Are tuna always hungry? A deep dive into stomach-fullness measures in the western and central Pacific Ocean

Stomach sampling

In total, 8229 stomachs of tropical tunas (3491 skipjack, 3436 yellowfin and 1302 bigeye, Supplementary Table S1) were withdrawn from the Pacific Marine Specimen Bank (PMSB). They were sampled from fish captured across the WCPO (125°E–120°W, 32°S–12°N) between 2001 and 2021 (Fig. 1) and stored frozen at −20°C until analysis. All the samples collected as part of the PMSB, including the stomach samples selected in this study, were processed in compliance with the PMSB standard operating procedures regarding methods of sampling, storage, and shipment, and were treated the same way. The time between sampling and analysis of the stomachs in this study varied from several months to 10 years, and we assume that the standard operating procedures used prevented differences in processing time affecting our measures of stomach fullness. The stomachs were examined and assessed for stomach fullness at SPC fisheries laboratory in Noumea, New Caledonia. Tunas were captured by commercial fishing and research vessels, mainly by purse seine, longline, pole-and-line and danglers-handline gear, some of which were associated to a FAD at capture. All stomachs selected within the framework of this study were examined following the same protocol regardless of the species, fishing method, sampling time and year of analysis.

Fig. 1.

Sampling locations (top) and distribution per year (bottom) for the 8229 tuna stomachs analysed in this study, colour coded by species. SKJ, skipjack tuna; YFT, yellowfin tuna; and BET, bigeye tuna.

Fullness coefficient

All stomachs were examined according to the following protocol. Each stomach was defrosted in the open air of the laboratory before being opened longitudinally without altering the content. If baitfish (such as sardines or anchovies) were used during fishing operations (i.e. on pole-and-line, handline or longline vessels), and recorded in the fishing logbook for proper identification, they were removed from the stomach contents prior to further analysis. The fullness coefficient was then determined by visually classifying the volume of prey in the stomach into one of the five categories (0, empty; 1, less than half full; 2, half full; 3, more than half full; 4, full).

Fullness metric

After removal of baits and the estimation of the fullness coefficient, all non-empty stomachs (i.e. those with fullness coefficients of 1, 2, 3 and 4) were examined as follows: the stomach contents were recovered using a 200-μm mesh sieve to rinse the prey from mucus to help with the identification, while retaining all small items such as otoliths or cephalopod beaks. Prey was sorted and identified to the lowest taxonomic level and each prey item was then weighed individually to the nearest 0.01 g. A ‘prey weight’ was calculated for each stomach by summing the weight of all the prey items measured in each stomach.

To calculate the fullness metric for each fish, we used a slight modification of Herbold’s (1986) ‘percentage fullness’ calculation, applying the following formula

where prey weight is the observed weight of all prey in the stomach and mean max. prey weight is the mean maximum prey weight at length predicted from species-specific log–log regression models of observed prey weight (g) for all stomachs classified as full (i.e. with a fullness coefficient of 4) (skipjack: n = 120, yellowfin: n = 106, bigeye: n = 55) as a function of fork length (mm) (see Supplementary Fig. S1–S3).

We did not filter for the maximum observed weight for each length class as Herbold proposed, excluding observation of decreasing weights for increasing size, because we wished to us establish a ‘typical’ full stomach baseline, not the maximum that was possible.

Binary fullness metric

We also derived a simpler binary measure of stomach fullness. Fish with a fullness coefficient or fullness metric equal to 0 were classed as having an ‘empty’ stomach. Fish with a fullness coefficient or fullness metric greater than 0 were coded as having an ‘non-empty’ stomach. Although being a coarser measure of fullness, this binary metric allowed us to compare the inference gained from using simple v. more complex measures of stomach fullness as response variables in our models (see ‘Modelling’ section for details).

Exploratory data analysis

As the fullness coefficient is a categorical variable and the fullness metric is continuous, the exploratory analysis of these two metrics consisted of simple descriptive statistics to compare them and to assess their relationship. We also inspected the relationship between the binary fullness metric and each covariate included in our models (see ‘Covariates’ section).

Covariates

We hypothesised that the following covariates may explain some of the observed variability in our measures of stomach fullness: two anthropogenic factors, the fishing gear type (‘Gear’), and the floating-object association (‘School’) and one environmental factor, the time of day (‘Time of day’). Spatial covariates such as Longhurst region were explored but did not explain variability. In the case of ‘Gear’, different modes of fishing require the target species to behave in particular ways to be susceptible to the fishing method. ‘Gear’ was a binary variable (i.e. passive or active gear). Passive gear are left passively soaking in the water and require fish to actively bite a hook or otherwise become entrapped by the gear (i.e. in our study: pole-and-line, handline, rod and reel, troll, and longline), whereas active gear are manipulated to catch fish regardless of their individual behaviour (i.e. in our study: purse seine). Such differences in gear have been hypothesised to influence the catchability of fish in response to their behaviour or satiation, and thus observations of stomach fullness. For example, Bertrand et al. (2002) suggested that longline, a passive gear, may bias catch towards fish with emptier stomachs when they are experiencing poor feeding success. The association of fish around floating objects, including around FADs, which are widely used in the Pacific, has been hypothesised to be linked to the search for food (Hallier and Gaertner 2008; Leroy et al. 2013) and may be responsible for some of the patterns we observe in stomach fullness. Skipjack and juvenile yellowfin and bigeye tuna often aggregate in dense schools around floating objects at dawn, and increased instances of empty stomachs have been observed in fish caught when associated with such objects (Ménard et al. 2000; Machful et al. 2021). To disentangle the potential effects of dawn and association with floating objects, we therefore include a ‘School’ covariate to indicate whether fish were caught in such an associated aggregation. ‘School’, which varies across purse seine, pole-and-line, danglers-handline gear, but does not apply to longline gear, was a binary variable, with ‘Associated’ including fish captured on anchored FADs, drifting FADs, and in fishing sets made around logs and whales, and ‘Not-associated’ including fish captured in free schools or at seamounts. The distribution of samples according to the categorical covariates (‘Gear’ and ‘School’) is summarised in Table S2. ‘Time of day’ may be linked to different feeding cycles, the availability of prey or evasion of predators, and is likely to affect the observations of stomach fullness temporally. Several studies have proposed that small tunas have maximal feeding success over the course of daylight hours (e.g. Ortega-Garcia et al. 1992; Ménard et al. 2000), and so here we test whether such a covariate influences our observations of stomach fullness. ‘Time of day’ was treated as a continuous variable covering the full 24-h cycle but cannot be estimated for tuna captured by longline due to the long soak times of this gear and, thus, the inability to know at what time the fish was hooked and, consequently, models using this covariate excluded longline samples. Furthermore, fish length was an available covariate to be used in the modelling, allowing us to test for any ontogenetic effects on stomach fullness. However, preliminary analysis suggested that there was no strong relationship between fish length and the fullness metric for each species (Fig. S4A). In addition, length was correlated with Gear (i.e. passive gear types caught bigger fish) and School (see Fig. S4B, C). Consequently, fish length was not explored further in the modelling.

Modelling

For each tuna species, we fitted a series of statistical models to the continuous (i.e. fullness metric) and discrete (i.e. fullness coefficient, binary fullness metric) response variables to explore the effects of the above-mentioned explanatory covariates and compare the utility of the different fullness metrics for this purpose (Table 1). Because the continuous ‘Time of day’ covariate was not available for the longline records, we fitted two models to the response variable from each dataset, per species (i.e. in total, we attempted to fit 18 models). The first model formulation, fitted to the ‘Reduced’ datasets (i.e. without longline data), incorporated fixed effects of ‘Gear’ and ‘School’, and a separate cyclic penalised cubic regression spline smooth was used to model ‘Time of day’ for each level of ‘School’ (i.e. M1, M3 and M5, see Table 1). The inclusion of an interaction term here enabled exploration of whether the response to ‘Time of day’ (by species) varied by ‘School’, or not. The second model formulation included the ‘Gear’ fixed effect only and was fitted to the ‘Full’ datasets (i.e. with longline data included, see M2, M4 and M6 Table 1). Because a large proportion of stomachs came from fishing events with very small sample sizes (i.e. 44% of SKJ, 80% of BET and 61% of YFT observations came from fishing events with fewer than five fish), it was not feasible to incorporate a random effect to account for any potential commonality of the fullness metric of stomachs within the same fishing event (see Discussion). All modelling was undertaken in R (ver. 4.4.0, R Foundation for Statistical Computing, Vienna, Austria, see https://www.r-project.org/) using the glmmTMB (ver. 1.1.9, see https://cran.r-project.org/package=glmmTMB; Brooks et al. 2017), MASS (ver. 7.3-60.2, see https://cran.r-project.org/package=MASS; Venables and Ripley 2002), mgcv (ver. 1.9-1, S. Wood, see https://cran.r-project.org/package=mgcv/; Wood 2011) and DHARMa (ver. 0.4.6, available at https://CRAN.R-project.org/package=DHARMa) packages. R code and data to reproduce all analyses are available on request to the PMSB.

Exploratory data analysis

The proportion of empty stomachs was high, especially for skipjack (44.6%) and bigeye tuna (33.5%) and slightly lower for yellowfin tuna (16.5%) (see Fig. 2a). The proportion of stomachs with a fullness coefficient of 1 (i.e. a stomach with a very small amount of food) was also high regardless of species (39.4% for skipjack, 41.3% for bigeye tuna and 63.2% for yellowfin tuna) (Fig. 2a). The distribution of the calculated fullness metric was very skewed across all species (skipjack: mean = 0.13, range: 0–3.46; yellowfin tuna: mean = 0.2, range: 0–5.5; bigeye tuna: mean = 0.22, range: 0–5.25) (see Fig. 2b). Looking at the relationship between the two measures, there was a large variation of observed fullness metric within each fullness coefficient level (see Fig. 2c).

Fig. 2.

Histograms of (a) fullness coefficient and (b) fullness metric for each species, and (c) box plots of fullness metric for each level of the fullness coefficient. SKJ, skipjack tuna; YFT, yellowfin tuna; and BET, bigeye tuna.

We observed with the binary fullness response that regardless of species, individuals are more likely to have empty stomachs if caught on active gear (purse seine) and non-empty stomachs if caught on passive gear (e.g. longline) (Fig. 3). For skipjack and bigeye tuna, individuals are more likely to have empty stomachs if caught in associated schools and non-empty stomachs if caught in not-associated schools. We did not observe a marked difference between associated and not-associated schools for yellowfin tuna. Finally, the proportion of fish having an empty stomach peaked just prior to dawn, falling away sharply as the day progressed, regardless of species.

Fig. 3.

Plots of the binary fullness metric, showing the proportion of empty stomachs (light colour) and non-empty stomachs (dark colour) by Gear (first column; full dataset), School (middle column; full dataset), and Time of day (last column; reduced dataset). (a) SKJ, skipjack tuna; (b) YFT, yellowfin tuna; (c) BET, bigeye tuna.

Investigating the fullness coefficient and the fullness metric

Regardless of species, models fitted to the fullness coefficient on the reduced dataset (i.e. categorical response variable, M1, see Table 1) and the full dataset (M2, see Table 1) had extremely poor discrimination and were able to predict only empty or non-empty stomachs. For the highly skewed continuous fullness metric, regardless of species, models fitted to the reduced dataset (M3, see Table 1) and the full dataset (M4, see Table 1) did not meet model assumptions or failed to converge. These response variables were not explored further.

Models for binary response variable

Models fitted to the binary response variable (empty or non-empty stomach) by using the reduced data set (M5, see Table 1) met model assumptions; however, the results varied in their ability to discriminate between empty (0) and non-empty stomachs (1) among species. The models for skipjack and bigeye tuna were the worst-performing models, with poor predictive discrimination between observed empty and non-empty stomachs (Fig. 4a, c). Non-empty yellowfin tuna stomachs typically had high predicted probability of being non-empty (Fig. 4b), whereas empty stomachs had large variation in predicted probability. Similar to the reduced dataset models, regardless of species, the full dataset models (M6, see Table 1) were found to meet model assumptions. For skipjack, 426 of a total of 1557 observed empty stomachs were correctly predicted to be empty (27.4%), and 1796 of a total of 1934 observed non-empty stomachs were correctly predicted to be non-empty (92.8%). For yellowfin, 284 of 566 observed empty stomachs were correctly predicted to be empty (50.1%), and 2726 of 2870 observed non-empty stomachs were correctly predicted to be non-empty (94.9%). For bigeye, 147 of a total of 436 observed empty stomachs were correctly predicted to be empty (33.7%), and 821 of a total of 866 observed non-empty stomachs were correctly predicted to be non-empty (94.8%). These models were better at predicting non-empty stomachs than empty ones, regardless of species.

Fig. 4.

Box plots of the predicted probability of the stomach being non-empty, on the basis of Model M5 (i.e. binary response variable (empty or non-empty) fitted to the reduced dataset, see Table 1). (a) SKJ, skipjack tuna; (b) YFT, yellowfin tuna; (c) BET, bigeye tuna.

Models using the binary response variable (M5 and M6) confirmed that ‘Gear’ was a significant explanatory variable regardless of species. Active gear always predicted a lower probability of having a non-empty stomach (than did passive gear), regardless of species. Reduced-dataset models (M5) confirmed that the ‘Time of day’ was also a significant explanatory variable, regardless of species, and that ‘School’ association was a significant explanatory variables for skipjack and bigeye tuna.

The interaction between ‘Time of day’ and ‘School’ inferred that in associated schools, the probability of non-empty stomachs was lowest before dawn and increased throughout the day, to a maximum just before sunset for skipjack and yellowfin tuna and a little earlier for bigeye tuna (Fig. 5). The effect of ‘Time of day’ was weaker but still present in not-associated schools. There was a lower chance of having a non-empty stomach early in the morning for skipjack and yellowfin tuna, and a shift towards dawn for bigeye tuna, which presented an uncertainty in its results because of the lack of data.

Fig. 5.

Partial effects of ‘Time of day’ by ‘School’ for the three species of tuna predicted from Model M5 (see Table 1) for the binary response variable (empty or non-empty) fitted to the reduced dataset. (a) SKJ, skipjack tuna; (b) YFT, yellowfin tuna; (c) BET, bigeye tuna. Solid lines represent the likelihood of having a non-empty stomach on the scale of the link function, and dashed lines represent the 95% confidence interval. Dark zones denote the night-time period.

Stomach-fullness measurements

The categorical fullness coefficient remains the first measure recorded when assessing fullness for all tuna stomach samples archived in the PMSB. Although classification to the ‘0’ category, representing an empty stomach, is unambiguous, the subjectivity inherent to this measure is confirmed by the large variability observed in fullness metric values within each of fullness coefficient categories from ‘1’ to ‘4’, representing the non-empty stomachs (see Fig. 2c). This spread of fullness metric values for non-empty stomachs is unsurprising, given the qualitative nature of the fullness coefficient. That said, its simplicity, ease of application and the positive relationship evident in Fig. 2c do highlight the potential value of the fullness coefficient as an initial estimate of stomach fullness, one particularly well suited to observers at sea where collection of precise prey-weight measurements is typically not possible.

The calculation of the continuous fullness metric in our study also warrants some discussion. The key difference between our fullness metric calculation and Herbold’s (1986) ‘Percentage fullness’ calculation lies in the derivation of the denominator term in Eqn 1. The log–log models depicted in Fig. S1–S3 aimed to predict the ‘typical’ maximum prey weight at length for each tuna species on the basis of all data available for the individuals we classified as having full stomachs (i.e. with a fullness coefficient of 4). By contrast, Herbold’s approach involved filtering the data for the maximum observed prey weight for each length class and excluding records based on the reasoning that larger fish have a potential maximum prey weight at least as heavy as that of a smaller fish. The multiple regression models fitted by Herbold (1986) to quantify the denominator in his ‘Percentage fullness’ calculation were therefore predicting the ‘absolute’ maximum prey weight at length, rather than the typical maximum prey weight that we use in Eqn 1.

Our decision to use the typical maximum prey weight as a reference follows the rationale that prey weights of full tuna stomachs are likely to vary considerably depending on the diversity of prey types consumed. Tropical tunas of the WCPO are known to predate on a wide array of taxa including fish, crustaceans, cephalopods and gelatinous prey (Ménard et al. 2000; Bertrand et al. 2002; Allain 2004, 2005). Little is known about how prey species’ diversity, availability, size, and mass vary across space and time in the WCPO, and how these are reflected in tropical tuna diets through ontogeny (but see Olson et al. 2016; Fuller et al. 2021). Under these circumstances, we considered it sensible to maximise the information available from our dataset, with the typical maximum prey weight arguably more accurately capturing the diversity of tuna diets exhibited across the study region, rather than reflecting only diets that lead to the heaviest full stomachs, which may be less representative of all possible diets.

We note also that our fullness metric is derived from stomach-content weight, a metric we believe was well-suited to estimating the typical maximum stomach capacity at length in these tropical tunas. That said, volumetric measures of stomach fullness have also been explored in other species (Knight and Margraf 1982; Phelps et al. 2007) and applying such a measure to tropical tunas could offer an informative proxy of absolute maximum stomach capacity at length in future studies.

Finally, we tested the utility of a binary fullness metric, a less subjective version of the fullness coefficient containing only two categories, and found that models using this simpler metric as a response variable were the most informative for understanding relationships between tuna stomach fullness and our chosen covariates. We discuss the outcomes of our modelling in further detail in the next section.

Model performance

The first series of models we fitted with the fullness coefficient (categorical response variable) and the fullness metric (continuous response variable) (i.e. M1–M4) were largely uninformative because they failed to converge, violated model assumptions, or demonstrated poor predictive performance. We cannot base any interpretation on these models other than to say that this fullness metric is highly variable across individuals. These results may be due to missing covariates that are predictive of stomach fullness, or a lack of sensitivity in the measures we are using to describe stomach fullness. It is also possible that the data we have for each species may not be representative of the true range of stomach-fullness values existing in nature, our response variable distributions therefore not fully capturing fullness values existing in the overall population. One way this might arise is if our sampling was somehow biased, for example, through space or time interacting with fishery selectivity. Although we cannot rule out this possibility, we suggest that the presence of any sampling bias in our data is likely to be minimal, given the broad spatial and temporal coverage of our sampling and the sample sizes available (see Fig. 1).

The binary models (empty or non-empty stomachs) using logistic regression (i.e. M5 and M6) generally had good model performance to predict non-empty stomachs (i.e. predicted probability of non-empty stomachs was higher for skipjack and yellowfin tuna, less so for bigeye tuna) but the model performance was variable for empty stomachs (i.e. predicted probability of empty stomachs was variable, regardless of species).

Our analysis allowed us to assess the influence of our covariates on possible feeding success of tuna caught by commercial fisheries. With reference to identifying empty stomachs, this is straightforward and unambiguous in the laboratory, and through our data (Fig. 3) and models (Fig. 5) we can clearly see at which levels of our covariates foraging is least effective or not occurring. Non-empty stomachs may correspond to a period of feeding or digestion; however, the degree to which the ‘non-empty’ stomachs are ‘non-empty’ (i.e. how full they are) is very uncertain and could not be reliably modelled.

Stomach fullness in relation to gear, school association and time of day

As in several previous studies, the results from the binary models confirmed the fact that fisheries-related covariates can predict observations of tuna stomach fullness.

As expected, fishing gear was a relevant predictor; all three tuna species were more likely to have empty stomachs if caught on active gear (i.e. purse seine) and non-empty stomachs if caught on passive gear (i.e. longline, pole-and-line or danglers-handline). However, it is important to again highlight that a non-empty stomach does not necessarily mean it is full. In work on yellowfin, bigeye and albacore (T. alalunga) from the South Pacific, Bard (2001) observed that individuals captured on longline gear typically exhibited empty stomachs or very low stomach fullness. They hypothesised that only tunas with low stomach-fullness levels would attack longline baits, and that such behaviour might be linked to a lack of more attractive, naturally occurring prey such as that available through the daily cycle of vertically migrating, mesopelagic micronekton. Because both pole-and-line and longline gear use baited fishing methods, the chance of catching a fish depends at least partly on the feeding motivation of the fish, i.e. how hungry they are, and if they are actively foraging. Under such circumstances, we would expect a priori that stomach fullness at capture would be generally low, which is in agreement with the observations of Bard (2001), but not necessarily empty, as supported by the prevalence of non-empty stomachs seen in our dataset. For active gear such as purse seine, catchability is less likely to be influenced by satiety and feeding behaviour, and more likely linked to other behaviours or factors related to the fishing strategy. Indeed, Moreno et al. (2007) explained that in the WCPO, purse seiners generally set up at the FADs just before dawn, because targeted tuna can more easily escape during the day with clearer water and a more favourable thermocline depth. As feeding is not a requirement to be caught by purse seine, we might expect tunas captured on this gear to exhibit a wider range of stomach fullness, with no clear trends towards empty or non-empty stomachs. Yet, our findings of a consistently higher probability of empty stomachs in purse-seine catches across all three species suggests that the fishing strategy, linked to the time of day and the use of floating objects, is also influential on our observations.

Our results suggested that skipjack and bigeye tuna were more likely to have non-empty stomachs when they were not associated with floating objects. For skipjack, this observation, which was also made in the equatorial Atlantic Ocean, supports the suggestion that FADs can act as a refuge or a meeting point, and that this species tends to gather in free schools to feed (Ménard et al. 2000). Again, from our results we cannot discern how full a non-empty stomach is in response to these covariates. Because the discrimination between empty and non-empty stomach was not significantly different for yellowfin tuna either associated or not associated to a floating object, it appears that foraging in this species is not strongly influenced by the school-association effect. However, the poor fit of our chosen models for the more detailed categorical fullness coefficient and continuous fullness metric clearly demonstrated that other processes are likely to be driving changes in these finer-scale variations in tuna feeding success. At the majority of the size classes examined here, these tuna species aggregate into groups in which schooling behaviour dominates (Robert et al. 2013; Scutt Phillips et al. 2019). In our results, we did not know the size of the schools (i.e. the number of individuals in the school) nor the size composition of the schools (i.e. schools composed of different species with fish of different size range). Dagorn et al. (1994) showed that as fish size increases, school size decreases and, for yellowfin tuna, this may be linked to a change in foraging. We expected a priori that variation in foraging behaviour among species and through ontogeny may help explain variation in tuna stomach fullness. Graham et al. (2007) and Weng et al. (2015) showed an ontogenetic shift in diet with size for yellowfin caught around coastal, anchored FADs through analysis of nitrogen stable isotopes, which they attributed to changes in foraging niche over size. However, the absence of a clear relationship between our fullness metric and fish length for each of the three tuna species (Fig. S4) suggests that any ontogenetic change in diet composition, if present, is not reflected in ontogenetic changes in the measures we use to gauge stomach fullness, at least not across the length range of the fish examined in our dataset. It should also be noted that while we are comparing associated schools with not-associated schools, we do not take into account the type of associated school (e.g. anchored FADs v. drifting FADs) and the location of floating objects (i.e. the environmental factors). Indeed, Jaquemet et al. (2011) showed a difference in the number of empty stomachs for tuna caught around drifting FADs in food-rich and food-limited areas of the western Indian Ocean. FADs drifting through areas where prey resources are scarce may affect feeding success of resident schools and this hypothesis may also influence the variability of our results.

Our modelling results indicated that tuna caught early in the morning (on gear other than longline) tended to have empty stomachs (see Fig. 5), suggesting several hypotheses on their behaviour. Fish caught at this time may not have fed during the night, potentially because they were not actively looking for food or because they were not successful in catching prey. Indeed, for individuals captured on passive gear, which inherently involves them choosing to take a bait, our results suggest that such active foraging behaviour may often just be commencing at dawn. However, despite their empty stomachs, we cannot assert with certainty that the tunas did not feed during the night. Information on prey-specific digestion times is lacking, and because tuna feed on a wide variety of prey types and sizes, it is difficult to know whether they may have fed on small or quickly digested prey species during the night, leaving an empty stomach at dawn. Data from electronic tagging experiments of Schaefer and Fuller (2005, 2007) have shown a modification in the vertical distribution of bigeye and skipjack, which tend to exhibit surface-oriented behaviours at night when associated with floating objects, suggesting that this may be linked to feeding on organisms of the deep scattering layers coming to the surface at night, including zooplankton and micronekton. Unfortunately, our dataset contained few observations from tunas captured at night that could help support these suggestions. Similarly, our observations for yellowfin showed that a majority of the empty stomachs come from fish caught in the morning. To obtain a more complete picture of tropical tuna diet, especially in terms of prey composition, it would be useful to obtain additional samples from the afternoon and the evening. While helping clarify patterns in stomach fullness through the full diel cycle, such samples would also allow us to gauge the influence of opportunistic sampling or fisheries-dependant bias on our results. Longline records may prove useful in this regard; however, uncertainty around the time of capture during the long soak times observed in our longline sets meant that we could not include these data in models incorporating the ‘Time of day’ covariate in this study.

As the PMSB archive of tuna stomachs from across the WCPO continues to grow, one priority for future modelling work is the incorporation of a fishing event variable into our stomach-fullness models. Its inclusion would allow us to gauge the degree of variability in fullness attributable to within- and between-fishing events, and to better quantify the influence (or not) of school cohesion on feeding behaviour around drifting FADs. Furthermore, integrating studies such as ours with those using telemetry, biotic oceanography or muscle stable isotope analyses may provide deeper insights into the factors driving tuna feeding success (Kirby 2001) and thus improve the utility of stomach-fullness measures as a valuable ecosystem indicator (Allain et al. 2021; Oceanic Fisheries Programme, Pacific Community 2023).

Several studies have examined diet and prey diversity in response to environment, fish length, and similar covariates as those we examined here (e.g. Bertrand et al. 2002; Fuller et al. 2021; Jiang et al. 2022), showing significant differences in diet among regions, species and size. However, our aim in this study was to more fundamentally examine measures of tuna stomach fullness and examine how the apparent short-term feeding success of individuals may be predicted by simple covariates that have been hypothesised to influence observations of stomach contents.

Future directions of the study

This study has highlighted how metrics of stomach fullness may allow us to answer questions about the ecosystem dynamics and the influence of fisheries and advise us on sampling methods for tuna-diet studies. We examined only a small number of explanatory variables in our study and further work is needed to identify other important covariates to better understand the mechanisms that underpin stomach fullness in tunas. Specifically, progress may lie in incorporating estimates of the habitat richness (e.g. primary production), availability of prey (e.g. zooplankton/micronekton estimates from models, Receveur et al. 2022), digestion rate of individual prey and prey composition.

At present, our results suggest that examining tuna stomach fullness as proportional binary data is most appropriate, but with ongoing sample collection, the continued development of fullness metrics, and incorporation of other data such a nitrogen isotope or lipid content measurements, it may be possible to more accurately characterise the true range of feeding success present in tuna populations.