annotated and illustrated catalogue of sharks species known to date. FAO Fisheries Synopsis Nº 125, 4 (1 and 2): 655 pp.
CLARKE, S.C., J.E. Magnussen, D.L. Abercrombie, M.K. McAllister, M.S. Shivji. 2006. Identification of Shark Species Composition and Proportion in the Hong Kong Shark Fin Market Based on Molecular Genetics and Trade Records. Conservation Biology 20, 201211.
FRÉON, P., O.A. Misund. 1999. Dynamics of pelagic fish distribution and behaviour: effects on fisheries and stock assessment. In: Science, B.s (Ed.), Fishing News Books. Oxford, London, p. 348p.
GULLAND, J.A., 1983. Fish stock assessment: a manual of basic methods, New York.
HAZIN, Fábio H. V.; A.A. Couto, K. Kihara, K. Otsuka and M, Ishino. 1990. Distribution and abundance of pelagic sharks in the southwestern equatorial Atlantic. J. Tokyo Univ. Fish., 77(1):5164.
HAZIN, Fábio H. V., 1993. Fisheries oceanographical study of tunas, billfishes and sharks in the southwestern equatorial Atlantic Ocean. University of Fisheries, Tokyo.
HAZIN, F.H.V., Hazin, H.G., 1999. Análise da viabilidade do emprego do espinhel monofilamento em pequenas embarcações da frota artesanal nordestina. Anais XV Premio Jovem cientista/CNPq. CNPq,
Brasilia, p. 212p.
HAZIN, H. G., Fabio Hazin, Paulo Travassos, Felipe C. Carvalho, and Karim Erzini, 2007. Standardization of swordfish CPUE series caught by Brazilian longliners in the Atlantic Ocean, by GLM, using the targeting strategy inferred by cluster analysis. Collect. Vol. Sci. Pap. ICCAT,60(6): 20392047.
JONHSON, R. and Wichern, D.W. 1988. Applied Multivariate Statistical Analysis. 2nd edn. Prentice Hall, New York, pp. 607
MENEZES de Lima, J.H., Kotas, J.E., Lin, C.F., 2000. A historical review of the brazilian longline
fishery and catch of swordfish. Collect. Vol. Sci. Pap. ICCAT, 51(4): 13291358.
MAUNDER, M. N. and A. E. Punt. 2004. Standardizing catch and effort data: a review of recent approaches. Fish. Res. 70: 141159.
Table 1. Distribution of 67,335 longline sets done by the Brazilian tuna longline fishery in the Atlantic Ocean, from 1978 to 2006, by cluster (values over 15% are in bold).
Cluster

1

2

3

4

5

6

Fishing Sets

24,385

12,746

13,276

10,037

6,088

805

Frequency of sets

36.2%

18.9%

19.7%

14.9%

9.0%

1.2%

other tunas

11.6%

0.6%

0.2%

0,1%

1,5%

0,0%

Yellowfin tuna

5,3%

6,3%

51,6%

11,1%

12,1%

3.9%

Albacore

4.4%

74.6%

8.7%

5.4%

9.5%

0.2%

Bigeye tuna

4.4%

4.4%

8.4%

58.0%

5.7%

0.7%

Bluefin tuna

0.0%

0.0%

0.1%

0.2%

0.2%

0.0%

Swordfish

36.4%

3.0%

8.5%

12.2%

7.1%

8.3%

Sailfish

1.9%

1.5%

3.2%

1.4%

4.3%

1.3%

White marlin

1.0%

0.7%

1.4%

0.9%

2.8%

0.6%

Blue marlin

1.5%

0.6%

1.7%

1.3%

0.8%

0.4%

Other billfishes

1.0%

0.3%

0.1%

0.0%

1.0%

0.0%

Wahoo

0.7%

0.7%

3.4%

0.6%

6.4%

0.0%

Dolphin fish

2.8%

0.4%

1.4%

0.7%

1.8%

0.3%

Blue shark

19.8%

1.6%

3.5%

2.8%

3.9%

4.4%

Hammerhead

1.2%

0.0%

0.2%

0.1%

0.8%

0.8%

Bigeye tresher

0.2%

0.0%

0.1%

0.0%

0.1%

0.2%

Shortfin mako

1.7%

0.4%

0.4%

0.2%

0.9%

0.5%

Grey sharks (Carcharhinus spp.)

0.5%

0.0%

0.3%

0.2%

0.1%

76.0%

Oceanic whitetip

0.1%

0.0%

0.0%

0.0%

0.0%

0.1%

Other shark

3.6%

2.2%

3.1%

2.9%

2.5%

1.6%

Other fish

2.1%

2.4%

3.6%

2.0%

38.6%

0.6%
 