Robbing Peter to pay Paul: replacing unintended cross-taxa conflicts with intentional tradeoffs by moving from piecemeal to integrated fisheries bycatch management

Bycatch in fisheries can have profound effects on the abundance of species with relatively low resilience to increased mortality, can alter the evolutionary characteristics and concomitant fitness of affected populations through heritable trait-based selective removals, and can alter ecosystem functions, structure and services through food web trophic links. We challenge current piecemeal bycatch management paradigms, which reduce the mortality of one taxon of conservation concern at the unintended expense of others. Bycatch mitigation measures may also reduce intraspecific genetic diversity. We drew examples of broadly prescribed ‘best practice’ methods to mitigate bycatch that result in unintended cross-taxa conflicts from pelagic longline, tuna purse seine, gillnet and trawl fisheries. We identified priority improvements in data quality and in understanding ecological effects of bycatch fishing mortality to support holistic ecological risk assessments of the effects of bycatch removals conducted through semi-quantitative and model-based approaches. A transition to integrated bycatch assessment and management that comprehensively consider biodiversity across its hierarchical manifestations is needed, where relative risks and conflicts from alternative bycatch management measures are evaluated and accounted for in fisheries decision-making processes. This would enable managers to select measures with intentional and acceptable tradeoffs to best meet objectives, when conflicts are unavoidable.


Introduction
Fisheries that target relatively fecund species can have large impacts on incidentally caught species that, due to their lower reproduction rates and other 'slow' life history traits, are relatively vulnerable to increased mortality, including seabirds, sea turtles, sea snakes, marine mammals, elasmobranchs and some teleosts (Goni 1998;Hall et al. 2000;Branch et al. 2013;Gray and Kennelly 2018). Affected populations can decline over short temporal scales (decades and shorter) and are slow to recover once depleted (Musick 1999;Hall et al. 2000;Stevens et al. 2000;Kitchell et al. 2002;Dulvy et al. 2008). Intraspecific selective mortality can cause fisheries-induced evolution (FIE). This occurs when fisheries selectively remove individuals based on highly heritable traits that vary within affected populations (Heino et al. 2015;Tuck et al. 2015;Audzijonyte et al. 2016;Lennox et al. 2017;Hollins et al. 2018). Selective mortality on heritable traits reduces the range of phenotypes for these traits within the exploited populations. This, in turn, can reduce population fitness, resistance and resilience to natural pressures, and reduce the ability of populations to evolve in response to changes in environmental conditions (Saccheri et al. 1998;Westemeier et al. 1998;Reed and Frankham 2003;Jorgensen et al. 2007;Evans and Sheldon 2008).
Objectives of managing fisheries bycatch include avoiding loss of unique alleles and genetic diversity, preventing protracted or irreparable harm, or permanent loss, of affected populations per se, and sustaining a desired state of ecosystem balance that produces targeted services, including fishery yields (FAO 1995(FAO , 2003(FAO , 2011. These objectives are interlinked as fisheries bycatch removals and alterations to the evolutionary characteristics of affected populations can alter ecosystem functions and structure through food web trophic links (Pace et al. 1999;Stevens et al. 2000;Ward and Myers 2005;Ferretti et al. 2010). In some cases, use of a bycatch mitigation method that achieves large reductions in catch rates and at-vessel mortality of one species or group of conservation concern is at the expense of another (e.g., Beverly et al. 2009;Salerno et al. 2010;Hall 1998Gilman et al. 2016b). When bycatch is managed in a piecemeal, species-by-species fashion, this can result in cross-taxa conflicts that are unplanned, with unintended consequences. Instead, holistically evaluating and identifying conflicts as well as mutual benefits amongst species groups from implementing bycatch management options supports carefully considered, intentional tradeoffs when cross-taxa conflicts cannot be avoided. The integrated bycatch management measures adopted by decision-makers would cause known and acceptable cross-taxa tradeoffs that best meet objectives.
Fisheries bycatch is currently largely managed in a piecemeal manner, both at broad regional levels and at the individual fishery level (Hall 1998Gilman et al. 2014aGilman et al. , b, 2016b. Global guidance on bycatch management, including from the Food and Agriculture Organization of the United Nations, is single-taxa specific (seabirds, FAO 1999a;sharks, FAO 1999b;sea turtles, FAO 2010). Some 'best practices' prescribed to mitigate bycatch recommended by environmental non-governmental organizations (NGOs) and their seafood corporate partners (e.g., pelagic longline circle hooks to benefit sea turtles, Wegmans 2016) likewise are single-taxa specific. Regional fisheries management organizations employ taxa-specific bycatch conservation measures, e.g., a stand-alone measure to manage fisheries bycatch of sea turtles, a separate measure for seabirds, etc. (Gilman et al. 2014b;ISSF 2017a). Domestic fisheries management systems employ similar taxa-specific bycatch mitigation measures (e.g., standalone U.S. longline regulations on sea turtles, NMFS 2001NMFS , 2015seabirds, NMFS 2005;odontocetes, NMFS 2012;sharks, NMFS 2014sharks, NMFS , 2016. These taxa-specific approaches to managing bycatch may not be integrated within an overarching holistic framework for managing bycatch.

123
As a result, because they do not account for conflicts between groups of conservation concern that result from their implementation, bycatch mitigation measures designed to address problems within one species group can exacerbate bycatch of other vulnerable taxa in some fisheries (Gilman et al. 2014a, b). Frameworks for bycatch risk assessment and management likewise largely do not account for effects at other levels of biodiversity, including effects of bycatch removals on genetic diversity within affected populations and broad changes in ecosystem functions and structure.
There are, however, numerous examples of bycatch mitigation methods that do not create conflicts and in some cases result in shared benefits to multiple bycatch groups. For example, non-entangling and biodegradable designs of fish aggregating devices (FADs) used in tuna purse seine fisheries reduce the entanglement of sharks, sea turtles and other organisms (ISSF 2015; Moreno et al. 2016). The backdown maneuver to release dolphins in tuna purse seine fisheries does not create conflicts . Various seabird bycatch mitigation methods that prevent seabird access to pelagic longline baited hooks during setting, such as side-setting, bird-scaring tori lines, and underwater setting devices, are unlikely to cause cross-taxa conflicts (Gilman et al. 2005;Clarke et al. 2014). Various gillnet bycatch reduction approaches likely pose low probability of causing cross-taxa conflicts. This includes increasing net tension (stiffness) to reduce the likelihood of entangling sea turtles and other large organisms, such as by: reducing the net profile, increasing filament diameter, using stiffer weaves, using larger floats on the top rope and heavier weights on the bottom rope, and infusing certain compounds (Werner et al. 2006;Larsen et al. 2007; Thorpe and Frierson 2009). Eliminating or increasing the length of tiedowns used in anchored gillnet eliminates a bag of slack webbing that entangles organisms such as sea turtles (Price and Van Salisbury 2007;Gilman et al. 2010). Making portions of gillnets more visible by using certain net colors, using thicker twine diameter, attaching corks or other materials within the net, and illuminating nets with chemical or battery-operated lightsticks can reduce catch rates of sea turtles, seabirds and marine mammals (Hayase and Yatsu 1993;Melvin et al. 2001a; Barlow and Cameron 2003;Werner et al. 2006;Wang et al. 2010Wang et al. , 2013Mangel et al. 2014). As another example, Turtle Excluder Devices, sorting and shepherding devices used in shrimp and prawn trawl fisheries, can exclude sea turtles as well as elasmobranchs and other large fishes, while largely retaining smaller organisms, such as targeted shrimp, that fit through the grid, without increasing catch of other taxa of conservation concern (Broadhurst 2000; FAO 2010; Willems et al. 2016). Spatial and temporal restrictions on fishing at predictable bycatch hotspots in some cases do not create cross-taxa conflicts (however, see the section ''Temporal and spatial fishery closures''). This includes both marine protected areas (MPAs) that protect static (e.g., shallow submerged features such as seamounts) (Morato et al. 2008) and dynamic sites (gyres, fronts) (e.g., dynamic spatial management of southern bluefin tuna (Thunnus maccoyii) bycatch, Hobday et al. 2010; real-time fleet communication to enable vessels to avoid fishing grounds with high bycatch, Gilman et al. 2006b;Little et al. 2015).
Drawing examples from pelagic longline, tuna purse seine, gillnet and trawl fisheries, we demonstrate how bycatch mitigation methods that effectively mitigate catch and mortality of one species or group of conservation concern can exacerbate catch and mortality of another of potentially greater conservation concern. The bycatch mitigation methods in the examples use gear technology, which involves changes in fishing methods and gear, and temporal and spatial fishery closures (for information on the full range of bycatch mitigation approaches, see : Hall 1996;Hall et al. 2000;FAO 2011;Gilman 2011). We also present theoretical examples of how employment of bycatch mitigation methods could inadvertently exacerbate FIE, reducing genetic diversity within affected populations. We describe examples of qualitative to model-based, quantitative methods for integrated ecological risk assessments (ERAs) of the effects of fishing which could be used to holistically assess the relative risks from fishing across affected taxonomic groups and across manifestations of biodiversity, from genetic diversity to the state of ecosystem-level components. Priority improvements in data quality and in understanding ecological effects of bycatch fishing mortality to support holistic semiquantitative and model-based ecological risk assessments of the effects of bycatch removals are identified. Findings from integrated risk assessments are needed to design bycatch management frameworks and define best-practice bycatch mitigation methods that account for cross-taxa conflicts and result in intentional and acceptable tradeoffs. These examples highlight the need to move beyond piecemeal, species-by-species approaches to holistic bycatch assessment and management. This will enable the relative risks to affected components of biodiversity, including genetic diversity, impacts on populations and species, and broader ecosystem-level effects resulting from alternative fisheries bycatch mitigation methods to be identified and accounted for in fisheries decision-making processes.

Gear technology
Pelagic longline hook shape Several environmental NGOs and retailers have adopted sustainable sourcing policies that call for the use of circle hooks instead of J-shaped hooks by pelagic longline fisheries as a solution to sea turtle bycatch (Monterey Bay Aquarium 2015; ALDI 2016; Wegmans 2016; WWF 2016). This taxa-specific recommendation could inadvertently exacerbate bycatch rates of other taxa of conservation concern. Four predominant hook shapes are used in global pelagic longline fisheries: circle, J, tuna and teracima ( Fig. 1) (Beverly 2009;IATTC 2011). The point of J-hooks is positioned parallel to the hook shank. Tuna and teracima hooks have a slightly curved shank, and like J-hooks, the shank does not protect the point, and as a result, are categorized as 'J-shaped' hooks (see Fig. 2 for fishing hook anatomy) (Beverly and Chapman 2007;Serafy et al. 2009). Circular shaped 'circle' hooks have points that are turned perpendicularly back toward the shank. This makes the point less exposed relative to J-shaped hooks.
When ingested, J-shaped hooks tend to hook deeply, in the esophagus and gut. Circle hooks with little or no offset tend to catch in the corner of the mouth (Cooke and Suski 2004;Curran and Beverly 2012;Epperly et al. 2012;Clarke et al. 2014;Parga et al. 2015). Hooks can be 'straight', where the point is in the same plane as the shank, or 'offset', with the point bent away from the plane of the shank. There is likely a threshold offset angle above which a circle hook's gape would be sufficiently large to cause it to hook similarly to J-shaped hooks Gilman and Hall 2015). Having a less exposed point, circle hooks also have a lower probability of foul-hooking (i.e., hooking externally in the body) than J-shaped hooks. Organisms that are foul-hooked or hooked in the mouth have higher at-vessel survival rates and possibly higher probability of pre-catch and post-release survival than those that are deeply hooked in the esophagus and gut (Borucinska et al. 2002;Chaloupka et al. 2004;Cooke and Suski 2004;Casale et al. 2008;Pacheco et al. 2011;Swimmer and Gilman 2012). Additionally, circle hooks' predominant hooking position in the mouth facilitates relatively easy access to remove all terminal tackle (hook, leader, weight, trailing branchline), potentially increasing post-release survival rates (Chaloupka et al. 2004;Cooke and Suski 2004;Godin et al. 2012;Serafy et al. 2012a;Swimmer and Gilman 2012;Parga et al. 2015).
Leatherback sea turtles (Dermochelys coriacea), the only extant sea turtle species to not have a hard shell, are most frequently caught in pelagic longline gear by becoming foul-hooked and entangled. They have lower catch rates on circle than J-shaped hooks (Watson et al. 2005;Sales et al. 2010;Pacheco et al. 2011;Foster et al. 2012;Coelho et al. 2015;Santos et al. 2012Santos et al. , 2013Gilman and Huang 2017). For hardshelled and leatherback turtles that ingest a hook, circle hooks result in a lower proportion of turtles swallowing the hook deeply relative to J-shaped hooks (Andraka et al. 2013;Gilman and Huang 2017). Thus, relative to J-shaped hooks of the same minimum width, circle hooks benefit sea turtles (both hardshelled species and leatherbacks) because a smaller proportion of those captured will be deeply hooked, and because they reduce leatherback sea turtle catch rates. However, use of circle hooks in place of similar sized J-shaped hooks will provide little benefit to hardshelled turtles in deep-set longline fisheries where most turtles drown during the gear soak and are dead upon gear haulback (e.g., Gilman et al. 2006c;Clarke et al. 2014;Hall et al. 2017).
While hook shape has the same effect on shark anatomical hooking position and survival rates as for other taxonomic groups, some pelagic sharks have a higher probability of capture on circle versus J-shaped hooks (Gilman et al. 2016b;Reinhardt et al. 2017). Based on results of a meta-analysis from 30 studies, pooled shark species had a 1.2 times significantly higher risk of capture on circle hooks than on J and tuna hooks. Blue shark (Prionace glauca), the main shark species captured in global longline fisheries (Gilman 2011;Clarke et al. 2014;Oliver et al. 2015), was the predominant caught shark species in 17 of the 30 compiled studies in the meta-analysis (Gilman et al. 2016b). Consistent with the findings of Gilman et al. (2016b), a meta-analysis by Reinhardt et al. (2017) found significantly higher catch risk for six individual species of sharks, and no shark species had significantly lower relative risk of capture on circle hooks relative to J-shaped hooks. Due to the larger sample size plus the number of independent studies, correctly designed meta-analyses can provide estimates with increased precision and accuracy over estimates from single studies, with increased statistical power to detect a real effect (e.g. Borenstein et al. 2009;Musyl et al. 2011Musyl et al. , 2015. Most of the studies compiled for the meta-analyses by Gilman et al. (2016b) and Reinhardt et al. (2017) employed designs with simultaneous variability in multiple potentially significant explanatory factors, including hook minimum width, such that the analyses were in effect measuring pooled relative catchability between wider circle and narrower J-shaped hooks. However, the small number of studies designed to enable an assessment of the single factor effect of hook minimum width have found higher probability of shark capture on circle hooks relative to J-shaped hooks of the same minimum width (Ward et al. 2009;Serafy et al. 2012b;Andraka et al. 2013;Caneco et al. 2014;Gilman et al. 2016b). While circle hooks increase shark catchability, they also reduce at-vessel mortality and the proportion of the catch that are deeply hooked relative to J-shaped hooks (Gilman et al. 2016b;Reinhardt et al. 2017).
Observations of lower shark probability of capture on J-shaped hooks relative to circle hooks may be partially due to a synergistic effect between hook shape and leader material, when monofilament leaders are used. When caught on circle hooks, sharks are most often hooked in the mouth and jaw, making it less likely that they can sever (bite through or abrade) a monofilament leader than when deeply hooked on a J-shaped hook (Afonso et al. 2012;Gilman et al. 2016b). Several studies have observed significantly lower shark catch rates on circle versus J-shaped hooks when wire leaders were used (Mejuto et al. 2008;Sales et al. 2010;Andraka et al. 2013), indicating that, when wire leaders are used, the single factor hook shape alone causes the observed difference in shark catch rates (Gilman et al. 2016b). The effect of hook shape on shark catch risk may be due in part to the way sharks consume their prey. Sharks chomp their prey (and baited hooks) repeatedly before swallowing, unlike teleosts who suck in and swallow their prey. While being munched by sharks, due to their shape, circle hooks roll and slide more so than J-shaped hooks, perhaps creating a higher probability that circle hooks will become oriented so that it lodges.
The relative magnitude of the effect of the single factor hook shape on catch rates is likely substantially larger for leatherback sea turtles (ca. 63% lower catch rate on circle vs. J-shaped hook when bait type was not variable, Watson et al. 2005;Sales et al. 2010;Foster et al. 2012, Coelho et al. 2015Santos et al. 2012Santos et al. , 2013 than for sharks (ca. 20% higher relative risk of capture on circle vs. J-shaped hook, Gilman et al. 2016b). While the benefit to leatherback sea turtles of using circle hooks relative to J-shaped hooks exceeds the cost to sharks in terms of catch rate effect, the effect on catch levels, fishing mortality levels and relative risks to affected populations must also be considered and requires fishery specific assessment.
Use of circle hooks to benefit sea turtles may result in conflicting effects with additional species of conservation concern. Reinhardt et al. (2017), pooling data from two records, found that Atlantic bluefin tuna (Thunnus thynnus) had a significantly higher relative risk of capture on circle hooks than J-hooks, as is the case for most principal market tunas (Gilman and Hall 2015). Overfishing is not occurring for either of the two Atlantic ocean stocks, but the status of biomass is unknown due to extensive uncertainty with historical catch data (ICCAT 2017). Unlike other principal market tuna species which have extensive spawning grounds and protracted spawning seasons, bluefin species spawn in limited areas and have relatively short spawning periods of 1-2 months, making them particularly vulnerable to stressors, including from fishing mortality and outcomes of climate change (Collette et al. 2011;Muhling et al. 2011). Effects of pelagic longline hook shape on other taxa of conservation concern are less certain (Gilman and Hall 2015). Circle hooks may reduce odontocete (toothed whale) catch rates relative to J-shaped hooks of a similar size (Gilman et al. 2006a;Forney et al. 2011). Circle hooks also result in a higher probability of odontocetes being able to escape by throwing the hook or disentangling from line, resulting in less serious injuries (Forney et al. 2011). Circle hooks increase pelagic stingray catch rates relative to rates on J-shaped hooks of a similar size (Andraka et al. 2013;Gilman et al. 2016b).
While not included as a case study here, it is worth briefly mentioning that pelagic longline bait type poses similar conflicts as hook shape: when using small species of fish (e.g., pelagic 'forage' fishes, such as mackerels and species with mackerel-like characteristics) for bait, the catch and deep hooking rates are higher for some shark species but lower for primarily hard-shelled species of sea turtles relative to when squid is used for bait (Gilman et al. 2016b;Gilman and Huang 2017). Fish bait also has higher catch rates of some seabird species relative to using squid (Li et al. 2012).
Pelagic longline time-of-day and gear depth Measures prescribing gear soak depth and time-of-day of fishing operations, such as deeper pelagic longline sets to reduce epipelagic sea turtle bycatch (e.g., NMFS 2004;FAO 2010), and longline night setting to reduce seabird bycatch (e.g., Monterey Bay Aquarium 2015; ACAP 2014, 2016), inadvertently exacerbates bycatch rates of other taxa of conservation concern in some fisheries. Gear soak depth and time-of-day of fishing operations significantly explain species-specific pelagic longline catch rates Young et al. 2010;Musyl et al. 2003Musyl et al. , 2011Gilman and Hall 2015;Gilman et al. 2017a). There is a substantial body of literature from electronic tagging and longline catch studies that indicate pelagic fishes and turtles partition themselves vertically (by depth and temperature) as well as horizontally (geospatially) by prey availability and primary environmental variables of hydrostatic pressure, temperature, and dissolved oxygen Bernal et al. 2010;Musyl et al. 2003Musyl et al. , 2011.
There are synergistic effects of soak depth and the time-of-day of fishing activities on catch rates. Pelagic species' vertical habitat preferences can vary temporally, within a day, due to diel vertical migration cycles, time of day of active foraging, and temporal variability in diving behavior (Sedberry and Loefer 2001;Nakano et al. 1997Nakano et al. , 2003Ward and Myers 2005;Beverly et al. 2009;Musyl et al. 2003Musyl et al. , 2011. Some species of pelagic apex predatory fishes conduct diel vertical migration cycles, mirroring movements of prey organisms comprising the deep scattering layer 123 (Carey et al. 1990;Schaefer and Fuller 2002;Weng and Block 2004;Young et al. 2010;Musyl et al. 2003Musyl et al. , 2011. For example, blue sharks make frequent deep dives during the daytime and forage primarily near the sea surface at night (Carey et al. 1990;Musyl et al. 2011), making them susceptible to capture primarily in shallow-set longline gear, but also in deep daytime sets. Thresher sharks infrequently come near the surface and thus are largely susceptible to capture only in deep-set longline gear (Carey et al. 1990;Nakano et al. 2003;Weng and Block 2004;Musyl et al. 2011). Sea turtles also exhibit diel vertical migration patterns where transitions between daytime and nighttime are not as pronounced. Turtles conduct relatively deeper dives during crepuscular hours to reach prey migrating vertically through the water column, but during the daytime, their prey might be too deep or not worth the energy investment for sea turtles to dive to reach (Hays et al. 2006). Depending on the vertical diel distribution pattern of a particular species, the time of day of gear setting and haulback can also significantly affect catch rates in deep set longline fisheries because the terminal tackle passes through a relatively large section of the water column for an extended time period (Kerstetter and Graves 2006).
These interacting effects of gear soak depth and time-of-day of fishing operations are illustrated in Fig. 3 for selected pelagic teleost and shark species. Based on pop-up satellite archival tagging (PSAT) data, Musyl et al. (2011) estimated that fishing longline hooks deeper than ca. 100 m during the day resulted in higher catch of target species of bigeye tuna and swordfish while avoiding epipelagic species of sharks relative to fishing shallow at night (Fig. 3). This fishing strategy would avoid nearly 98% and 86% of oceanic whitetip and silky shark interactions, respectively, would reduce blue shark encounters by an estimated 53% and shortfin mako by 46%. This would, however, increase the catch rate of deeper-habitat mesopelagic II species, including bigeye thresher sharks (Fig. 3), which have long generation times and slow growth relative to other elasmobranchs (Dulvy et al. 2008).
Consistent with findings of Musyl et al. (2011), Table 1 summarizes the effects of various combinations of soak depth and time-of-day of gear soak on catch rates of the Palau tropical western and central Pacific Ocean pelagic longline fishery, based on analyses of observer and logbook program data (Gilman et al. 2017a). To summarize, in general, daytime deep-set longline gear, with all hooks soaking [ 100 m, can achieve economically viable catch rates of swordfish and some target tunas, and have substantially lower catch rates of olive ridley and other hard shelled sea turtle species (Polovina et al. 2003(Polovina et al. , 2004Rice and Balazs 2008;Hall et al. 2012;Cambiè et al. 2013;Watson and Bigelow 2014), epipelagic sharks (Galeana-Villasenor et al. 2008;Musyl et al. 2011;Bromhead et al. 2012;Caneco et al. 2014;Watson and Bigelow 2014), and istiophorid billfishes Musyl et al. 2011;Watson and Bigelow 2014) relative to shallow-set gear, where the majority or all hooks soak at depths \ 100 m. However, deeper-set daytime gear has higher catch rates of some mesopelagic II species such as thresher sharks Musyl et al. 2011).
Variability in time-of-day of foraging behavior observed in pelagic fishes and sea turtles (e.g., Bigelow et al. 2002;Ward et al. 2004;Musyl et al. 2003Musyl et al. , 2011 also occurs for pelagic seabirds. In some regions, night setting, prescribed in some fisheries to protect albatrosses and other primarily diurnal foraging seabird species has led to higher bycatch of nocturnal foragers (e.g. northern fulmars Fulmarus glacialis, Melvin et al. 2001b).
There can also be variability in vertical habitat distributions by age class, body size and sex within species. As a result, the time-of-day and depth of fishing affects the length frequency distribution, mean length and sex ratios of the catch of individual species (Musyl et al. 2003Coelho et al. 2012;Gilman et al. 2017a). Sex-skewed catches can in turn affect mean lengths for species with sexual size dimorphism (differences in size by sex of the same age class, i.e., differential growth by sex).
Gear soak depth also significantly affects at-vessel survival rates. Because shallower, warmer water depths generally have lower dissolved oxygen concentration, higher stress occurs for fish caught at shallower depths due to the combination of temperature, limited oxygen and higher metabolic rates (Moyes et al. 2006;Gallagher et al. 2014). However, turtles and marine mammals caught on deeper hooks are more likely to drown before gear retrieval if they cannot reach the surface to breathe during the gear soak, and are more likely to experience decompression sickness (the bends) if they survive the gear soak (Gilman et al. 2006c;Moore et al. 2009;García-Párraga et al. 2014).
Vertical habitat preferences also vary geospatially and temporally for individual species, and within species by age class, body size, and sometimes by sex. Therefore, the effect of pelagic longline soak depth on catch rates will vary spatially and temporally. The effect of soak depth on pelagic longline catch rates will vary both at relatively small spatial scales, such as whether gear is set near a shallow submerged feature or in the open ocean, and over very broad scales (100 s to 1000 s of km) due to changes in pelagic species' distributions, such as in response to large scale climate variability (e.g., inter-annual El Niño-Southern Oscillation and decadal Pacific Decadal Oscillation phases) and responses to outcomes of climate change (Gilman et al. 2016a). The effect of fishing depth will also vary due to regional and seasonal differences in variables that determine habitat preferences and migration corridors, including variability in the mixed-layer depth/thermocline, oxygen/temperature gradients, and other variables that determine habitat preferences, such as prey availability (Musyl et al. 2003. Tuna purse seine set type Tuna purse seine vessels make different set types, which can have substantially different catch Fig. 3 Depth and temperature preferences for various pelagic teleost and shark species, determined from pop-up satellite archival tags. Boxes are 25% and 75% quartiles, solid horizontal line is the median (50% quartile), dashed horizontal line is the mean, and bars represent 10-90th deciles. M = male, F = female. [adapted from Musyl et al. (2011)] 123 compositions. They make 'unassociated' or free swimming school sets (hereafter called school sets), where a school of tunas and other pelagic fishes is at or near the sea surface. Fishers detect the free swimming school when the fish disturb the sea surface and through the presence of seabirds foraging from above (Hall 1998;Hall and Roman 2013). Purse seine vessels also make sets on floating objects that aggregate pelagic organisms. This includes drifting logs, drifting algae, live and dead large marine organisms, marine debris (e.g. crates, pallets, nets), as well as artificial objects that are built and deployed by fishers and are designed specifically to aggregate pelagic fishes (Castro et al. 2002;Hall and Roman 2013). This latter category of artificial floating object is referred to as FADs . Drifting FADs are typically equipped with satellite buoys, and sometimes also with echo-sounders, to track the spatial position and estimate the biomass of fish aggregated at the device (Lopez et al. 2016;Gilman et al. 2018a). FADs can also be anchored, called rumpons in Indonesia, and payaos in the Philippines, usually used by smallscale fisheries, but in some regions (e.g., Papua New Guinea) are used by industrial purse seine fisheries (Beverly et al. 2012). Fishing vessels may also be used like a drifting FAD (e.g., seiners cooperating with bait boats, and purse seine vessels can be used to draw an aggregation away from a FAD) (Hall and Roman 2013). And in the Indian and Atlantic Oceans, purse seine vessels set on anchored supply vessels positioned over shallow seamount summits, similar to an anchored FAD (Amandè et al. 2008;Koehler and Moreno 2016). Primarily in the eastern Pacific, purse seiners make sets on schools of yellowfin tuna in association with dolphin pods, referred to as dolphin sets (Hall and Roman 2013).
The most appropriate approach to combine various set types to understand effects on the species composition and sizes of the catch is not always clear. Different school set types and different associated set types can have substantially different catch compositions (Dagorn et al. 2013;Hall and Roman 2013;ISSF 2017c;Peatman et al. 2017). Despite this, here we refer to the catch composition of combined school set types and combined associated set types. This enables simpler comparisons that are consistent with higher resolution purse seine set type categorizations employed by seafood certification and ranking programs and major seafood buyers (e.g., preferential sourcing of school tuna, ALDI 2016; certification under the Marine Stewardship Council standard differentiates between school and floating object sets, Acoura Marine 2017; Seafood Watch program ranks 'floating object purse seines' as red/avoid and 'unassociated purse seine' as yellow/good alternative, Monterey Bay Aquarium 2018), and because data  (Gilman et al. 2017c).
Conflicting ecological risks to bycatch species between purse seine sets on dolphins versus drifting FADs have been recognized for over 2 decades (Hall 1998). Measures adopted by fishers and management authorities to reduce dolphin mortality levels in eastern Pacific Ocean purse seine sets on dolphin pods (including conducting a backdown maneuver, deploying crew during backdown to help dolphins escape, using a Medina dolphin safety panel to reduce dolphin entanglement and increase sinking of the corkline during backdown, and fleet-wide and vessel-based dolphin mortality limits) have effectively achieved an order of magnitude reduction in annual, direct dolphin mortality levels. The number of dolphins killed per dolphin school set declined by about 98%, and the low annual dolphin mortality levels have been effectively maintained over the past 2 decades (Hall et al. 2001;Hall and Roman 2013). In 1990, when dolphin mortality levels had already declined by about 60% from a peak in 1986, in response to advocacy efforts by Earth Island Institute, an environmental NGO, some tuna canneries adopted policies to not purchase tuna caught in trips that included dolphin sets. As a result, U.S. purse seine vessels began making drifting FAD sets (Hall 1998). There was a temporary decline in the number of annual dolphin sets, which eventually reverted to pre-1990 levels (Hall et al. 2001;Hall and Roman 2013). However, the development of the drifting FAD fishery in the eastern Pacific resulting from the adoption of dolphin-safe policies has remained in place. The temporary benefit to affected dolphin populations from the dolphin-safe policy came at a lasting cost of an increase in bycatch of sharks and sea turtles, as well as in catch and discards of undersized tunas and non-marketable teleost species in drifting FAD sets (Hall 1998;Hall and Roman 2013).
Similar to the conflicting effects on bycatch rates between purse seine sets on dolphins versus drifting FADs, purse seine school and FAD, log and other associated set types have different catch compositions (Dagorn et al. 2013;Hall and Roman 2013;ISSF 2017c;Peatman et al. 2017). There is very large regional variability in purse seine catch rates by set type and taxonomic group (Table 2). Table 2 compares catch rates of selected species and groups in purse seine associated and school sets. Two rates are reported for some records when rates were available both in units of catch per 1000 sets, which may be an appropriate rate to use in fisheries subject to input controls (i.e., limits on effort, such as the annual number of sets or number of days fishing) and per 1000 t of landed target tunas (bycatch-to-catch ratio), which may be an appropriate rate to use in fisheries subject to output controls (i.e., limits on catch, such as annual catch levels of target tuna species).
Log and drifting FAD sets catch a larger number of species and a much larger biomass of nontarget species (i.e., species other than skipjack, yellowfin and bigeye tunas) than school sets (Hall 1998;Amandè et al. 2010;Hall and Roman 2013;Torres-Irineo et al. 2014;Gaertner et al. 2016;Peatman et al. 2017;Lezama-Ochoa et al. 2017). This is in part due to the slow drifting speed of FADs and logs relative to that of free-swimming tuna schools chasing prey and of schools of tunas and dolphins escaping speedboats. The majority of nontarget catch is comprised of teleost species that are not considered to be of conservation concern (e.g., kawakawa Euthynnus affinis and other small species of tunas, ocean triggerfish Canthidermis maculata, mackerel scad Decapterus macarellus, mahi-mahi Coryphaena hippurus, rainbow runner Elagatis bipinnulata and other carangids) (Dagorn et al. 2013;Hall and Roman 2013;ISSF 2017c). In most regions, a large proportion of the nontarget teleost catch is discarded (Hall and Roman 2013;ISSF 2017c). However, some tuna RFMOs are considering requiring full retention of these species.
Not accounting for entanglements in FAD subsurface appendages (which can be addressed by using non-entangling materials; ISSF 2015), shark catch rates, expressed as the weight of caught sharks per 1000 sets, were an order of magnitude higher in sets on drifting FADs and logs than in school sets (Atlantic and Indian Oceans, Amandè et al. 2008Amandè et al. , 2010. When the shark catch rate is expressed as the weight of caught sharks per weight of retained market tunas, shark catch rates in school and associated sets are the same order of magnitude (ISSF 2017c). School sets have higher catch rates of mobulids (manta and devil rays, Mobula spp.) (Amandè et al. 2008(Amandè et al. , 2010Dagorn et al. 2013;Hall and Roman 2013), of which several species are categorized as IUCN Endangered, Vulnerable and Near Threatened (Croll et al. 2016;IUCN 2017).
123 Table 2 Catch rates of selected species and groups in purse seine tuna fishery associated and school sets. Associated sets refers to sets on both drifting FADs and logs unless specified otherwise (Williams et al. 2009;Amandè et al. 2008Amandè et al. , 2010Dagorn et al. 2013;  The number of sea turtles captured in tuna purse seine sets (ca. hundreds per year globally) is extremely small relative to other gear types (e.g., gillnet, trawl and pelagic longline, Lewison and Crowder 2007;Gilman 2011). When captured, more than 90% are released alive (Williams et al. 2009;Amandè et al. 2008Amandè et al. , 2010Gilman 2011;Dagorn et al. 2013;Hall and Roman 2013). Sea turtle mortality also occurs from entanglement in FAD appendages and surface structures but data are unavailable to estimate mortality levels from these interactions. Turtle catch rates are higher in combined (drifting FAD and log) associated sets than in school sets in the Atlantic, Eastern Pacific and Indian Oceans (Amandè et al. 2008(Amandè et al. , 2010Hall and Roman 2013). However, based on data from the western and central Pacific Ocean, turtle capture rates are higher in school sets than drifting FAD sets (Table 2) (Williams et al. 2009). Not shown in Table 2, turtle capture rates are highest in sets on large marine animals (e.g., whales, whale sharks Rhincodon typus, large mobulid rays), and are higher in log sets than in both school and drifting FAD sets (Williams et al. 2009). Leatherback turtles, an IUCN Vulnerable species, are caught ca. 90% more frequently in school than associated sets, as they tend not to aggregate at floating objects (Hall and Roman 2013).
Billfishes catch rates are the same order of magnitude in associated and school sets (Table 2). There were lower billfishes catch rates in associated sets in the western and central Pacific, and higher in associated sets in the other three regions (Amandè et al. 2008(Amandè et al. , 2010ISSF 2017c). Some marlin stocks are overexploited (e.g., western central north Pacific Ocean striped marlin, ISC 2015).
Associated sets have higher skipjack and bigeye tuna catch rates than school sets, while school sets have higher yellowfin tuna catch rates (Dagorn et al. 2013;Harley et al. 2015). Drifting FAD sets also have higher catch of small, juvenile yellowfin and bigeye tunas than school sets (Fonteneau et al. 2013;Restrepo et al. 2017). Some bigeye and yellowfin tuna stocks are overfished (ISSF 2017b).
Environmental NGOs have prescribed seafood sourcing practices for canned tuna and several major retailers and food service companies have adopted canned tuna sourcing policies that identify canned tuna sourced from purse seine FAD sets as less ecologically sustainable than school sets, due to relatively higher bycatch rates in FAD sets (e.g., WWF 2011; ALDI 2016; Giant Eagle 2016; Wegmans 2016). These recommendations were very likely not based on assessments that accounted for the population effects of increased fishing mortality levels of species and groups with higher catch rates in purse seine school sets than in drifting FAD sets, including sea turtles, mobulids and yellowfin tuna.
This comparison does not account for other ecological risks, including habitat damage, fishing mortality from entanglement in in-use and derelict FAD surface and subsurface structures (when nonentangling designs are not employed), and potential effects of FADs on the behavior and ecology of species that associate with floating objects, possibly causing collateral population-level and broader ecosystem-level effects (Marsac et al. 2000;Hallier and Gaertner 2008;Amandè et al. 2010;Dagorn et al. 2013;Hall and Roman 2013;Sempo et al. 2013;Maufroy et al. 2015;Gaertner et al. 2016;Escalle et al. 2017;Gilman et al. 2018a).

Gillnet mesh size
Gillnets of a particular mesh size are relatively sizeselective for most species of fishes, but can have poor species selectivity (Hamley 1975;Kirkwood and  Of various gillnet gear design factors that affect size selectivity, including mesh size, netting color, twine material, number of filaments in the twine, hanging ratio, net dimensions, and twine diameter, mesh size has the largest effect for most species (He and Pol 2010). Mesh size has been regulated to balance the efficiency of catch rates of target species with the mitigation of catch rates of unwanted catch of juvenile and large adults of target and non-target species, including endangered, threatened and protected bycatch species (Price and Van Salisbury 2007;Gilman et al. 2010). However, conflicts can result from a prescribed gillnet mesh size. For example, as mesh size increases, catchability of sea turtles, marine mammals, some seabirds and other large organisms generally increases while catchability of small, juvenile age classes of target fish species decreases (Dagys and Zydelis 2002; Price and Van Salisbury 2007;Murray 2009;Orphanides 2010). Mesh size can also have conflicting effects on different elasmobranch species: The U.S. northwest Atlantic Ocean monkfish (Lophius americanus) set gillnet fishery had a 74% significantly lower spiny dogfish (Squalus acanthias) but 126% significantly higher thorny skate (Amblyraja radiata) bycatch rates in 35.6 cm (14 inch) than 25.4 cm (10-inch) mesh gillnets (Salerno et al. 2010).

Gillnet acoustic pingers
Acoustic pingers are used to reduce bycatch rates of some species of small cetaceans and seabirds in drift and set gillnets, and bycatch of some species of small cetaceans in pelagic and mid-water trawls, traps and other gears (e.g., Lien et al. 1992;Melvin et al. 2001a;Koschinski et al. 2006;Leeney et al. 2007;Berrow et al. 2009;Morizur et al. 2008;Carretta and Barlow 2011;Dawson et al. 2012). For example, a coastal drift gillnet with acoustic pingers attached to the float line and ends of each net significantly reduced the bycatch rate of common murres (Uria aalge) by 50% relative to control gear without pingers (Melvin et al. 2001a). Coastal set gillnets with active pingers attached to the float line had a 86% significantly lower Franciscana dolphin (Pontoporia blainvillei) catch rate than control nets (Bordino et al. 2002). In a controlled experiment in the Gulf of Maine coastal set gillnet fishery, there was a 92% significantly lower harbour porpoise (Phocoena phocoena) bycatch rate in gear with pingers than in control gear (Kraus et al. 1997). Analyses of observer data from the Gulf of Maine set gillnet fishery found a 60% lower porpoise bycatch rate in nets with pingers spaced at B 92 m apart than in nets lacking pingers (Palka et al. 2008). Fisher compliance with prescribed pinger use can be one obstacle to achieving the magnitude of pinger bycatch mitigation efficacy observed in experiments (Dawson et al. 2012). For some species, some characteristics of the sound emitted by pingers may be unpleasant, causing the animals to avoid the area near the pingers, or the pinger may alert some species to the presence of the gear and enable them to avoid capture (Dawson et al. 2012).
However, for bottlenose dolphins (Turisops truncatus) and some species of pinnipeds, pinger broadcasts become a 'dinner bell', alerting and attracting them to the location of the gear, increasing their local abundance and rate of depredation of catch from the fishing gear, providing a food subsidy (Kraus et al. 1997;Melvin et al. 2001a;Bordino et al. 2002;Dawson et al. 2012;Stansbury et al. 2015). For example, Bordino et al. (2002) observed a significantly higher rate of sea lion (Otaria flavescens) depredation in set gillnets with pingers than in control gear, and the sea lion depredation rate on nets with pingers increased throughout the study period. Similarly, Melvin et al. (2001a) observed a significantly higher abundance of harbour seals attending a drift gillnet with pingers than gillnets lacking pingers. This increased abundance and depredation activity by pinnipeds in the vicinity of the fishing gear in turn increases their catch risk (Hamer et al. 2011;Dawson et al. 2012), which may offset population-level benefits resulting from the food subsidy from depredation. For example, bottlenose dolphins have been captured in trammel nets and gillnets containing relatively high-intensity (loud) pingers designed to deter depredation (Northridge et al. 2003;Dawson et al. 2012). It is as yet unclear if pingers can be designed to avoid and minimize detection by species of pinnipeds and small cetaceans that depredate the catch from fishing gear, such as by using a frequency that they cannot detect, emitting randomized signals, or emitting a range of frequencies (Dawson et al. 2012).

Temporal and spatial fishery closures
There are numerous examples of how marine area closures can result in the spatial or temporal displacement of bycatch of an individual species or age classes (e.g., Powers and Abeare 2009;Diamond et al. 2010;Murray et al. 2000;Suuronen et al. 2010;SPC 2010;Sibert et al. 2012). Although less well documented, there are also examples of closed areas designed to reduce the bycatch of one species of conservation concern increasing the bycatch of another, and increasing fisheries degradation of benthic communities at other sites.
For example, area closures for a U.S. north Pacific Ocean demersal trawl fishery effectively met the objective of reducing the bycatch rate of king crab (Paralithodes camtschaticus), but resulted in an increased bycatch rate of Pacific halibut (Hippoglossus stenolepis) due to the spatial displacement of effort (Abbott and Haynie 2012). The fishery, which targets rock sole (Lepidopsetta polyxysra) roe and Pacific cod (Gadus microcephalus), had seasonal king crab and Pacific halibut bycatch limits due to concerns over the conservation status of king crab and the allocation of Pacific halibut to coastal fisheries that target this species. An additional unintended consequence of the closures was a change in fisher behavior to increase catch rates of cod and decrease catch rates of rock sole (Abbott and Haynie 2012). Baum et al. (2003) modelled the effects on catch rates from the closure of areas to the U.S. north Atlantic swordfish longline fishery. A portion of the fishing grounds called the Northeast Distant statistical area was closed for about 3 years, staring in 2000, to reduce loggerhead (Caretta caretta) and leatherback (Dermochelys coriacea) sea turtle bycatch (NMFS 2001). The spatial displacement of effort was modeled to simulate either a change in effort to achieve constant levels of swordfish catch, or to maintain a constant level of effort. Both scenarios for the closure of the Northeast Distant area were predicted to reduce catch rates of loggerhead and leatherback sea turtles and two species of sharks of relatively low conservation concern (blue and shortfin mako sharks, categorized as Low Risk by the IUCN Red List of Threatened Species), but increase catch rates of 10 shark species, including two categorized as Vulnerable by the IUCN Red List (blacktip Carcharhinus limbatus and dusky C. obscurus sharks) (IUCN 2017). The model simulations predicted mixed effects on catch rates of teleosts, with increased catch rates of some species of discarded teleosts (Baum et al. 2003).
In 2001, to reduce pressure on the depleted North Sea cod (Gadus morhua) stock, the European Commission temporarily closed to certain fishing fleets a large cod spawning area in the central North Sea. Rijnsdorp et al. (2001) observed that the closed areas resulted in the displacement of beam trawling effort to areas with increased catch rates of sensitive demersal and benthic species. Beam trawling effort redistributed to the western North Sea, where according to Rijnsdorp et al. (2001), restricted-range populations of rays and benthic vulnerable marine ecosystems occur. Similarly, Dinmore et al. (2003) predicted that continuation of the seasonal cod area closures, and concomitant continued displacement of trawling effort to areas that had not been trawled prior to the seasonal area closures, would increase cumulative impacts to benthic invertebrate production and cause localized reductions in benthic biomass.

Are bycatch mitigation methods drivers of fisheries-induced evolution?
FIE has unequivocally been demonstrated using transcriptomics on specific heritable growth traits for fish populations in the laboratory (van Wijk et al. 2013) and, by inference, genomic diversity in the wild (e.g., Hauser et al. 2002;Hutchinson et al. 2003;Thériault et al. 2008). But the full impact and scope of FIE has not yet been investigated (Heino et al. 2015). Employment of a bycatch mitigation method may inadvertently cause or strengthen a fishery's 123 intraspecific heritable trait-selective mortality, resulting in FIE. These changes in genetic diversity within species can be protracted or permanent (Kuparinen and Merila 2007;Heino et al. 2015) where causes of changes in traits from human predation outpaces most natural causes (Darimont et al. 2009). The relative catchability of individuals of a population susceptible to capture by a fishing gear type is explained, in part, by various heritable traits that show inter-individual variability. This includes, for example, behavioral traits for shyness/boldness; life-history traits such as age and size at maturation and growth rate (size-atage); physiological traits such as metabolic rate (which may be correlated with shyness/boldness), visual acuity and swimming performance; and morphological traits such as mouth dimensions (e.g., gape width and height) and body shape (Heino et al. 2015;Lennox et al. 2017;Hollins et al. 2018). We are unaware of any empirical research on bycatch mitigation method effects on FIE, and more generally, FIE in taxa other than teleosts and FIE resulting from selection on physiological, morphological and behavioral traits are not well understood (Heino et al. 2015;Hollins et al. 2018). As a result of these gaps, here we present only theoretical examples of potential ways that employment of bycatch mitigation methods might cause or increase a fishery's intraspecific selectivity of heritable traits that are variable within an affected population, in particular for populations with small effective sizes.
Changes in passive (static) fishing gear designs to mitigate unwanted bycatch can affect the gear's visual and olfactory detectability. The capture process for passive fishing gears (e.g., gillnets, longlines, traps, hook-and-line gears) relies, in part, on the movement of organisms into the gear, where, among many other factors, detectability of the gear affects fishing efficiency (Bjordal 2002; He and Pol 2010). For instance, replacing wire with monofilament for leaders used in pelagic longline branchlines conducted to reduce shark catch rates has resulted in higher catch rates of some teleosts, perhaps due to lower detectability of the monofilament leaders (Ward et al. 2008;Gilman et al. 2016b). This change in gear design might increase the probability of catching turtles with traits for relatively poorer vision: While the continuum of vision capacity of all individuals of a turtle population might enable the entire population to detect a wire leader next to a baited hook, only turtles with traits for acute vision or flicker fusion (to detect contrast) might be able to see monofilament leaders, eliciting avoidance and concomitant lower probability of capture than if they could not detect this component of the branchlines. This would increase the gear's selective removal (increasing the relative risk of capture) of the subset of the population with heritable traits for relatively poor vision (Gilman et al. 2016b).
Prescribing a minimum hook size to reduce sea turtle catchability (Gilman and Huang 2017) increases the size selectivity of some components of the catch (Gilman et al. 2018b), favoring traits for faster life histories, resulting in, for example, a reduction in the average size at first reproduction (Zhou et al. 2010). The larger the hook, the larger an organism's mouth dimensions needs to be to fit it in its mouth (Erzini et al. 1997;Gilman et al. 2018b). As a result, for species that tend to be caught by ingesting a baited hook, hooks with a larger minimum width reduce the relative catchability of smaller length classes within a species (Lokkeborg and Bjordal 1992;Cooke et al. 2005). This same concept could be applied to other gear types where a change in gear design or fishing method to mitigate problematic bycatch alters the length frequency distribution of the catch, including for example from changes in gillnet mesh size, discussed in the section ''Gillnet mesh size''.
The capture process for mobile fishing gears (e.g., trawls, purse seines) involves actively encircling or towing the gear into the path of the target species (Bjordal 2002). Some pelagic trawl captains, on very rare occasions when they detect a dolphin or porpoise in the net, reduce their tow speeds to attempt to allow the marine mammals to escape. A slower tow speed might increase intraspecific selectivity of fishes and marine mammals susceptible to capture, as it may reduce relative catch risk and hence favor genotypes possessed by members of a population for physiological traits such as for anaerobic swimming (i.e., sprinters), other swimming performance traits, and perhaps behavioral traits supporting escapement with slower trawl tow speeds (e.g., Gabr et al. 2007;Hollins et al. 2018).
Measures that reduce tuna purse seine sets on FADs and increase school sets in order to reduce shark bycatch rates (refer to the section ''Tuna purse seine set type'') might increase selectivity for individuals of a species with heritable traits favoring maintaining schooling behavior when experiencing 'predation' by purse seine gear. This could occur if the intraspecific variability in behavior of remaining in or leaving a school during a predator attack (Pitcher 1995;Pavlov and Kasumyan 2000) occurs when experiencing predation when in a free swimming school but not when aggregated in a shoal in a fixed location such as under a drifting or anchored FAD. Under natural conditions, individuals possessing traits for atypical behavior of discontinuing schooling behavior when the school is being preyed upon would have the highest predation risk (Pitcher 1995;Pavlov and Kasumyan 2000), but when an unassociated school is being pursued by a purse seine, this minority trait would improve their survival probability (assuming that they survive following escape in a suboptimal school size, such as by merging with another school).
Vessels fish deeper to avoid surface-habitat bycatch species, such as by pelagic trawls to avoid dolphin bycatch (e.g., Thompson et al. 2013), and pelagic longlines to avoid sea turtles and epipelagic species of elasmobranchs (FAO 2010;Gilman 2011;Musyl et al. 2011; and see the section ''Pelagic longline time-ofday and gear depth''). This increases the relative catchability and selective fishing mortality of individuals of a population possessing physiological traits that enable them to make use of deeper habitat where lower temperatures and oxygen concentrations occur Hollins et al. 2018).
Spatio-temporal bycatch mitigation measures can also result in unintended FIE. This would occur if a time-area fisheries closure displaces fishing effort to times (time-of-day, season) and sites where a narrower proportion of the variations of a trait that exists for an affected population occurs. For example, because many marine species exhibit temporal and spatial variability in habitat use by size (e.g., sharks, Stevens and McLoughlin 1991;teleosts, Solmundsson et al. 2015), an MPA that displaces fishing effort to an area with a narrower length frequency distribution of a species could result in an increase in size selectivity. More specifically, if small size classes of a species occur primarily in nearshore habitats while larger individuals are distributed offshore, an MPA that restricts fishing to offshore waters that increases selection for larger organisms would favor heritable traits for faster life histories (e.g., earlier maturation) (Heino et al. 2015). MPAs could, however, be designed to reduce FIE (e.g., Hall 2002;Dunlop et al. 2009).
If employment of a bycatch mitigation method inadvertently causes intraspecific heritable trait-based selectivity, this would reduce genetic diversity of affected populations by reducing the occurrence of phenotypes for traits associated with higher catchability. Reduced intra-population genetic diversity reduces a population's fitness, resistance and resilience to natural pressures and concomitant persistence, and can compromise natural selection and the ability to evolve in response to environmental changes and stressors, also increasing the risk of population extirpation (local extinction) as a result of the reduction in adaptive genetic variation (Saccheri et al. 1998;Westemeier et al. 1998;Reed and Frankham 2003;Ehlers et al. 2008;Evans and Sheldon 2008). Not accounting for FIE-influenced changes in life-history characteristics in stock assessments can compromise the robustness of model estimates (Audzijonyte et al. 2016). Similarly, bycatch mitigation measures may inadvertently exacerbate fishing mortality of phylogenetically distinct species, compromising the potential for evolution at community and ecosystem levels (Kareiva and Marvier 2003;Redding and Moores 2006;Gilman et al. 2014a, b). These examples highlight the importance of assessing and managing fisheries bycatch impacts across levels of biodiversity, including genetic diversity, impacts on individual populations and species, and broader community-and ecosystem-level effects (e.g., McNeely et al. 1990), discussed in the next section.

How to transition to integrated bycatch assessment and management
The examples presented here illustrate how some bycatch mitigation methods reduce catch and mortality of one species or group of conservation concern but exacerbate catch and mortality of another. Theoretically, employment of some bycatch mitigation methods might inadvertently reduce genetic diversity within populations, causing or exacerbating FIE. Bycatch mitigation can also affect evolutionary processes of communities by inadvertently increasing direct or collateral fishing mortality of phylogenetically distinct species. Some bycatch mitigation measures, through food web links, cause or strengthen drivers for broad community-and ecosystem-level changes in functions, structure and resilience, with 123 concomitant effects on the provision of targeted ecosystem services, including fishery yields (Pace et al. 1999;Stevens et al. 2000;Ferretti et al. 2010). These examples support moving from piecemeal to holistic bycatch assessment and management so that the relative risks across affected components of biodiversity resulting from fisheries bycatch are identified and accounted for in fisheries decisionmaking processes. Intentional cross-taxa bycatch tradeoffs could be designed to balance exploitation, distributing fishing mortality across populations at sustainable levels, preserving the state of the ecosystem characteristic of unexploited conditions (Hall 1996;Zhou et al. 2010;Garcia et al. 2012).

Ecological risk assessments
The transition to holistic bycatch assessments would be supported by adapting Productivity-Susceptibility Analyses (PSAs) and other methods for semi-quantitative and quantitative, model-based ERAs of the effects of fisheries bycatch to assess relative risks to affected populations, cumulative effects from regional fisheries, and to assess broader ecosystem-level effects. ERAs of the effects of fisheries bycatch need to move beyond single taxa species-level assessments to broader, more holistic assessments of risks across levels of biodiversity, including genetic diversity, impacts on individual populations and species, and broader ecosystem-level effects. Methods for ERA of the effects of fishing have recently been developed for the continuum of data-poor to data-rich fisheries. ERA methods include rapid, first order, qualitative evaluations, semi-quantitative assessments, and modelbased quantitative assessments (Dambacher et al. 2009;Hobday et al. 2007Hobday et al. , 2011Weijerman et al. 2015).
The objectives of analysis of most semi-quantitative fisheries ERAs have been to determine population-and species-level relative risks from fishing mortality of individual taxonomic groups of conservation concern, most employing PSAs (e.g., Stobutzki et al. 2002;Waugh et al. 2008;Cortes et al. 2015). PSAs assess productivity through use of attributes for intrinsic factors, such as demographic characteristics of a population, stock or species. These productivity attributes provide an indicator of relative resistance to fishing mortality and resilience or ability to recover from depletion. Susceptibility considers extrinsic factors that influence catch risk, such as encounterability, selectivity and post-release mortality, where some attributes are modified by the use of mitigation strategies (Patrick et al. 2009;Hobday et al. 2011;Gilman et al. 2017b). Species with high productivity and low susceptibility have lower risk from a fishery than species with lower productivity and higher susceptibility scores (Fig. 4).
PSAs and other semi-quantitative ERAs of the effects of fisheries bycatch provide a first order assessment of relative risk, and support prioritizing the focus of more rigorous quantitative assessments that estimate absolute population-level effects from fishery removals . PSAs identify the relative risk to individual populations from the effects of fisheries bycatch. Populations identified in the PSAs as having the highest risk could then undergo robust quantitative modeling to simulate absolute risks to individual populations that would result from implementing alternative bycatch management measures. This would identify any cross-taxa conflicts, information needed to holistically manage bycatch.
The objectives of analysis of model-based quantitative methods for ERAs of the effects of fishing include to assess absolute risk, e.g., of population extirpation, irreparable harm to a stock, habitat loss, and loss in genetic diversity Zhou et al. 2011). Conventional single stock assessment methods are the most common form of a quantitative ERA, used to assess the status and temporal changes in stock status and predict stock responses to different Relatively low risk (high productivity, low susceptibility) Relatively high risk (low productivity, high susceptibility) Fig. 4 Template productivity-susceptibility analysis plot, an example of a semi-quantitative ecological risk assessment of the effects of fishing on populations, stocks, species or habitat.
[adapted from Hobday et al. 2011;Williams et al. (2011)] management options (Hilborn and Walters 1992). They are typically used for principal market species but more recently have also been used for stocks of sharks and other incidentally caught species of conservation concern, including through new assessment methods designed for data-poor stocks (e.g., Fu et al. 2016). Numerous analytical approaches have been used in quantitative ERAs of the effects of fishing to define threshold population sizes and fishing mortality rates beyond which populations are at risk of irreparable harm or extirpation, with a range of data requirements and concomitant range in certainty in outcomes (Wade 1998;Musick 1999;Chaloupka 2003;Lewison and Crowder 2003;Dulvy et al. 2004;Hobday et al. 2011;Zhou et al. 2011). Findings from population models can identify the relative risks from alternative bycatch mitigation methods on absolute population-level effects, guiding management responses. For example, there are now numerous adaptations of a model that does not require detailed demographic information, designed to determine mortality thresholds for marine mammals, known as Potential Biological Removal (PBR) (Wade 1998), including for seabirds (Dillingham and Fletcher 2011;Richard and Abraham 2013) and sea turtles (Curtis et al. 2015;Casale and Heppell 2016). A more rigorous population simulation modeling approach than PBR that also can be applied given minimal data compares relative competing risks of various mortality sources, including from fisheries bycatch (Chaloupka 2003). Examples of biological reference points that define a threshold beyond which increased fishing effort and mortality risks irreparably damaging a fish stock include F crash , the fishing mortality rate that will drive a population to 1/1000 of virgin biomass, and B crit , the minimum viable population size below which population extirpation is imminent (Mace 1994;Zhou et al. 2011;Gilman et al. 2014a).
As applied in stock-specific harvest strategies, multispecies and ecosystem-based limit reference points can also be defined as thresholds for undesirable conditions for an ecosystem. When stock-specific limits are exceeded, this triggers pre-agreed controls designed to rebuild the stock, thus eliminating the risk of serious or irreversible harm or extirpation. Similarly, when ecosystem-level thresholds are exceeded, controls would be triggered that alter the level of pressures that caused the unwanted change in ecosystem state (Sainsbury et al. 2000;ICES 2001;Collie and Gislason 2001;Gilman et al. 2014bGilman et al. , 2017b. Global biodiversity research and conservation activities have generally focused on species-level criteria (see review by Gilman et al. 2011). Methods for ERAs of the effects of bycatch fishing mortality that comprehensively consider biodiversity across its hierarchical manifestations are needed. Multispecies and ecosystem models are additional examples of quantitative ERAs that can be designed to determine ecosystem changes in response to pressures, including from bycatch removals, and simulate ecosystem effects of alternative bycatch management approaches (e.g., Ecopath with Ecosim: Polovina 1984;Walters et al. 1997;Christensen and Walters 2004;Atlantis: Fulton and Smith 2004;; Models of Intermediate Complexity for Ecosystem assessment: Plaganyi et al. 2014;andqualitative modeling: Dambacher et al. 2009, 2015). Different ecosystem models vary in their balance between realism, accuracy and complexity, and have a wide range of data requirements (Weijerman et al. 2015). For example, the potential ecological and economic benefits of eliminating bycatch (the ''maximum dexterity'', or ''clean fleet'' scenario) were explored using simulations with food web models of ecosystems in northern British Columbia, Canada and in Raja Ampat, Indonesia (Pitcher and Ainsworth 2010). In both systems, increased economic returns were suggested by longterm simulations encompassing 100% or 80% bycatch reductions.
ERAs should also assess risks from the effects of bycatch removals on intraspecific genetic diversity and phylogenetic diversity resulting from selective fishery removals. The relative risks of intra-population FIE from alternative bycatch management methods could be predicted (e.g., Kenchington et al. 2003;Kuparinen and Merila 2007;Heino et al. 2015;Hollins et al. 2018). Similarly, evaluation of effects of bycatch management options on genetic diversity among species should also be conducted. An ERA of the effect of a pelagic longline fishery by Gilman et al. (2014a) compared species' PD 50 indices, which is the expected phylogenetic diversity (PD) loss if the species goes extinct, assuming all other species have a 50% probability of persistence. This index of phylogenetic uniqueness enables an assessment of species' relative importance for maintaining unique 123 evolutionary history and for the potential continuation of evolutionary processes (Faith 2008;Sgro et al. 2011). However, evolutionary histories are not available for all taxonomic groups, and, discussed below, there is no agreed standardized way to compare the relative taxonomic distinctness of species from unrelated groups.

Data quality
Robust assessments of the relative risks to individual populations from fisheries bycatch require adequate data quality. Improvements in some fisheries monitoring and research programs are needed to identify cross-taxa conflicts resulting from bycatch mitigation methods to support holistic bycatch management. Fisheries management systems, even those that are rudimentary and governing data-poor fisheries, can successfully transition to meeting many of the data requirements of integrated bycatch assessment and management (Gilman et al. 2017b). Cross-taxa conflicts likely occur in many gear types, but have not been documented due to inadequate monitoring or gaps in research. For example, data gaps need to be filled to determine if increasing the fishing depth of pelagic trawls to reduce dolphin bycatch (e.g., Thompson et al. 2013) increases bycatch of mesopelagic species, such as thresher sharks. Similarly, data assessing fishers' response to bans on wire leaders in pelagic longline fisheries to reduce shark catch rates are needed to determine if this resulted in increased seabird catchability. When required to replace wire with monofilament leaders, pelagic logline fishers may respond by attaching branchline weights further from the hook due to safety concerns-if a branchline breaks during hauling, which frequently occurs when sharks are caught and sever leaders not made of wire or other durable material, the weight can fly at the vessel at high velocity when the weight is located near the hook-reducing the baited hook sink rate and increasing seabird bycatch rates (Gilman et al. 2016b). Monitoring methods require adjustments in some fisheries to enable the identification of cross-taxa conflicts resulting from alternative bycatch mitigation methods. For example, data fields and data collection protocols by onboard observers require expansion (e.g., Gilman and Hall 2015).
Data also need to be collected to determine which populations of species of conservation concern are captured in a fishery, a prerequisite to conducting quantitative ERAs that estimate absolute populationlevel effects from bycatch removals. As an example, there is minimal information on which cetacean species are captured in most Pacific Island pelagic longline fisheries, and little knowledge of the structure and conservation status of populations of cetacean species that overlap with these fisheries (SPREP 2007;Gilman et al. 2013). Minimally invasive tissue biopsy samples could be collected and analyzed to determine which populations are caught in a fishery.
Even when information on which populations are captured in a fishery is available, data requirements for robust population models are frequently not met. For many bycatch species of conservation concern, improved information is required on key life history characteristics, information to determine population structure, and accurate estimates of all sources of mortality, including total fishing mortality by age class and sex (e.g., NMFS 2012;Camargo et al. 2016;Gilman et al. 2017b).
Similarly, ERAs on the effects of bycatch require improvements in understanding of effects on genetic diversity within and between species. Increased knowledge is needed of the phenotypic ranges of intraspecific diversity of traits that explain susceptibility to fisheries capture (Hollins et al. 2018). FIE in species of conservation concern (i.e., taxa other than teleosts) and FIE driven by selectivity for physiological, morphological and behavioral traits (i.e., traits other than life-history attributes) require improvements in understanding (Heino et al. 2015;Hollins et al. 2018). Expansions of data collected by fisheries monitoring programs are also required. For instance, monitoring temporal trends in genetic diversity within populations requires sufficient time series of genetic markers for traits affected by selective fishery removals, which requires long-term collection of tissue samples for the extraction of DNA (Kenchington et al. 2003;Heino et al. 2015).
Improvements are needed in many data inputs for models of ecosystem-level effects of fishing and evaluation of bycatch management options (one component of fisheries management strategy evaluation), including from bycatch removals (e.g., Weijerman et al. 2015). Monitoring methods require adjustment to supply data needed for fisheries managers to track temporal changes in indicators of the state of ecosystem components that are sensitive to bycatch fishing mortality. Ecosystem state indicators that are sensitive to bycatch fishing mortality include, for example: size structure within and across trophic levels, the number and biomass of functional groups within trophic levels, mean trophic level of the catch, the biomass of bycatch species' prey, the biomass of vulnerable species, the ratio of the biomass of pelagic to demersal fishes, and response diversity (the variety of responses to pressures and environmental changes by components of a facet of biodiversity-genotypes within a species, species within a functional group, functional groups within a trophic level, etc.) (Elmqvist et al. 2003;Fulton et al. 2005;Link 2005;Suding and Hobbs 2009;Coll et al. 2016).
Decision support tool for integrated bycatch assessment and management Progress towards holistic bycatch management could be expedited through developing and having fisheries management authorities use decision support tools for bycatch assessment and management. The tools would benefit fisheries managers who struggle with conventional, piecemeal, bycatch assessment and management, as an integrated approach would be more efficient through employing a single, comprehensive bycatch program in place of separate taxa-specific programs. The tool's identification of which risk assessment methods are feasible given available data inputs would reduce complexity. The tool could augment the ability of fisheries management authorities to account for findings of integrated bycatch ERAs that identify cross-taxa conflicts resulting from alternative bycatch mitigation approaches. Use of the decision support tool could reduce the likelihood of having political attention respond to taxa-specific advocacy and instead maintain a holistic focus. The decision support tool would need to address several complex issues where there is no unequivocal, most rigorous approach, including how to determine the relative risk to species of different taxonomic groups, how to compare indices of phylogenetic uniqueness between species of unrelated groups, and how to account for uncertainty in ecosystem model simulations of alternative bycatch management approaches (Isaac et al. 2007;Pitcher and Ainsworth 2010;Gilman et al. 2014b). The tool could identify approaches to increase the likelihood of fisher uptake of integrated bycatch management measures, where fisher-led methods to reduce discarded bycatch can improve food security through a resilient fishing strategy ''across the food web'' (Pitcher and Lam 2010). It would provide managers with requisite information to enable them to then make agonizing choices when designing holistic, integrated bycatch management measures so that unavoidable tradeoffs are intentional and acceptable. Existing tools for management strategy evaluation could be employed for this purpose. An Atlantis ecosystem model developed for the southeast Australian Commonwealth scalefish (Seriolella brama) and shark fishery, a complex ecosystem model used for quantitative management strategy evaluation Fulton et al. 2014), is an example of an existing management strategy evaluation method that could be employed as a decision support tool for integrated bycatch management. Integrated bycatch management strategy evaluation would simulate what effect alternative bycatch management measures would have on catch rates and levels of each species susceptible to capture in the fishery, providing information to support decisions on what are acceptable tradeoffs.
In addition to being designed to account for crosstaxa conflicts of alternative bycatch management measures, the bycatch decision support tool could adapt wetlands and terrestrial sequential mitigation hierarchy (e.g., EPA 2015) to provide decision support for broader management of fisheries bycatch (Wilcox and Donlan 2007;Finkelstein et al. 2008;UNEP-WCMC 2016). The sequence of bycatch management steps are: avoidance, minimization and offsets or compensatory mitigation. The tool could be designed to account for the fishery-specific context, including the existing management system, such as resources for surveillance and monitoring, and current conservation and management measures (input, output, temporaland spatial-restrictions, gear and fishing method restrictions). And, the tool could evaluate and rank alternative bycatch mitigation methods, and combinations of methods, against criteria for efficacy, the economic viability, practicality, safety and ability to facilitate compliance monitoring of each bycatch mitigation method, in addition to a criterion on cross-taxa conflicts resulting from each method (WPRFMC 2018).
A decision support tool should be designed so that it could be effectively used across types of fisheries, from data-poor to data-rich fisheries, from artisanal/ 123 small scale to industrial/large scale fisheries, and from fisheries with relatively rudimentary management systems with relatively low institutional and financial resources to fisheries with relatively robust management systems and ample resources. The tool could be designed to enable employment from local to regional fisheries, where the latter approach enables comprehensively assessing and managing cumulative impacts.
In summary, tools would support integrated ecological risk assessment and management of the effects of fisheries bycatch could: • Characterize and benchmark existing knowledge of bycatch ecological risks, contemporary bycatch management framework, and fishing practices and gear designs that affect bycatch • Guide end users to select ecological risk assessment methods that are feasible to implement given available data quality • Evaluate and rank bycatch mitigation methods following sequencing hierarchy, and considering cross-taxa conflicts, bycatch mitigation efficacy, practicality, crew safety, economic viability and feasibility to monitor compliance • Conduct bycatch management strategy evaluation, enabling the adoption of acceptable tradeoffs where unavoidable • Be designed to be as simple as possible to maximize the likelihood of successful on-theground implementation • Be adaptable for use across fisheries, from datapoor to data-rich, artisanal/small-scale to industrial/large scale, and from those with rudimentary to robust management frameworks, and from local (stage 2, Fig. 5) to regional scale (stage 3, Fig. 5).
Approaches to bycatch management can be described as progressing along a continuum from conventional narrow-scale, fishery-by-fishery and species-by-species, to nested-scale ecosystem-based fisheries management (EBFM), to cross-sectoral marine ecosystem-based management (EBM) (Fig. 5). The capacity to progress along this continuum will vary by fishery, country and region (Pitcher et al. 2009;Gilman et al. 2017b). Most fisheries management frameworks that manage bycatch are at the first stage of Fig. 5, where bycatch is managed fishery-by-fishery and species-by-species. The Transitioning from conventional narrow-scale and species-by-species bycatch management to nested-scale EBFM, to cross-sectoral marine EBM. (Sainsbury et al. 2000;Link 2005;Gilman et al. 2017b) decision support tool would aim to get fisheries to stage 3, with integrated bycatch management at a nested-scale, across regional fisheries. EBFM augments conventional target species single-stock, singlefishery management approaches to address broader effects of fishing (FAO 2003;Garcia et al. 2003). The EBFM extension to single-stock reference points is ecosystem-based reference points, which account for ecosystem effects of fishing activities (Rice 2000;Pitcher 2001;Rochet and Trenkel 2003). Similarly, EBFM extends single stock management through ecosystem-based harvest strategies. EBFM is one component of holistic marine EBM, which requires cross-sectoral planning and management of marine pressures (Pikitch et al. 2004;Crowder and Norse 2008;Link and Browman 2014). Successful mitigation of the main global drivers of change and loss in marine biodiversity that adversely affect the fishing industry but are largely caused by other industry sectors, including marine pollution, outcomes of climate change, habitat degradation, and invasive alien species (Leadley et al. 2010) will increasingly require effective cross-sectoral collaboration. These developments would lead, hopefully, from old-style management that tends to be utopian, narrow, and prescriptive, to a new era of practical, holistic, adaptive and industry-supported strategies (Pitcher and Lam 2010