Estimating trends and magnitudes of bycatch in the tuna fisheries of the Western and Central Pacific Ocean

Minimising the unintended capture of fish, marine mammals, reptiles, seabirds and other marine organisms is an important component of responsible fisheries management

Quantifying the magnitude of bycatch is an important first step for prioritising policy making to minimise bycatch (Alverson et al., 1994;Lewison et al., 2005;Pauly et al., 2002). Once an understanding is obtained on the magnitude by taxa, fishery, gear type and region, strategies to monitor and mitigate risks can be developed and implemented (Gilman, 2011;Gilman et al., 2019;Wilcox & Donlon, 2007). However, estimating levels of commercial fisheries bycatch can be challenging due to a lack of observer coverage (Kennelly, 2020; Lewison et al., 2014;. While reporting by the vessel on the quantity of retained catch of commercially important species is a requirement for many fisheries (FAO, 2020) the requirement to report on discarded (or minor retained) species is often a duty of independent observers (Gilman et al., 2014). Observer coverage rates, however, are typically low and often not representative of the fishery effort (Nicol et al., 2013). Despite such challenges global estimates of bycatch and discarding have been attempted (Gilman et al., 2020;Pérez Roda et al., 2019;Zeller et al., 2018). Comparing across fishery types and regions, these analyses indicate that discarding rates for fisheries targeting tunas and pelagic fish are lower than for other fisheries (Gilman et al., 2020) and that discard rates in the western and central Pacific Ocean are lower in comparison to neighbouring regions and other ocean basins (Zeller et al., 2018).
The global tuna catch in 2020 has been provisionally estimated at just over 4.4 million tonnes, with the fisheries of the western and central Pacific Ocean accounting for 56% of this catch (Williams & Ruaia, 2022). Longline and purse seine fisheries account for 80% of the catch in the western and central Pacific Ocean (Williams & Ruaia, 2022). The stocks of the western and central Pacific Ocean (elasmobranch) and numerous species of non-target finfish (teleosts; Pacific Community Data Holdings). However, the magnitude and temporal trend of these interactions is largely unknown or not accurately quantified.
The Western and Central Pacific Fisheries Commission has required all Purse seine fishing activity undertaken within the area bounded by 20°N and 20°S (excluding trips with fishing activity within the national jurisdiction of a single coastal state) to be independently observed and reported upon since 2010 (WCPFC CMM 2008-01). It has also required that at least 5% of longline fishing activity on vessels >24 m in length is observed since 2012. In practise, coverage rates of available observer data for longline and purse seine fisheries in the WCPFC Convention Area have often been lower than the mandated minimum rates, and in the case of longline fisheries have varied markedly between fleets (Panizza et al., 2022).
The observer data were used to estimate catch rates, which were then applied to reported effort data from vessel logbooks to obtain catch estimates for the period 2003 to 2019. Estimates for 2020 and 2021 were not included due to the reduced levels of observer coverage caused by the COVID-19 pandemic. The derived estimates provide the first whole of jurisdiction analyses on the magnitude and catch trends of bycatch associated with the industrial tuna fisheries operating in the WCPFC Convention Area.

| Data-longline fisheries
The analysed longline observer data set included data from the WCPFC Regional Observer Programme, as well as additional data held by the Pacific Community (SPC

| Data-purse seine fisheries
The analysed purse seine observer data set included data from the Set type has been demonstrated to have a strong effect on catch rates and compositions in purse seine fisheries (Amandè et al., 2010;Pilling et al., 2015). Intentional setting on schools associated with whales and whale sharks has been prohibited since 2013 (WCPFC CMM 2011-03).

| Catch estimation groups
Catch estimates were generated for species, or groups of species, referred to as estimation groups.

| Longline catch rate and catch estimation
Reported catches from vessel logbooks were used for albacore, bigeye, yellowfin and skipjack tuna (Katsuwonus pelamis, Scombridae), and for billfish species, and were assumed to be known without error. For the remaining 34 estimation groups, Generalised Estimating Equations were fitted to observer data and used to model longline catch rates (Table 1). Models were fitted using the R package 'geepack' (Højsgaard et al., 2006) in R v4.1.1 (R Core Team, 2021). Poisson-like error structures were used where possible, with a two-stage delta-lognormal modelling approach implemented where necessary to account for zero-inflation (see Table 1). An 'exchangeable' working correlation structure was assumed, where residuals from observations from the same observer trip are correlated, with a shared correlation parameter for all observer trips. It was not possible to fit models with exchangeable correlation structures for all estimation groups, in which case independence between residuals within trips was assumed ( Table 1).
where k is the number of estimated parameters. The random draws of parameter values were then used to generate 1000 estimated catch rates for each record of effort. Estimated catches were then obtained by taking the product of the catch rates and the effort. Porbeagle shark catch rates and catches were estimated using data south of 20°S as the species is likely absent north of this latitude in the Pacific Ocean (Francis et al., 2008). The unit of estimated longline catch was individuals for all estimation groups, that is the unit used by observers when recording the catch.

| Purse seine catch rate and catch estimation
Reported catches from vessel logbooks were used for skipjack, yellowfin and bigeye tuna. For the remaining 45 estimation groups (Table 2), observer data were used to estimate catch rates for unobserved sets. Presence/absence models were fitted to observer data using Generalised Estimating Equations using the R package 'geepack' (Højsgaard et al., 2006)  The specification of the presence/absence models was: where subscripts i and j refer to observer trip and set number, respectively, P ij denotes whether captures of the estimation group were observed, f n represent natural cubic splines and ϕ is a variance inflation parameter.  (Neubauer et al., 2018). To mitigate downwards bias in catch estimates, whale and whale shark sets were treated as free school sets when estimating catch rates and catches of whale sharks and marine mammals, both in the observer and reported effort data set.

| Coverage of catch estimates
The catch estimates cover longline and purse seine fishing in the WCPFC Convention Area (Figure 1)

| Post-processing of catch estimates
Estimated catches were summed across relevant strata to obtain estimates at more aggregated resolutions, for example, annual totals.
Estimates of total catch were generated for different 'species types' TA B L E 2 Estimation groups for purse seine catches and their corresponding species type and catch unit (ordered by species type and then alphabetically by scientific name). and 2). Estimates were combined across estimation groups by assuming that estimation group catch estimates were independent.
Summary statistics were then computed, using the 2.5 and 97.5 percentiles to generate 95% confidence intervals.

| Precision of estimated longline catch rates at differing levels of monitoring coverage
Simulations were used to explore the precision of estimated longline catch rates at differing levels of fisheries observer coverage, using a stratified sub-sampling approach similar to Lawson (2004; Supporting Information). There are a range of options available for allocating electronic and/or observer monitoring coverage within a longline fleet. Two approaches were used: a target coverage rate of 5%, 10% or 20% of trips, with full coverage of sets within a trip, and, partial coverage of all trips, with a target coverage rate of 10%, 20% or 50% of sets for each trip.
Simulations were undertaken separately for two broad regions within the WCPFC Convention Area: the area from 10°S to 30°S, primarily vessels targeting albacore tuna, and the area from 10°S to 20°N, primarily vessels targeting yellowfin and/or bigeye tuna.
Eleven species were selected for each region, covering a range of species including target species, bycatch species and species of conservation interest.

| Estimation of hooks between float (HBF) for reported longline effort data
The accuracy of the predictive model of HBF was considered adequate. HBF was estimated with a classification accuracy of 66% for the testing data set and predictions accurate to ± one HBF class for 91% of records in the testing data set (Table S1). Uncertainty in the overall proportions of inferred shallow set and deep set effort was lower for 2007 onwards, when HBF information was more widely available in reported longline effort data.

| Patterns in reported effort and observer data coverage
Total reported longline effort in the WCPFC Convention Area aver-  S5). Observer coverage pre-2010 was relatively low for free school sets compared to coverage rates of associated sets ( Figure S6).

| Longline catch, catch rates and composition
The estimated catch composition of modelled longline fisheries in the WCPFC Convention Area from 2003 to 2019 was dominated by tropical tunas and albacore, representing 67% of the catch by numbers ( Figure 2 Estimated longline catch compositions and catch rates varied by set depth as inferred from HBF ( Figure 3). Catch rates of tropical tunas and albacore were higher for deep sets than shallow sets, whereas catch rates of billfish, other finfish species, elasmobranchs, marine mammals and sea turtles were highest for shallow sets. hooks-between-float specific effort. For example, the temporal trends for estimated mahi mahi were more influenced by shallow F I G U R E 2 Total estimated annual catch ('000 individuals; grey region provides 95% CIs) of longline fisheries in the WCPFC Convention Area by species type (see Table 1). Estimated catches do not cover domestic fisheries of the Philippines, Vietnam and Indonesia and former shark-targeted fisheries in the exclusive economic zones of Papua New Guinea and the Solomon Islands. Reported catches were used where available, covering tropical tuna, albacore and billfish and were assumed to be known without error.

| Purse seine catch, catch rates and composition
The estimated catch composition of modelled purse seine fisheries in the WCPFC Convention Area from 2003 to 2019 was dominated by tropical tunas (Figure 4) Table 1). Estimated catch rates do not cover domestic fisheries of the Philippines, Vietnam and Indonesia and former shark-targeted fisheries in the exclusive economic zones of Papua New Guinea and the Solomon Islands. Reported catch and effort from hook-betweenfloats specific aggregated data were used to calculate catch rates for tropical tuna, albacore and billfish.
accounted for by blue marlin (53% of total billfish catch), black marlin  Table 2). Catch units are '000 tonnes for tropical tuna and other finfish, and '000 individuals for elasmobranchs, sea turtles and marine mammals. Estimated catches do not cover domestic fisheries of the Philippines, Vietnam and Indonesia and purse seiners operating in temperate waters off Japan and New Zealand. Reported catches were used for tropical tuna and assumed to be known without error.  Table 1) and set type. Catch rate units are tonnes per set for tropical tuna and other finfish, and otherwise individuals per set (for billfish, sharks & rays, marine mammals and turtles). Reported catches were used to calculate catch rates for tropical tuna. Catch rates do not cover domestic fisheries of the Philippines, Vietnam and Indonesia and purse seiners operating in temperate waters off Japan and New Zealand. Purse seine sets are defined as: schools associated with anchored fish aggregating devices (aFAD), schools associated with drifting fish aggregating devices (dFAD), schools associated with drifting natural logs (Log), free schools (FS), schools associated with whales (Whale), schools associated with whale sharks (Whale shark).

| Precision of estimated longline catch rates at differing levels of monitoring coverage
Coefficients of variation (CVs) were generally higher in years with lower numbers of observed sets and for species that were more rarely caught (Tables S5-S8). CVs demonstrated strong between species variation for a given target coverage rate. CVs at a departure year bin resolution for a target coverage rate of 10% of sets (and partial coverage of all trips) were generally lower or equivalent to those for a target coverage rate of 20% of trips (with full coverage of an observed trip). Exceptions to this were leatherback and green turtle, the rarest observed species considered, for which CVs for a target coverage rate of 10% of sets (and partial coverage of all trips) were more consistent with those for a target coverage rate of 10% of trips. CVs at a resolution of departure year bin and flag were higher, and more variable, than at a resolution of departure year bin.

| DISCUSS ION
Assessing the impact of global fisheries has become increasingly important as responsible authorities seek to demonstrate that they are employing ecologically sustainable fishing practices. Demonstrating Historically, estimates and analyses of bycatch have been hampered by inadequate temporal and spatial observation across fishery fleets and gears (Lewison et al., 2014;Mace et al., 2014). Our estimates were similarly influenced by the inadequacies of sufficient observation. This uncertainty is present both due to the level of coverage by observers across fleets and gears and by their capacity to monitor all catch related activities. In particular, our estimates of bycatch for the longline fisheries were complicated by the coverage of available observer data, and for some years, the coverage of aggregated effort data specific to hooks between floats. As such, the catch estimates presented here must be viewed in the context of the limitations of the data set, and the methodology used to obtain the estimates.
A recent study has also highlighted the uncertainty that is generated by the limited capacity to observe across all activities during a catch event on purse seiners (Forget et al., 2021). Across three trips, observations by fisheries observers underestimated shark catches for the majority of sets, resulting in underestimation of shark catch at a trip level of between 10% and 40%. As such, it is reasonable to expect that the estimates presented here underestimate the actual number of individuals caught in the large-scale equatorial purse seine fishery even when fisheries observer coverage rates are 100%.
Additionally, there may be inaccuracies in species identifications by observers . In this context our bycatch estimates should be interpreted as the bycatch that would have been recorded by observers with 100% coverage of fishing events, rather than estimates of the total bycatch encountered.
These observational inadequacies are compensated to some degree by the approach used to generate uncertainty in catch estimates. However, residual diagnostics indicated a lack of fit for a number of the lognormal longline model components, and spatial patterns in residuals for most longline and purse seine catch rate models, particularly for longline models for commonly observed estimation groups. This appears to reflect the inability of the longline catch rate models to adequately capture both targeting behaviour and spatial variation in catch rates more generally. Recent increases in the spatial coverage of longline observer data should support explicit inclusion of spatial effects in catch rate models in the future.
There may also be value in considering other approaches to account for targeting in future analyses of longline catches, for example, by fitting separate catch rate models to appropriate subsets of available observer data informed by variables such as catch compositions or the spatial distribution of fishing effort. Further refinements to the modelling approach should also be considered in future work, for example, separate estimation of catch rate and catches for marine mammals at more detailed taxonomic groupings. More recently, observer coverage rates of longline and purse seine fisheries in the WCPFC Convention Area have been impacted by COVID-19. Overall coverage rates have been lower since mid-2020 and the spatial coverage of available observer data has been less representative of overall fishing effort (Panizza et al., 2022). It is reasonable to assume that future estimates of bycatch and bycatch rates will be less precise for the period with reduced observer coverage rates. Additionally, the reduction in the representativeness of observer data may introduce bias in catch estimates. In combination, the impacts of COVID-19 are likely to compromise the ability to detect temporal trends in bycatch in recent years.
To evaluate the efficacy of the methodology applied we compared the reported catches from longline vessel logbooks to the estimates generated from the observer data using our modelling approach for the target albacore, bigeye, yellowfin, skipjack and billfish species ( Figure S7).  Our estimates provide important information for calculating population sizes, trends and mortalities that inform extinction risk assessments (e.g. IUCN Red List) and vulnerability assessments (Walker et al., 2021). Sustainability certifications and international trade requirements (e.g. the Convention on the International Trade of Endangered Species of Wild Flora and Fauna, CITES) are increasingly requiring quantification of the magnitude and trends in bycatch. For example, determining and restricting marine mammal bycatch to be within or below potential biological removals is a requirement for import of seafood into the United States (Félix et al., 2021).
Parameterisation and tuning of ecosystem models constructed to evaluate the ecosystem effects of differing fishing activities are also dependent on a baseline of information on the magnitude and trends in bycatch species (Griffiths et al., 2019). We did not estimate seabird bycatch due to insufficient observer coverage for a number of longline fleets pre-2015 operating in high latitude areas where fisheries pose the greatest risk to seabirds (Waugh et al., 2012).
However, at least nine threatened seabirds interact with WCPFC fisheries (IUCN, 2022). With improved observer coverage the methods applied here for other taxa could be adapted for application to seabird observations to generate time series of bycatch estimates.  . Food security is increasingly becoming an issue of concern for Pacific Island Countries and Territories and the leaders of the Pacific Islands have committed to increase the availability of fish from tuna fisheries for local consumption by 40,000 tonnes per year to meet the food security needs of their populations (FFA and SPC, 2015). Developing opportunities to access fish from tuna fisheries through offloading of bycatch (e.g. non-threatened teleost species) from the industrial fisheries is considered an important component to this supply chain (James et al., 2018). Our baseline estimates for bycatch magnitude and trends provide a necessary first step for evaluating the feasibility of this option to meet all or some of the identified food security gap.
Our study has established a baseline of catches and catch com- The contribution of the many fisheries observers who accurately recorded the catches of bycatch used in this study during their deployment on tuna fishing vessels is duly acknowledged. We thank the two anonymous reviewers whose valuable comments helped improve and clarify the manuscript. The study was conducted through support provided by the Western and Central Pacific Fisheries This publication was produced with the financial support of the European Union and Sweden. Its contents are the sole responsibility of the authors and do not necessarily reflect the views of the European Union and Sweden.

DATA AVA I L A B I L I T Y S TAT E M E N T
The analysed observer and aggregated catch and effort data sets are not in the public domain. Access to these data sets can be requested (see https://www.wcpfc.int/doc/data-02/rules -and-proce dures -prote ction -acces s-and-disse minat ion-data-compi led-commi ssion).