Collaborative Study of Bigeye Tuna Cpue from Multiple Atlantic Ocean Longline Fleets in 2018

SCRS/2018/058 Collect. Vol. Sci. Pap. ICCAT, 75(7): 2033-2080 (2019)

COLLABORATIVE STUDY OF BIGEYE TUNA CPUE FROM MULTIPLE ATLANTIC OCEAN LONGLINE FLEETS IN 2018

S.D. Hoyle1, J. Hsiang-wen Huang2, D. Nam Kim3, M. Kyung Lee3, T. Matsumoto4, and J. Walter5

SUMMARY

In April 2018 a collaborative study was conducted between national scientists with expertise in Chinese, Japanese, Korean, Chinese-Taipei, and USA longline fleets, and an independent scientist. The meetings addressed Terms of Reference covering several important issues related to bigeye tuna CPUE indices in the Atlantic Ocean. The study was funded by the International Commission for the Conservation of Atlantic Tunas (ICCAT) and the International Seafood Sustainability Foundation (ISSF). The meeting developed joint CPUE indices based on analysis of combined data from the Japanese, Korean, Chinese-Taipei, and US fleets. The meeting also welcomed the availability of data from the Chinese longline fleet, and began the process of preparing and exploring this new dataset for future analysis.

RÉSUMÉ

En avril 2018, une étude collaborative a été menée entre des scientifiques nationaux possédant une expertise en matière de flottilles palangrières de la Chine, du Japon, de la Corée, du Taipei chinois et des États-Unis, et un scientifique indépendant. Les réunions ont abordé les termes de référence couvrant plusieurs questions importantes liées aux indices de CPUE du thon obèse dans l'océan Atlantique. L'étude a été financée par la Commission internationale pour la conservation des thonidés de l'Atlantique (ICCAT) et l’International Seafood Sustainability Foundation (ISSF). La réunion a élaboré des indices communs de CPUE basés sur l'analyse de données combinées provenant des flottilles japonaises, coréennes, américaines et du Taipei chinois. La réunion a également salué la disponibilité des données de la flottille palangrière chinoise et a entamé le processus de préparation et d'exploration de ce nouveau jeu de données en vue d'une analyse future.

RESUMEN

En abril de 2018 se llevó a cabo un estudio colaborativo entre científicos nacionales con experiencia en las las flotas de palangre de China, Japón, Corea, Taipei Chino y Estados Unidos, y un científico independiente. Las reuniones abordaron los términos de referencia cubriendo varios temas importantes relacionados con los índices de CPUE del patudo en el océano Atlántico. El estudio fue financiado por la Comisión Internacional para la Conservación del Atún Atlántico (ICCAT) y la International Seafood Sustainability Foundation (ISSF). La reunión desarrolló índices de CPUE conjuntos basándose en el análisis de los datos combinados de las flotas de Japón, Corea, Taipei Chino y Estados Unidos. La reunión acogió también con satisfacción la disponibilidad de datos de la flota de palangre chino e inició el proceso de preparar y explorar este nuevo conjunto de datos para futuros análisis.

KEYWORDS

Catch/effort, Bigeye tuna

1 ICCAT consultant, New Zealand. Email: [email protected] 2 Chinese Taipei Scientist, National Taiwan Ocean University, Keelung. Email: [email protected] 3 Korean Scientists, National Institute of Fisheries Science, Busan. [email protected], [email protected] 4 Japanese scientist, National Research Institute of Far Seas Fisheries, Shimizu. [email protected] 5 US Scientist, NOAA Fisheries, Miami, Florida. [email protected] 2033 Introduction

In April 2018 a collaborative study of longline data and CPUE standardization for bigeye tuna was conducted between scientists with expertise in Chinese, Japanese, Chinese-Taipei, Korean, and US fleets, and an independent scientist. The study was funded by the International Commission for the Conservation of Atlantic Tunas (ICCAT) and the International Seafood Sustainability Foundation (ISSF). The study addressed the Terms of Reference outlined below, which cover important issues that had previously been highlighted by different working parties. Work was carried out, for those factors relevant to them, for the following:

- Area: Atlantic Ocean - Fleets: Chinese longline, Japanese longline; Chinese-Taipei longline, Korean longline, US longline - Stocks: Bigeye tuna

Terms of Reference

- Provide indices of abundance for bigeye tuna to be presented at the bigeye tuna data preparation meeting (Madrid, Spain 23-27 April 2018) and the bigeye stock assessment meeting between 16-20 July 2018 (Pasaia, Spain).

- The analyses will consider data to be provided by key industrial fisheries operating in the Atlantic Ocean, including Japan, USA, Chinese-Taipei, the Republic of Korea, and potentially also People’s Republic of China.

- Analyses will be carried out prior to and during the data preparatory meeting and after if necessary to include comments/decisions taken by the SCRS Tropical Tuna Species Group. Preparations will be carried out for each dataset using R code submitted by the consultant to participating national scientists. Further and final analyses will be carried out during a joint CPUE meeting between all participating scientists and the consultant before (or during) the data preparatory meeting.

Tasks will include the following, to the extent possible in the available time:

- Provide R data preparation code to all participants who are experienced in R use, and liaise with them to help preparing their data sets in the correct format.

- Load, prepare, and check each dataset, given that data formats and pre-processing vary between fleets and through time.

- Explore data coverage and representativeness of the data.

- Conduct the following analyses to prepare indices:

• Apply cluster analyses or alternative methods for identifying targeting; • Develop CPUE standardizations for bigeye tuna using reliable data from each CPC. Prepare separate indices for each fleet, and joint indices, using the same statistical approach (annual and quarter indices); • Thoroughly check all code and results in order to validate the final standardized indices series for the use during the 2018 bigeye stock assessment session; • Explore spatial and temporal patterns in residuals by fleet and cluster, in order to better understand the factors driving CPUE changes, to explore potential confounding effects and possible seasonal catchability changes.

- An SCRS document should be prepared, including detailed descriptions of the: data set compiled, the methods used and analyses made, and the standardized CPUE indices for each fleet and the joint indices (by areas and annual/quarters). The document to be presented during the Bigeye Data Preparatory meeting, shall be submitted to the ICCAT Secretariat ([email protected]) by 16/04/2018 at the latest and should be prepared in accordance with the ICCAT Guidelines for authors of scientific papers for the SCRS and ICCAT Collective Volume Series.

- Undertake any additional analyses deemed relevant before the bigeye stock assessment meeting, according to decisions taken by the SCRS Tropical Tuna Species Group during the bigeye tuna data preparation meeting.

2034 All work is subject to the agreement of the respective fisheries agencies to make the data available.

Methods

Data cleaning and preparation

The five datasets had many similarities but also significant differences. Variables differed somewhat among datasets, as did other aspects such as the sample sizes, the data coverage and the natures of the fleets.

Data preparation and analyses were carried out collaboratively among all participants, using a standard set of scripts developed for this purpose in R version 3.3.4 (R Core Team 2016). The approaches followed those described by Hoyle et al. (2015) and Hoyle et al. (2016).

Data

In this section we describe the datasets provided, and the methods used to prepare and clean the data for analysis. As the provided datasets were prepared for this collaborative study, the data do not include all information potentially included in logbook data. The cleaning described here differs from the standard cleaning procedures by national scientists when producing CPUE indices. All operational data were available only for the purpose of this collaborative project, and no operational data is available after the project.

More thorough descriptions of the datasets and their preparation are provided in papers prepared by the national scientists.

Japanese data were available from 1958-2017, with fields year, month and day of operation, location to 1° of latitude and longitude, vessel call sign, logbook id, no. of hooks between floats, number of hooks per set, date of the start of the fishing cruise, and catch in number of bluefin tuna, albacore, southern bluefin, bigeye, yellowfin, swordfish, white marlin, Atlantic blue marlin, and black marlin. Data from 1958 comprised only 31 sets and were not used. Data for 2017 were preliminary.

The Chinese-Taipei operational data were available for the indices for the period 2005-2017, although earlier data were provided to help understand the fisheries. Available fields were year, month and day of operation; vessel call sign; operational area (a code indicating fishing location at 5° resolution); operation location at 1° resolution (from 1995); number of hooks between floats (from 1995); number of hooks per set; catches in number for the species albacore, bigeye, yellowfin, bluefin, southern bluefin, other tuna, swordfish, white marlin, blue marlin, black marlin, other billfish, skipjack, shark, and other species; equivalent values in weight for all species; SST; bait type fields for ‘Pacific saury’, ‘mackerel’, ‘squid’, ‘milkfish’, and ‘other’; depth of hooks (m); set type (type of target, from 2006); remarks (indicating outliers).

Korean operational data were available for 1979 to 2016, with fields vessel id, operation date, operation location to 1°, number of hooks, number of floats, and catch by species in number for albacore, bigeye, black marlin, blue marlin, white marlin, other species, bluefin tuna, sailfish, shark, skipjack, swordfish, and yellowfin.

United States operational data were available for 1986 to 2017, with fields yrqtr, 5° cell, anonymised vessel id, vessel-month, hbf, SST, hooks, bluefin, albacore, bigeye, yellowfin, swordfish, white marlin, blue marlin, black marlin, sailfish, shark, blue shark, shortfin mako shark, porbeagle shark.

In the Japanese data international call sign was available 1979 - present, and was selected as the vessel identifier. Call sign is unique to the vessel and held throughout the vessel’s working life. Prior to 1979 the logbook id was available, which could be used to identify the effort by a vessel within one year, but not between years. In the Chinese-Taipei data, the registered number of the vessel was available for each set, and was also selected as the vessel identifier. The first digit of the Chinese-Taipei vessel number indicated the tonnage of the vessel. In the Korean data the call signs were understood to have changed through time to some extent, and so vessel ids were assigned based on a combination of vessel names and vessel call signs. For all fleets, the vessel id was rendered anonymous by changing it to an arbitrary integer. Sets without a vessel call sign were allocated a vessel id of ‘1’. For joint analyses, a fleet code was added to differentiate vessels from different fleets.

2035 In all Japanese and Korean data, and in most Chinese-Taipei data from 1994, latitude and longitude were reported at 1° resolution, with a code to indicate north or south, west or east. Chinese-Taipei fishing locations were otherwise reported at 5° cell resolution using a logbook code. US data were reported at 5° cell resolution. All data were adjusted to represent the south-western corner of the 1 x 1° square, and longitudes translated into 360° format. Each set was allocated to regions according to region definitions. Data outside these areas were ignored. Location information was used to calculate the 5° square (latitude and longitude).

Hooks per set were reported in all datasets, and the few sets without hooks were deleted. For the purposes of further analyses, we cleaned the data by removing data likely to be in error. The criteria were selected after discussion with experts in the respective datasets. Hooks per set above 5000 and less than 200 were removed.

In the Chinese-Taipei logbook, columns for bluefin were added in 1994. Prior to this bluefin were only recorded in the database when individuals changed the heading in the logbook. The number of reported bluefin increased substantially in 1994. With the three fields for ‘other’ species, ‘other tunas’ are thought to be mostly neritic tunas, ‘other billfish’ may represent mostly sailfish and longbill spearfish, and ‘other fish’ included pomfrets, dolphinfish, ocean sunfish, and, particularly in recent years, oilfish.

In the logbooks of each fleet some very large catches were reported at times for individual species, but were not removed since there was anecdotal evidence that they may be genuine, and because they are unlikely to affect results substantially. Further investigation should consider the pros and cons of retaining these values.

In the Japanese logbook hooks between floats (HBF) were available for almost all sets 1971-2017, and for a high proportion of sets 1959-1966. Sets after 1975 with HBF missing or > 25 were removed. Sets before 1975 with missing HBF were allocated HBF of 5, according to standard practice with Japanese longline data (e.g. Langley et al. 2005; Hoyle et al. 2013; Ochi et al. 2014). In the Chinese-Taipei logbook hooks between floats (HBF) were available from 1995. In the Korean logbook HBF was not available but the number of floats was reported, so we calculated HBF by dividing the number of hooks by the number of floats and rounding it to a whole number.

The remarks section of the Chinese-Taipei dataset indicated outliers and other anomalies. Codes and criteria for outliers changed in 2012. Before 2012 an outlier was flagged if there was catch of more than 5 tons of a species per set, or outliers in the distribution of species catch number per set. From 2012 an outlier was flagged according to the ‘IQR rule’. 1. Arrange average catch numbers per set (within a year) for all vessels in order. 2. Calculate first quartile (Q1), third quartile (Q3) and the interquartile range (IQR=Q3-Q1). 3. Compute Q1-1.5 x IQR and Compute Q3+1.5 x IQR. Anything outside this range is an outlier. This outlier information is used in the standard data cleaning procedures for Chinese-Taipei standardisations. We did not use the outlier information in data cleaning for this paper.

This was the first time that Chinese operational data have been made available at an ICCAT meeting. Chinese scientists provided data covering the period January 2008 to August 2017. Fields provided were vessel, set date, year, month, latitude and longitude (to 0.01 degree), hooks, hooks between floats, and albacore, bigeye, and yellowfin tuna catch in number and weight (kg). There was insufficient time to prepare, explore, and understand the data to a level where it could be included in the joint analyses.

After data cleaning, a standard dataset was produced for each fleet to be used in subsequent analyses.

Each set was allocated to regions, based on the region definitions used in the stock assessment (Figure 1). Data outside these regions were ignored. Subsequent analyses were performed separately for each region in each regional structure.

Cluster analysis

The proportions of each species in the catch varied considerably between fleets, spatially, and through time. We used cluster analysis to explore factors contributing to these patterns, and to identify effort groups with similar species composition and therefore, presumably, similar fishing behaviour and targeting.

We clustered the data using the approach applied by Hoyle et al. (2015). We removed all sets with no catch of any of the species, and then aggregated by vessel-month. Set level data contains variability in species composition due to the randomness of chance encounters between fishing gear and schools of fish. This variability leads to some misallocation of sets using different fishing strategies. Aggregating the data tends to reduce the variability, and therefore reduce misallocation of sets. For these analyses we aggregated the data by vessel-month, assuming that 2036 individual vessels tend to follow a consistent fishing strategy through time. One trade-off with aggregation in this way is that vessels may change their fishing strategy within a month, which will result in misallocation of sets. For the purposes of this paper we refer to aggregation by vessel-month as trip-level aggregation, although the time scale is (for distant water vessels) in most cases shorter than a fishing trip. For Japanese data prior to 1979 vessel id was not available, but we were able to cluster effort by vessel-month using the logbook id.

We calculated proportional species composition by dividing the catch in numbers of each species by catch in numbers of all species in the vessel-month. Thus the species composition values of each vessel-month summed to 1, ensuring that large catches and small catches were given equivalent weight. The data were transformed by centring and scaling, so as to reduce the dominance of species with higher average catches. Centring was performed by subtracting the column (species) mean from each column, and scaling was performed by dividing the centred columns by their standard deviations.

We clustered the data using the hierarchical Ward hclust method, implemented with function hclust in R, option ‘Ward.D’, after generating a Euclidean dissimilarity structure with function ‘dist’. This approach differs from the standard Ward D method which can be implemented by either taking the square of the dissimilarity matrix or using method ‘ward.D2’ (Murtagh & Legendre 2014). However in practice the method gives similar patterns of clusters to other methods, more reliably than ward.D2 (Hoyle et al. 2015).

Data were also clustered using the kmeans method, which minimises the sum of squares from points to the cluster centres, using the algorithm of Hartigan and Wong (1979). It was implemented using function kmeans in the R stats package (R Core Team 2014).

Selecting the number of groups

We used several subjective approaches to select the appropriate number of clusters. In most cases the approaches suggested the same or similar numbers of groups. First, we applied hclust to transformed trip-level data and examined the hierarchical trees, subjectively estimating the number of distinct branches. Second, we ran kmeans analyses on untransformed trip-level data with number of groups k ranging from 2 to 25, and plotted the deviance against k. The optimal group number was the lowest value of k after which the rate of decline of deviance became slower and smoother. Third, following Winker et al (2014) we applied the nScree() function from the R nFactors package (Raiche & Magis 2010), which uses various approaches (Scree test, Kaiser rule, parallel analysis, optimal coordinates, acceleration factor) to estimate the number of components to retain in an exploratory PCA. Where there was uncertainty about the number of clusters, we selected the option with more clusters.

Plotting and data selection

We plotted the hclust clusters to explore the relationships between them and the species composition and other variables, such as HBF, number of hooks, year, and set location. Plots included boxplots of a) proportion of each species in the catch, by cluster; b) the distributions of variables by cluster; and c) maps of the spatial distribution of clusters, one map for each cluster.

In some analyses clusters that caught very few of the species of interest were omitted, because they provide little relevant information and may cause analysis problems due to large numbers of zeroes, and memory problems due to large sample sizes. Cluster selection was based on review and discussion of the plots of covariates and species compositions by cluster.

We pooled data from multiple fleets into a single dataset which was then used to carry out analyses for the years 1959-2017.

For standardization of each region, data were selected for vessels that had fished for at least N1 quarters in that region. The standard level of N1 was 5 quarters. Subsequently, vessels were included in the full 1959 – 2017 dataset if they reported at least 100 sets, whereas for the shorter 1959 – 1978 and 1979 – 2017 datasets the criterion was reporting at least 50 sets. Spatial cells and year-quarters were each included if their data comprised at least 50 sets. Year-quarter * 5° cell strata with fewer than 5 sets were also removed, to avoid giving too much statistical weight to individual sets via the area-weighting process – described below. The distributions of sets per stratum by region without this final cleaning step are shown in Figure 4.

2037 For datasets with more than 60,000 sets the number of sets in each stratum (year-quarter * 5° cell) was limited by randomly selecting 40 sets without replacement from strata with more than 40 sets. Testing suggested that this approach did not cause bias, and random variation was low at 15 sets and very low at 30 sets per stratum (Hoyle & Okamoto 2011a), suggesting that 40 sets were more than adequate.

CPUE standardization

CPUE standardization methods generally followed the approaches used by Hoyle and Okamoto (2011b) with some modifications. The operational data were standardized using generalized linear models in R. Indices were prepared for each species and region using several approaches, summarised in Table 1 and further described below.

Distributions

Lognormal constant analyses were carried out using generalized linear models that assumed a lognormal distribution. In this approach the response variable log⁡(퐶푃푈퐸 + 푘) was used, and a Normal distribution assumed. The constant k, added to allow for modelling sets with zero catches of the species of interest, was 10% of the mean CPUE across all sets in the analysis. CPUE was defined at the set level as catch in number divided by hooks set.

Covariates included year-quarter (yq), and 5° cell (latlong5) fitted as categorical variables. Analyses including the continuous variable hooks fitted it using a cubic spline function h with 10 degrees of freedom. Analyses including the vessel identifier (vessid) fitted it as a categorical variable. Analyses including hooks between floats (hbf) fitted it as a continuous variable using a cubic spline , while those including cluster (cl) fitted it as a categorical variable.

푙표푔(퐶푃푈퐸 + 푘)~푦푞 + 푣푒푠푠𝑖푑 + 푙푎푡푙표푛푔5 + 휙(ℎ푏푓) + 푐푙 + ℎ(ℎ표표푘푠) + 휖

Delta lognormal analyses (Lo et al. 1992; Maunder & Punt 2004) used a binomial distribution for the probability w of catch rate being zero and a probability distribution f(y) , where y was log(catch in number/hooks set), for non- zero (positive) catch rates. The index estimated for each year-quarter was the product of the year effects for the two model components, (1 − 푤). 퐸(푦|푦 ≠ 0). Covariates in the binomial and lognormal models were the same as in the lognormal constant model described above.

푤, 푦 = 0 Pr(푌 = 푦) = { (1 − 푤)푓(푦) 표푡ℎ푒푟푤𝑖푠푒 g(푤) = (퐶푃푈퐸 = 0) ~⁡푐표푣푎푟𝑖푎푡푒푠 + 휖, where g is the logistic function. f(푦) = 퐶푃푈퐸 ~⁡푐표푣푎푟𝑖푎푡푒푠 + 휖

Data in all models except the binomial model were ‘area-weighted’, with the weights of the sets adjusted so that the total weight per year-quarter in each 5° square would sum to 1. This method was based on the approach identified using simulation by Punsly (1987) and Campbell (2004), that for set j in area i and year-quarter t, the 푙표𝑔(ℎ푖푗푡+1) weighting function that gave the least average bias was: 푤푖푗푡 = 푛 . Given the relatively low variation ∑푗=1 log(ℎ푖푗푡+1) ℎ푖푗푡 in number of hooks between sets in a stratum, we simplified this to 푤푖푗푡 = 푛 . ∑푗=1 ℎ푖푗푡

In the binomial component of the model, fitted probabilities of 0 or 1 may occur, associated with categories of vessel, year-quarter, or 5° cell. This is known as ‘perfect separation’ and is problematic for uncertainty estimation (Venables & Ripley 2002). This is one reason why the binomial component of the uncertainty is not used. Perfect separation should not bias prediction of the year-quarter effects, as long as a category at which separation occurs is not used in the predictor. Perfect separation in a year-quarter results in an accurate prediction of 0 or 1.

For the lognormal constant and positive lognormal GLMs, model fits were examined by plotting the residual densities and using Q-Q plots.

Data periods

Vessel identity information for Japan was only available from 1979, and all of the data before 1979 was Japanese. Overlap between vessels with the same id across years is required to avoid confounding between year effects and vessel ids. Thus we could not apply a consistent approach across all years when including vessel ids in the model. The discontinuity in vessel 1979 could be addressed in several different ways. We therefore analysed the data in several ways so as to provide the assessment scientists with appropriate data. 2038 First, we standardized the full dataset from 1959 to the present without including vessel effects.

Then we standardized the full dataset with vessel effects, assigning an identical dummy vessel ID to all sets that lacked vessel identity information. However, using a dummy value introduces several problems. First, most Japanese vessels begin to report their call sign in 1979, but a few do not, and these may be regarded as self-selected and not randomly selected from the vessel population. We therefore omitted all sets without vessels ids starting in 1979.

We estimated two time series: 1959-1978 without vessel effects, and a second time series 1979-2017 with vessel effects (omitting all sets without vessel IDs). Subsequently the analyst may use the two time series as desired, either as separate indices in the assessment, or concatenating them after adjusting the averages so that the estimates for 1979 are the same.

The effects of covariates were examined using influence plots, using the R package influ (Bentley et al. 2011).

Indices of abundance

Indices of abundance were obtained by applying the R function predict.glm to model objects. The datasets used for prediction included all year-quarter values, with all other variables fixed at either the median for continuous variables, or the mode for categorical variables. Binomial time effects were obtained by a) generating logit time effects from the glm, and b) adding a constant to these logit time effects so that the mean of the back-transformed proportions was equal to the proportion of positive sets across the whole dataset. The main aim with this approach is to obtain a CPUE that varies appropriately, since variability for a binomial is greater when the mean is at 0.5 than at 0.02 or 0.98, and the multiplicative effect of the variability is greater when the mean is lower. The outcomes were normalised and reported as relative CPUE with mean of 1.

An approach to combine the early 1959-1978 and late 1979 – 2017 time series was developed. The approach adjusted the two time series to have the same relationship in the 1977 - 1980 period as observed in the long term time series estimated without vessel effects. We took the averages of the two periods 1977 – 1978 and 1979 – 1980 in both the separate time series (m1early and m1late) and the continuous time series (m2early and m2late). The adjusted time series was created by concatenating the two series after multiplying each by m2 / m1. We applied this approach to both the quarterly and the annual time series.

Uncertainty estimates were provided by applying the R function predict.glm with type = ”terms” and se.fit=TRUE, and taking the standard error of the year-quarter effect. For the delta lognormal models we used only the uncertainty in the positive component. Uncertainty estimates from standardizing commercial logbook data are in general biased low and often ignored by assessment scientists, since they assume independence and ignore autocorrelation associated with (for example) consecutive sets by the same vessels in the same areas. There may be a very large mismatch between the observation error in CPUE indices and the process error in the indices that is estimated in the assessment. This is particularly true for distant water longline CPUE, where the very large sample sizes often generate small observation errors.

Residual distributions and Q-Q plots were produced for all but the binomial analyses. For the lognormal positive analyses that included cluster in the model, median residuals were plotted by cluster. For all lognormal positive analyses, residuals by year-quarter were plotted by flag; median residuals by year-quarter were plotted by flag; and median residuals by 5° cell were mapped onto a contour plot for each flag.

Results

Coverage estimates for each fleet (Figure 2) showed high levels of coverage for the Japanese fleet since 1960, and almost complete coverage since 1976. Coverage for the Chinese-Taipei fleet dropped from 1980 until 2004 and was particularly low in the late 1980s and early 1990s, but reached almost complete coverage in 2005. The Korean fleet has had relatively low bigeye catches in the Atlantic and coverage has been highly variable as a consequence. Coverage of the US fleet has been high since the mid-1990s, with some variability due to uncertainty about average fish weights.

Comparison plots are provided that show the preliminary indices developed for each fleet and the combined dataset, plotted together by region (Figure 3). Each series is normalised to average 1.

2039 We present figures for the indices likely to be used in assessments, so as to conserve space. In all regions we select figures from the analysis that omits low-target clusters from the dataset, and includes both HBF and cluster in the model.

We estimated delta lognormal indices for all regions of the bigeye stock assessment (Figures 5 - 7, Tables 2 and 3). For each region we present 4 sets of indices: for 1959 – 2017 without vessel effects, for the same period with vessel effects, for 1959 – 1978 without vessel effects, and for 1979 – 2017 with vessel effects. Index values and their coefficients of variation are provided in Tables 2 and 3.

The joined plots for region 2 (Figures 8 and 9) illustrate that the trend in the early time series differs from the trend over the same period in the long-term time series. The patterns are very similar after 1971, but they differ before this period. The influence plots show that the transition in about 1971 is associated with more influence from spatial effects in the early period analysis (Figure 12-14). This results in the combined index being higher on average during the early period.

We provide influence plots for each region and for the analyses 1959 – 1978 without vessel effects, 1979 – 2017 with vessel effects, and for the full period both with and without vessel effects for region 2 only (Figures 10 – 17). In region 1 prior to 1979 (Figure 10) the spatial effects had the largest influence on the temporal trend, and cluster had a smaller impact, whereas HBF, which varied little, had almost no impact. In region 1 after 1979 the cluster had the greatest impact on the trend, while spatial effects, HBF, and vessel effects were also moderately influential (Figure 11).

In region 2 prior to 1979, both spatial effects and cluster were very influential, suggesting a big increase in fishing power through time, particularly at the end of the time series (Figure 12). After 1979 there was some impact from HBF, vessel effects and cluster, but the scale of the impact was much smaller than in the earlier period (Figure 13). Fishing power for bigeye increased from 1979 until the mid-1990s with increasing use of larger HBF. Fishing power associated with individual vessels appeared to increase over the whole time series. However there was a general decline associated with the cluster effect, suggesting a reduction in the proportion of effort targeted at bigeye.

In the southern region 3 prior to 1979 (Figure 16) there was an approximately four-fold increase in fishing power associated with fishing location, with an increasing proportion of effort in northern parts of the region, where bigeye catch rates are higher. The cluster effect also showed increasing fishing power. Towards the end of the time series. After 1979 (Figure 17) the influence of the fishing location changed with a dip in fishing power in the mid- 1980s followed by recovery. Hooks between floats had surprisingly little impact on fishing power. The effects of individual vessels and cluster were not substantial, though vessel had a short-term positive influence on fishing power that matched the dip associated with fishing location. With the small sample sizes there may be some confounding effects.

Plots showing the spatial effects estimated in each cell in the 1979-2017 models indicate that in the northern region 1, catch rates are higher further south and in the middle of the ocean between about 40 and 60° W (Figure 18). In the equatorial region 2 catch rates are highest at about 10° S and 40° W, and lowest in the Caribbean. There is also an area of low longline catch rates in the Gulf of Guinea (Figure 19). In southern region 3 data are too sparse to estimate reliable spatial effects in all location, but in general catch rates tend to be higher further north and east (Figure 20).

Diagnostics for the lognormal positive distribution indicate a small amount of negative skewness in the distributions of residuals (Figures 21-23), but otherwise the distributions conform well to assumptions of normality.

Trends through time in temporal residuals from the 1979-2017 models (Figures 24-26) do not show clear patterns in the northern region 1 or the southern region 3. However in equatorial region 2 there is more catch rate decline than elsewhere between 15N and 10S, corresponding to areas with more fishing effort, including purse seine effort. The largest declines that are relatively spatially continuous are close to the African coast between the equator and 10N.

Medians of residuals by cluster and year-quarter for the three regions (Figures 27-29) do not show strong patterns that might indicate problems in the analysis. Median residuals by cluster and 5 cell show patterns that suggest different spatial effects may be associated with different fishing behaviour (Figures 30-32). 2040 Discussion

The development of joint indices was motivated by concern about differing trends in the time series developed for individual fleets. Joint standardization allowed the comparison of data from all fleets using identical methods, which permitted us to distinguish the influences of methods and data. The comparison plots show sufficiently small differences between the individual fleets’ time series to suggest that they could be used together in a single model.

The joint indices were in general less variable than the individual series, with fewer gaps, and with information extending to the most recent period. For example, in region 1 the Japanese time series ends in 2015 due to diminishing effort, but the joint time series continues to 2017 using information in the US dataset.

The joint indices included Japanese, Korean, Chinese-Taipei (from 2005), and US data. Future analyses should seek to include additional datasets such as the operational data from the Chinese fleet. The short timelines prior to and during the meeting made it too difficult to prepare this new dataset to the required level, but it was very encouraging that the Chinese data were made available for the first time, and we look forward to exploring and potentially including the dataset in future.

In all regions the joint series is most similar to the Japanese time series developed using the same methods, because there is more Japanese effort and bigeye catch than available from the other fleets, and the Japanese fishery also fishes more of the area. Within each stratum, each set receives the same statistical weight, so there is no explicit consideration of fleets as such.

Catch rates vary spatially. We have assumed that the spatial pattern is consistent through time, at least within individual analyses. This assumption does not hold between the pre- and post-1979 analyses, since the analyses are done separately and the spatial effects are therefore estimated independently. It is apparent from the influence plots and the indices that these differences in spatial patterns, combined with the change in fleet distribution in the early 1970s, have affected the indices. This may be due to changing distribution of the stock through time, or due to differing spatial catchability through time. Both are plausible. Stock movement has been demonstrated by the analysis of residual trends which shows spatial variation in trends between 1979 and 2017. Catchability change is also plausible given that the early period includes only Japanese effort while the latter period includes data from all fleets; Japanese fishing methods and targeting have also changed since the 1970s.

Given the evidence for changes in spatial patterns through time, it would be useful in future to explore time-area variation in catch rates.

The long-term time series with vessel effects failed at the transition between the 1959-1978 period without vessel effects and the 1979-2017 period with vessel effects, as is evident from the influence plot (Figure 15). This is due to lack of vessels overlapping between the periods. One approach to provide a continuous series would be to assume that the (uniform) vessel catchability in the early period was the same as the mean realised vessel catchability in 1979. This is equivalent to the assumption in the long term series without vessel effects, but with the advantage that vessel-level catchability variation is modelled from 1979. We encourage the documentation of vessel identity information for the Japanese fleet prior to 1979, which would allow us to resolve this problem.

In the ‘joined’ indices, the early and late series are combined by imposing upon them the 1977-1980 trend observed in the long term series without vessel effects. This approach also maintains approximately consistent average catchability in 1977-1980, allowing for some differences between the series in spatial effects and other covariates. In this analysis we removed 5° cell by year-quarter strata that included fewer than 5 sets. This was done to avoid giving too much influence to individual sets, since each stratum was weighted equally in the model. Choosing the five-set criterion required a balance between removing the most uncertain strata and retaining sufficient strata to estimate indices. This balance is likely to vary between regions and through time.

Acknowledgments

Thanks to the International Commission for Conservation of Atlantic Tunas (ICCAT) and the International Seafood Sustainability Foundation (ISSF) for funding the work. We are grateful to ICCAT and the ICCAT Secretariat for facilitating the meeting. Thanks to Hilario Murua, Victor Restrepo, and Ana Justel for their support. Thanks to the Korean National Institute of Fisheries Science, Chinese-Taipei Overseas Fisheries Development Council, Japanese Institute for Far Seas Fisheries, and US National Marine Fisheries Service for allowing us to access their data. 2041 References

Bentley N., Kendrick T.H., Starr P.J., Breen P.A. 2011. Influence plots and metrics: tools for better understanding fisheries catch-per-unit-effort standardizations. ICES Journal of Marine Science 69(1): 84-88.

Campbell R.A. 2004. CPUE standardisation and the construction of indices of stock abundance in a spatially varying fishery using general linear models. Fisheries Research 70(2-3): 209-227.

Hartigan J.A., Wong M.A. 1979. Algorithm AS 136: A k-means clustering algorithm. Journal of the Royal Statistical Society.Series C (Applied Statistics) 28(1): 100-108.

Hoyle S., Davies N., Chang S.-K. 2013. Analysis of swordfish catch per unit effort data for Japanese and Chinese Taipei longline fleets in the southwest Pacific Ocean, WCPFC-SC9-2013/SA-IP-03. WCPFC Scientific Committee, Ninth Regular Session, 7-15 August 2012, Busan, Republic of Korea.

Hoyle S., Kim D., Lee S., Matsumoto T., Satoh K., Yeh Y. 2016. Collaborative study of tropical tuna CPUE from multiple Indian Ocean longline fleets in 2016.

Hoyle S.D., Okamoto H. 2011a. Analyses of Japanese longline operational catch and effort for bigeye and yellowfin tuna in the WCPO, WCPFC-SC7-SA-IP-01. Western and Central Pacific Fisheries Commission, 7th Scientific Committee. Pohnpei, Federated States of Micronesia.

Hoyle S.D., Okamoto H. 2011b. Analyses of Japanese longline operational catch and effort for bigeye and yellowfin tuna in the WCPO, WCPFC-SC7-SA-IP-01. Western and Central Pacific Fisheries Commission, 9th Scientific Committee. Pohnpei, Federated States of Micronesia.

Hoyle S.D., Okamoto H., Yeh Y.-m., Kim Z.G., Lee S.I., Sharma R. 2015. IOTC–CPUEWS02 2015: Report of the 2nd CPUE Workshop on Longline Fisheries, 30 April – 2 May 2015. 126 p.

Langley A., Bigelow K., Maunder M., Miyabe N. 2005. Longline CPUE indices for bigeye and yellowfin in the Pacific Ocean using GLM and statistical habitat standardisation methods. WCPFC-SC1, Noumea, New Caledonia SA WP-8. 8-19 p.

Lo N.C.H., Jacobson L.D., Squire J.L. 1992. Indices of relative abundance from fish spotter data based on delta- lognormal models. Canadian Journal of Fisheries and Aquatic Sciences 49(12): 2515-2526.

Maunder M.N., Punt A.E. 2004. Standardizing catch and effort data: a review of recent approaches. Fisheries Research 70(2-3): 141-159.

Murtagh F., Legendre P. 2014. Ward’s hierarchical agglomerative clustering method: which algorithms implement Ward’s criterion? Journal of Classification 31(3): 274-295.

Ochi D., Matsumoto T., Satoh K., Okamoto H. 2014. Japanese longline CPUE for bigeye tuna in the Indian Ocean standardized by GLM. IOTC–2014–WPTT16–29 Rev_1. 28 p.

Punsly R. 1987. Estimation of the relative annual abundance of yellowfin tuna, Thunnus albacares , in the eastern Pacific Ocean during 1970-1985. LA JOLLA, CA ( ), I-ATTC.

R Core Team 2014. R: A Language and environment for statistical computing. Vienna, Austria, R Foundation for Statistical Computing. R Core Team 2016. R: A Language and Environment for Statistical Computing. Vienna, Austria, R Foundation for Statistical Computing.

Raiche G., Magis D. 2010. nFactors: Parallel analysis and non graphical solutions to the Cattell Scree Test. R package version 2(3).

Venables W.N., Ripley B.D. 2002. Modern Applied Statistics with S, Springer.

Winker H., Kerwath S.E., Attwood C.G. 2014. Proof of concept for a novel procedure to standardize multispecies catch and effort data. Fisheries Research 155: 149-159. 2042 Table 1. Species, regions, distributions and variables used in CPUE analyses. Target Species Regions Fleets Vessel ID Period Distribution variable Lognormal constant, Cluster delta lognormal BET 1, 2, 3 All Y, N 1959-2017 & HBF (binomial and lognormal). N 1959-1978 Y 1979-2017

Table 2. Indices for early period (1959 – 1978), standardized with vessel effects, for regions 1 to 3. Year Quarter R1 R2 R3 Estimate CV Estimate CV Estimate CV 1959 1 0.5018 0.0575 1959 2 0.8661 0.0387 1959 3 0.8524 0.0418 1959 4 1.2521 0.033 1.1619 0.0955 1960 1 0.7447 0.0369 NA NA 1960 2 0.919 0.0304 NA NA 1960 3 0.934 0.0344 NA NA 1960 4 1.2046 0.0269 NA NA 1961 1 0.7852 0.0369 NA NA 1961 2 1.2122 0.0262 6.5909 0.0991 1961 3 1.6092 0.0277 5.636 0.0616 1961 4 1.1809 0.0262 2.254 0.0477 1962 1 0.7499 0.0332 1.7259 0.0672 1962 2 1.055 0.0234 NA NA 1962 3 1.1492 0.1507 1.104 0.0282 NA NA 1962 4 NA NA 1.2073 0.028 1.7566 0.0424 1963 1 NA NA 0.8091 0.0269 0.5418 0.0428 1963 2 0.8639 0.1073 1.3394 0.0236 NA NA 1963 3 0.6697 0.0654 1.38 0.0251 1.8402 0.066 1963 4 0.6953 0.1472 1.1812 0.0219 0.8434 0.0393 1964 1 NA NA 1.0744 0.0227 0.4306 0.0352 1964 2 0.4768 0.0532 1.4747 0.0235 0.5103 0.0935 1964 3 0.4886 0.044 0.9696 0.0219 1.3578 0.0352 1964 4 1.8921 0.0707 1.259 0.0219 1.0414 0.0243

2043 Table 2. (Continued) Year Quarter R1 R2 R3 Estimate CV Estimate CV Estimate CV 1965 1 1.9462 0.0629 1.3443 0.0226 0.7189 0.0373 1965 2 1.0566 0.0416 1.2581 0.0208 0.8005 0.0434 1965 3 0.63 0.0527 1.2008 0.0193 0.9768 0.0272 1965 4 1.9128 0.0601 1.2164 0.0197 0.9801 0.0261 1966 1 1.1281 0.0806 1.1356 0.0232 0.3987 0.0364 1966 2 1.0301 0.0497 0.9799 0.0255 0.4392 0.052 1966 3 1.4345 0.0927 1.1462 0.0318 0.7429 0.0306 1966 4 1.339 0.0638 1.1497 0.03 0.9419 0.0336 1967 1 1.0933 0.0572 1.2045 0.0281 0.4188 0.0429 1967 2 0.6578 0.0534 1.0749 0.0278 0.6077 0.0536 1967 3 0.5734 0.0853 0.7681 0.0314 0.8778 0.0354 1967 4 1.4478 0.0721 1.2251 0.0241 1.0438 0.0298 1968 1 1.4626 0.068 1.1879 0.0318 0.3634 0.0543 1968 2 1.3021 0.054 1.1616 0.0294 0.6328 0.0353 1968 3 1.7801 0.1181 1.2243 0.0279 0.9201 0.0291 1968 4 0.856 0.0941 1.3007 0.0319 0.6781 0.0344 1969 1 1.242 0.0595 1.1516 0.0374 0.6712 0.042 1969 2 1.0358 0.0798 1.034 0.0277 0.9178 0.0418 1969 3 0.5649 0.085 1.0658 0.0291 1.3675 0.0439 1969 4 NA NA 1.1454 0.0303 1.2466 0.0596 1970 1 1.3637 0.055 1.0918 0.0313 0.4071 0.0363 1970 2 1.568 0.0509 0.9018 0.0267 0.4143 0.0543 1970 3 0.8337 0.0464 0.8609 0.0281 0.5222 0.0595 1970 4 0.9419 0.0483 1.0011 0.0397 1.1512 0.0469 1971 1 1.3348 0.0413 0.961 0.0314 0.3771 0.0659 1971 2 0.8669 0.0354 0.876 0.0276 0.444 0.0418 1971 3 0.5777 0.0532 0.7788 0.0293 0.7602 0.0368 1971 4 0.9157 0.035 0.8522 0.0326 0.6094 0.0424 1972 1 0.9793 0.0485 0.8903 0.0333 0.2401 0.0445 1972 2 0.5564 0.0572 0.7549 0.0384 0.3867 0.0548 1972 3 0.5729 0.0574 0.9318 0.0432 0.5805 0.067 1972 4 0.8206 0.0593 0.9325 0.0551 0.6091 0.0436 1973 1 0.9463 0.0574 1.1362 0.0417 NA NA 1973 2 1.2766 0.0689 0.7557 0.0518 0.6063 0.0653 1973 3 0.6628 0.0548 0.9272 0.0485 0.8179 0.0596 1973 4 1.5564 0.0446 0.7951 0.0446 0.5487 0.0472 1974 1 1.2748 0.0548 1.0543 0.061 NA NA 1974 2 1.5844 0.0618 0.7235 0.1025 0.8005 0.0762 1974 3 0.7119 0.0509 0.6622 0.0385 0.6919 0.0658 1974 4 1.0803 0.0459 0.9344 0.0462 0.5855 0.0552

2044 Table 2. (Continued) Year Quarter R1 R2 R3 Estimate CV Estimate CV Estimate CV 1975 1 0.9755 0.0532 0.7082 0.0425 NA NA 1975 2 0.7687 0.0465 0.7005 0.0446 0.407 0.1308 1975 3 0.4089 0.0508 0.7012 0.0304 1.2098 0.0582 1975 4 0.8639 0.0486 0.6277 0.0548 0.6637 0.0464 1976 1 1.1124 0.0545 0.7171 0.0524 NA NA 1976 2 0.7352 0.0619 0.6672 0.0422 NA NA 1976 3 0.5182 0.0485 0.6128 0.0361 NA NA 1976 4 0.9 0.0594 0.9627 0.0555 1.1221 0.0614 1977 1 0.8047 0.0579 1.3057 0.0709 NA NA 1977 2 0.8656 0.0626 0.957 0.0654 NA NA 1977 3 0.4886 0.0662 0.8505 0.0465 NA NA 1977 4 0.7826 0.0588 1.1565 0.0437 0.7012 0.0924 1978 1 0.8851 0.0585 1.1715 0.072 NA NA 1978 2 0.8518 0.0763 0.7746 0.062 NA NA 1978 3 0.764 0.0685 0.9744 0.049 1.0316 0.0657 1978 4 1.1208 0.0573 0.6498 0.0622 0.8767 0.0749

2045 Table 3. Indices for late period (1979 – present ), standardized with vessel effects, for regions 1 to 3. Year Quarter R1 R2 R3

Estimate CV Estimate CV Estimate CV 1979 1 1.9644 0.0766 2.1795 0.0645 1979 2 1.136 0.09 1.5771 0.0416 1979 3 1.7571 0.0832 1.6896 0.0351 1979 4 2.717 0.0669 1.998 0.039 2.4446 0.0812 1980 1 3.2128 0.067 1.957 0.04 NA NA 1980 2 3.8218 0.0853 1.8981 0.0347 NA NA 1980 3 1.9365 0.071 1.5272 0.0315 NA NA 1980 4 2.2734 0.0699 1.6904 0.0297 1.0746 0.0838 1981 1 1.6888 0.0579 2.3419 0.0293 1.299 0.1087 1981 2 1.4325 0.0629 1.7118 0.0282 NA NA 1981 3 1.2679 0.0641 1.5194 0.0274 1.0216 0.1055 1981 4 1.4355 0.0579 1.5727 0.0247 1.5889 0.0841 1982 1 2.3721 0.0744 1.7611 0.0266 1.0498 0.1282 1982 2 2.0265 0.072 1.2921 0.0265 1.0776 0.1109 1982 3 1.5022 0.0696 1.2583 0.028 1.2937 0.0984 1982 4 2.0562 0.0725 1.4856 0.0252 1.2865 0.0765 1983 1 2.2144 0.0641 1.5447 0.0326 NA NA 1983 2 1.5082 0.0613 1.4429 0.0498 NA NA 1983 3 2.1273 0.0716 1.3982 0.0346 1.1804 0.0992 1983 4 1.5124 0.0699 1.7131 0.0273 2.3621 0.0947 1984 1 1.7104 0.0573 1.7196 0.028 NA NA 1984 2 1.2166 0.0689 1.5919 0.0335 NA NA 1984 3 1.9495 0.0903 1.446 0.0286 1.6883 0.0871 1984 4 2.028 0.0722 1.6403 0.0271 1.6041 0.0844 1985 1 1.8123 0.0825 1.8728 0.0268 NA NA 1985 2 1.0751 0.0658 1.4975 0.0254 2.7322 0.0877 1985 3 1.7785 0.0881 1.5057 0.0231 1.4439 0.0654 1985 4 1.7948 0.0587 1.5119 0.0223 1.0967 0.0679 1986 1 1.5072 0.0619 1.9562 0.0288 1.3391 0.0959 1986 2 0.7201 0.0693 1.6438 0.0379 2.2372 0.1063 1986 3 2.143 0.0859 1.5329 0.0285 1.4326 0.0655 1986 4 1.6126 0.0615 1.7166 0.0271 1.0554 0.0657 1987 1 1.8259 0.0513 1.8186 0.0289 1.0076 0.1091 1987 2 1.5562 0.0735 1.7743 0.0293 2.455 0.1039 1987 3 1.3421 0.0583 1.9854 0.0297 2.2106 0.0773 1987 4 0.9113 0.0567 2.098 0.029 1.5276 0.077 1988 1 1.0094 0.0489 2.0804 0.0277 1.1223 0.1019 1988 2 0.5511 0.0653 1.7963 0.0282 1.6293 0.0743 1988 3 1.1351 0.0649 1.8013 0.0255 1.3973 0.0651 1988 4 1.1311 0.0524 1.8625 0.025 1.4431 0.0633

2046 Table 4. (Continued). Year Quarter R1 R2 R3 Estimate CV Estimate CV Estimate CV 1989 1 1.0895 0.0434 1.712 0.0262 0.7742 0.1022 1989 2 1.1236 0.0783 1.4051 0.024 1.5053 0.0754 1989 3 1.1642 0.0663 1.4383 0.0227 1.4788 0.0658 1989 4 1.0474 0.0512 1.3327 0.0223 0.8855 0.0643 1990 1 1.4369 0.0426 1.5287 0.0257 1.2198 0.0845 1990 2 1.1584 0.0717 1.1447 0.0247 1.0456 0.0695 1990 3 1.2832 0.0665 1.0289 0.0227 1.0549 0.0604 1990 4 1.5657 0.0622 1.0714 0.0229 1.5615 0.0604 1991 1 1.0269 0.0457 1.3526 0.0261 0.6004 0.0548 1991 2 1.0609 0.0701 1.1016 0.025 0.7444 0.0462 1991 3 0.9999 0.063 1.0981 0.0219 0.9717 0.05 1991 4 1.2558 0.0619 1.2048 0.0236 1.2701 0.0702 1992 1 0.929 0.0435 1.3786 0.0248 0.6422 0.0658 1992 2 0.7143 0.0599 0.9308 0.0311 0.6915 0.0554 1992 3 0.8375 0.0522 1.0503 0.0245 1.2089 0.0621 1992 4 1.2225 0.0595 1.1371 0.0262 1.5302 0.0657 1993 1 0.76 0.0452 1.2118 0.0257 NA NA 1993 2 0.8474 0.0598 1.0102 0.0257 0.985 0.0566 1993 3 1.094 0.0644 1.1551 0.0235 1.3006 0.0505 1993 4 1.0681 0.0605 1.0666 0.0248 1.3218 0.0616 1994 1 0.6192 0.0548 1.1611 0.026 0.7414 0.0573 1994 2 0.7899 0.0667 0.9266 0.0245 1.2299 0.047 1994 3 0.7749 0.0747 0.8175 0.0223 1.0541 0.0434 1994 4 1.4827 0.0576 0.9068 0.0238 0.8985 0.0617 1995 1 1.1191 0.0493 1.175 0.0253 1.3989 0.0682 1995 2 0.8991 0.056 0.9829 0.0249 1.0096 0.0459 1995 3 0.5927 0.0531 0.9686 0.0229 1.0609 0.0414 1995 4 0.7318 0.0485 0.8726 0.0226 1.0419 0.061 1996 1 0.9711 0.0556 0.9847 0.0244 0.9116 0.0542 1996 2 0.713 0.0563 0.7845 0.0248 0.6791 0.0478 1996 3 0.8437 0.0516 0.7227 0.0219 1.0879 0.0537 1996 4 1.0905 0.0476 0.716 0.0217 0.9671 0.0603 1997 1 1.0786 0.0487 0.9033 0.024 0.5756 0.0631 1997 2 0.7554 0.0478 0.6538 0.0231 0.6858 0.0548 1997 3 0.6298 0.0685 0.6361 0.0214 0.8713 0.0581 1997 4 1.0717 0.0467 0.6677 0.0222 0.6984 0.0637 1998 1 0.9097 0.0395 0.8529 0.0259 NA NA 1998 2 0.8043 0.0472 0.8084 0.0253 0.5488 0.0508 1998 3 0.6673 0.0588 0.7211 0.0232 1.3781 0.0585 1998 4 0.923 0.0461 0.6889 0.0211 0.5825 0.0557

2047 Table 5. (Continued). Year Quarter R1 R2 R3 Estimate CV Estimate CV Estimate CV 1999 1 1.1549 0.0409 0.773 0.0239 0.5126 0.0743 1999 2 1.2668 0.0461 0.7308 0.0239 0.6729 0.0575 1999 3 0.5853 0.0619 0.6114 0.0241 1.1948 0.0555 1999 4 1.19 0.0501 0.8097 0.0255 0.8287 0.0507 2000 1 1.1542 0.0432 0.9908 0.0225 0.7358 0.0952 2000 2 1.2247 0.0411 0.8172 0.0231 0.7715 0.0525 2000 3 0.4835 0.0514 0.7289 0.0231 0.7144 0.05 2000 4 0.6625 0.0441 0.6372 0.0221 0.5218 0.0608 2001 1 1.4021 0.0393 0.8417 0.0215 NA NA 2001 2 1.1839 0.039 0.705 0.0233 0.4511 0.0655 2001 3 0.5788 0.0514 0.5506 0.0252 1.1941 0.0657 2001 4 0.6522 0.0434 0.4985 0.0239 0.7764 0.0645 2002 1 0.7296 0.0393 0.5775 0.0224 NA NA 2002 2 0.5532 0.0419 0.6582 0.0254 1.0371 0.0894 2002 3 0.6849 0.0917 0.6367 0.0286 1.1913 0.0795 2002 4 0.651 0.042 0.6011 0.024 1.0582 0.0831 2003 1 0.7329 0.045 0.7461 0.0241 NA NA 2003 2 0.9363 0.044 0.6502 0.0232 0.6125 0.0634 2003 3 0.2825 0.0647 0.5574 0.0238 1.2975 0.0582 2003 4 0.6541 0.05 0.4628 0.0234 0.7091 0.0772 2004 1 0.8406 0.0395 0.6433 0.0224 NA NA 2004 2 0.6519 0.0462 0.5127 0.0227 0.3022 0.0624 2004 3 0.564 0.0752 0.4459 0.0247 1.4377 0.0587 2004 4 0.4126 0.051 0.5137 0.0233 0.9199 0.0584 2005 1 0.6477 0.0354 0.6376 0.0201 NA NA 2005 2 0.7442 0.0466 0.5322 0.0203 NA NA 2005 3 0.4902 0.0733 0.4752 0.0211 0.7045 0.0592 2005 4 0.6829 0.0489 0.5346 0.0217 0.9058 0.084 2006 1 0.6422 0.039 0.7525 0.0231 NA NA 2006 2 0.6855 0.0585 0.5958 0.023 0.8619 0.0981 2006 3 0.6889 0.1048 0.5778 0.0234 0.8393 0.0654 2006 4 0.5613 0.0773 0.6081 0.0251 0.7887 0.0782 2007 1 0.5949 0.0623 0.7678 0.0251 NA NA 2007 2 0.5879 0.0946 0.6777 0.0243 NA NA 2007 3 0.6114 0.1084 0.5795 0.022 0.9369 0.0696 2007 4 0.5765 0.0943 0.6192 0.0234 0.7673 0.0691 2008 1 0.5841 0.0724 0.5727 0.0246 0.6724 0.1646 2008 2 0.3615 0.0777 0.5298 0.0212 0.3225 0.0922 2008 3 0.5289 0.0813 0.4807 0.0207 0.945 0.053 2008 4 0.6917 0.087 0.5638 0.0229 1.0887 0.0552

2048 Table 6. (Continued). Year Quarter R1 R2 R3 Estimate CV Estimate CV Estimate CV 2009 1 0.4393 0.0817 0.5817 0.024 0.4616 0.1052 2009 2 0.436 0.0853 0.4878 0.0212 NA NA 2009 3 0.4302 0.0766 0.3955 0.0207 0.8675 0.0675 2009 4 0.6614 0.0577 0.5637 0.0219 0.5509 0.0667 2010 1 0.4234 0.0811 0.5202 0.0228 NA NA 2010 2 0.512 0.083 0.4281 0.0219 0.3978 0.0702 2010 3 0.4699 0.0818 0.5203 0.0213 0.6764 0.053 2010 4 0.492 0.072 0.509 0.0223 0.6531 0.0841 2011 1 0.4185 0.0792 0.5371 0.0229 0.4708 0.0874 2011 2 0.4429 0.0887 0.4271 0.0213 0.6173 0.0626 2011 3 0.4453 0.0809 0.4097 0.0202 1.1663 0.0601 2011 4 0.7429 0.0853 0.4928 0.0203 0.4799 0.1212 2012 1 0.4797 0.0811 0.4556 0.022 0.7198 0.0797 2012 2 0.4691 0.1201 0.4305 0.021 0.4442 0.0676 2012 3 0.5477 0.0765 0.4182 0.0229 1.3591 0.0524 2012 4 0.4476 0.0759 0.6532 0.0223 1.1711 0.058 2013 1 0.5928 0.0935 0.7113 0.0236 1.1095 0.1173 2013 2 0.4198 0.0814 0.6102 0.0247 0.5563 0.0537 2013 3 0.4419 0.0815 0.5606 0.0259 1.2243 0.0571 2013 4 0.558 0.0713 0.82 0.0242 0.9367 0.07 2014 1 0.5956 0.0813 0.7827 0.0265 NA NA 2014 2 0.5852 0.0996 0.5407 0.0242 0.3921 0.058 2014 3 0.5072 0.0808 0.6043 0.0251 0.7806 0.067 2014 4 0.7591 0.0726 0.7045 0.027 0.4071 0.0764 2015 1 0.4631 0.0952 0.8081 0.0283 0.7669 0.0977 2015 2 0.4434 0.0888 0.6739 0.025 0.4021 0.0623 2015 3 0.6766 0.0822 0.661 0.0283 0.5228 0.0537 2015 4 0.6188 0.078 0.7877 0.0257 0.7057 0.0667 2016 1 0.4917 0.091 0.6818 0.0271 NA NA 2016 2 0.3819 0.096 0.4992 0.0262 0.4691 0.0736 2016 3 0.5181 0.0939 0.6014 0.0253 0.7605 0.0547 2016 4 0.9225 0.0817 0.7362 0.0263 0.7758 0.0775 2017 1 0.5527 0.091 0.6301 0.027 0.5274 0.0888 2017 2 0.3225 0.1 0.5724 0.0271 0.3428 0.0623 2017 3 0.9358 0.0879 0.5482 0.025 0.5643 0.0586 2017 4 0.5173 0.1572 0.7082 0.0246 0.6326 0.0785

2049 1

Figure 1. Map of the regional structures used to estimate bigeye CPUE indices.

2050

Figure 2. Estimates of coverage per year by each fleet in the analysis. Coverage is estimated by dividing bigeye catch reported in the available operational data by the Task II bigeye catch.

2051

Figure 3. Comparison plots showing the preliminary indices developed for each fleet, plotted together by region along with the joint indices. Each series is normalised to average 1. The annual versions of the plots are shown here to facilitate comparison.

2052

Figure 4. Frequency histograms of numbers of sets per year-quarter-grid cell stratum in three regional models for 1979-2017 with vessel effects, before removal of strata with fewer than 5 sets.

2053

Figure 5. Estimated CPUE series for bigeye region 1 (north), including time series for all years (top) both with (right) and without (left) vessel effects, and time series for 1959-78 without vessel effects, and 1979-2017 with vessel effects.

2054

Figure 6. Estimated CPUE series for bigeye region 2 (tropical), including time series for all years (top) both with (right) and without (left) vessel effects, and time series for 1959-78 without vessel effects, and 1979-2017 with vessel effects.

2055

Figure 7. Estimated CPUE series for bigeye region 3 (south), including time series for all years (top) both with (right) and without (left) vessel effects, and time series for 1959-78 without vessel effects, and 1979-2017 with vessel effects.

2056

Figure 8. Quarterly time series for region 2, including time series for 1959-78 without vessel effects, and 1979- 2017 with vessel effects (top left), time series for the full period without vessel effects (top right), and the combined time series (bottom left).

2057

Figure 9. Annual time series for region 2, including time series for 1959-78 without vessel effects, for 1979-2017 with vessel effects, for the full period without vessel effects, and combined time series covering the full period.

2058

Figure 10. Influence summary plot for each parameter in the standardization model for bigeye region 1 (north) for 1959-78 without vessel effects.

2059

Figure 11. Influence summary plot for each parameter in the standardization model for bigeye region 1 (north) for 1979-2017 with vessel effects.

2060

Figure 12. Influence summary plot for each parameter in the standardization model for bigeye region 2 (tropical) for 1959-78 without vessel effects.

2061

Figure 13. Influence summary plot for each parameter in the standardization model for bigeye region 2 (tropical) for 1979-2017 with vessel effects.

2062

Figure 14. Influence summary plot for each parameter in the standardization model for bigeye region 2 (tropical) for 1959-2017 without vessel effects.

2063

Figure 15. Influence summary plot for each parameter in the standardization model for bigeye region 2 (tropical) for 1959-2017 with vessel effects.

2064

Figure 16. Influence summary plot for each parameter in the standardization model for bigeye region 3 (south) for 1959-78 without vessel effects.

2065

Figure 17. Influence summary plot for each parameter in the standardization model for bigeye region 3 (south) for 1979-2017 with vessel effects.

2066

Figure 18. Plot showing the relative catch rates by grid cell, as estimated in the region 1 model for 1979 – 2017. Densities are indicated by colour and contour lines, with yellow indicating higher density than red.

Figure 19. Plot showing the relative catch rates by grid cell, as estimated in the region 2 model for 1979 – 2017. Densities are indicated by colour and contour lines, with yellow indicating higher density than red.

2067

Figure 20. Plot showing the relative catch rates by grid cell, as estimated in the region 3 model for 1979 – 2017. Densities are indicated by colour and contour lines, with yellow indicating higher density than red.

2068

Figure 21. Diagnostics plots for bigeye lognormal positive models in northern region 1, for all years (above) both with (right) and without (left) vessel effects, and (below) for 1959-78 without vessel effects, and 1979-2017 with vessel effects..

2069

Figure 22. Diagnostics plots for bigeye lognormal positive models in equatorial region 2, for all years (above) both with (right) and without (left) vessel effects, and (below) for 1959-78 without vessel effects, and 1979-2017 with vessel effects..

2070

Figure 23. Diagnostics plots for bigeye lognormal positive models in southern region 3, for all years (above) both with (right) and without (left) vessel effects, and (below) for 1959-78 without vessel effects, and 1979-2017 with vessel effects.

2071

Figure 24. Trends in temporal residuals by grid cell for bigeye region 1, from the model for 1979 to 2017 with vessel effects. The trends within each cell are estimated by regression of the residuals against year-quarter. Darker red represents decline and lighter yellow represents increase relative to the model average.

2072

Figure 25. Trends in temporal residuals by grid cell for bigeye region 2, from the model for 1979 to 2017 with vessel effects. The trends within each cell are estimated by regression of the residuals against year-quarter. Darker red represents decline and lighter yellow represents increase relative to the model average.

2073

Figure 26. Trends in temporal residuals by grid cell for bigeye region 3, from the model for 1979 to 2017 with vessel effects. The trends within each cell are estimated by regression of the residuals against year-quarter. Darker red represents decline and lighter yellow represents increase relative to the model average.

2074

Figure 27. Median residuals from the lognormal constant model per year-quarter (x-axis), by cluster (subplots), for bigeye in region 1 (north). Residuals are shown for 2 models: 1959-1978 without vessel effects (above), and 1979-2017 with vessel effects (below).

2075

Figure 28. Median residuals from the lognormal constant model per year-quarter (x-axis), by cluster (subplots), for bigeye in region 2 (north). Residuals are shown for 2 models: 1959-1978 without vessel effects (above), and 1979-2017 with vessel effects (below).

2076

Figure 29. Median residuals from the lognormal constant model per year-quarter (x-axis), by cluster (subplots), for bigeye in region 3 (south). Residuals are shown for 2 models: 1959-1978 without vessel effects (above), and 1979-2017 with vessel effects (below).

2077

Figure 30. Bigeye residuals for region 1 (north) by cluster. Median residuals are mapped by 5° cell for the periods 1959-1978 without vessel effects (above), and 1979-2017 with vessel effects (below).

2078

Figure 31. Bigeye residuals for region 2 (tropical) by cluster. Median residuals are mapped by 5° cell for the periods 1959-1978 without vessel effects (above), and 1979-2017 with vessel effects (below).

2079

Figure 32. Bigeye residuals for region 3 (south) by cluster. Median residuals are mapped by 5° cell for the periods 1959-1978 without vessel effects (above), and 1979-2017 with vessel effects (below).

2080