Fig 1.
Organization of a seabird bycatch model and the associated data sources.
The components of the integrated bycatch model include bycatch loss (A), the origin and the condition of the bycatch (B) and a traditional loss-free bycatch estimation and prediction model (C). Three data sources, Brothers et al. [7], Gilman et al. [17] and Atlantic POP (grey rectangles with solid borderlines), were employed to inform important parameters of the bycatch estimation model; U.S. Atlantic PLL logbook data were used to project the fleet level total bycatch. In particular, Brothers et al. [7] informed the estimation of bycatch loss rate (Ploss) with details in Fig 2; Gilman et al. [17] and POP data together informed the estimation of the survival rate of haul-captures (Plive) and the probability of set-captures (Pset) with details in Fig 3.
Fig 2.
Classification of seabird interactions and subsequent carcass retrieval.
At the line setting stage (A), bait-taking interactions of seabirds are classified into three broad categories, i.e., observed caught, possibly caught and observed not caught (clear circles on the left). The true state of the interaction, i.e., either hooked or not hooked (gray circles in the middle), is not directly observable to both the observers at the line setting and hauling stages. On the arrows running from a true state to a classified type are the estimated classification probabilities. For example, an interaction that does not lead to a hooking event (top middle circle) may be classified as observed not caught (lower left circle) with a 90.8% probability. Seabirds hooked during the line setting stage could drop off the hook with probability (ploss). Additional seabirds may get caught at the line hauling stage (newly hooked in B).
Fig 3.
Two-step processes for both the bycatch loss (A) and mortality (B). The probability for each stage is in italics, and the probabilities of final outcomes can be found in parentheses under each outcome.
Table 1.
Counts of bait-taking interaction categories and the associated counts of retrieved carcasses from Brothers et al. [7].
Table 2.
Estimated posterior mean and 95% credible interval of the classification probabilities (in %) of seabird interactions with fisheries.
Fig 4.
Prior (dashed line) and posterior (solid line) distributions of the bycatch loss rate (ploss).
The prior is a uniform distribution from 0 to 1. Count data extracted from Brothers et al. [7] were used to estimate ploss.
Fig 5.
Probability of set-capture (pset) in the U.S. Atlantic PLL fleet for different bycatch loss scenarios.
The low, medium and high bycatch loss scenarios use, respectively, point estimates of bycatch loss rate (ploss) at the 2.5th, 50th and 97.5th percentiles of the posterior distribution of the bycatch loss rate based on observations from Brothers et al. [7], and the expected bycatch loss scenario uses the entire posterior distribution of ploss in the calculation.
Table 3.
Model selection summary for the low, medium and high bycatch loss scenarios.
Fig 6.
Estimated total seabird bycatch from 1992 to 2016 by the U.S. Atlantic PLL fleet.
Estimates of total seabird bycatch were generated for four bycatch loss scenarios and a loss-free model.
Table 4.
Mean, standard deviation (std.) and coefficient of variation (CV) of projected fleet level total (1992–2016) seabird bycatch from different bycatch loss scenarios in the Western North Atlantic.
Fig 7.
Relative differences in annual bycatch between different bycatch loss scenarios and the loss-free model.