Selection of indicators for assessing and managing the impacts of bottom trawling on seabed habitats

Bottom trawl fisheries are the most widespread source of anthropogenic physical disturbance to seabed habitats. Development of fisheries-, conservation- and ecosystem-based management strategies requires the selection of indicators of the impact of bottom trawling on the state of benthic biota. Many indicators have been proposed, but no rigorous test of a range of candidate indicators against nine commonly agreed criteria (concreteness, theoretical basis, public awareness, cost, measurement, historical data, sensitivity, responsiveness, specificity) has been performed. Here, we collated data from 41 studies that compared the benthic biota in trawled areas with those in control locations (that were either not trawled or trawled infrequently), examining seven potential indicators (numbers and biomass for individual taxa and whole communities, evenness, Shannon–Wiener diversity and species richness) to assess their performance against the set of nine criteria. The effects of trawling were stronger on whole-community numbers and biomass than for individual taxa. Species richness was also negatively affected by trawling but other measures of diversity were not. Community numbers and biomass met all criteria, taxa numbers and biomass and species richness satisfied most criteria, but evenness and Shannon–Wiener diversity did not respond to trawling and only met few criteria, and hence are not suitable state indicators of the effect of bottom trawling. Synthesis and applications. An evaluation of each candidate indicator against a commonly agreed suite of desirable properties coupled with the outputs of our meta-analysis showed that whole-community numbers of individuals and biomass are the most suitable indicators of bottom trawling impacts as they performed well on all criteria. Strengths of these indicators are that they respond strongly to trawling, relate directly to ecosystem functioning and are straightforward to measure. Evenness and Shannon–Wiener diversity are not responsive to trawling and unsuitable for the monitoring and assessment of bottom trawl impacts.


| INTRODUC TI ON
Bottom trawls, here defined as any towed bottom fishing gear including otter trawls (OT), beam trawls, scallop dredges and hydraulic dredges (HD), are used to catch fish and shellfish living in, on or near the seabed (Sainsbury, 1986). Bottom trawling is by far the largest source of human physical disturbance in the marine environment, but also makes an important contribution to global food supply, accounting for 19-25 M tonnes of annual fish landings (Amoroso et al., 2018). It is therefore important to quantify trawl impacts to assess sustainability and guide management in the context of wider ecosystem management and conservation (Clark et al., 2016;McConnaughey, Hiddink, Jennings, Pitcher, et al., 2020). (Sciberras et al., 2018) and selection for communities dominated by short-lived fauna have been documented in response to bottom trawling (van Denderen et al., 2015). This can lead to changes in community production, trophic structure and ecological function (Duplisea, Jennings, Malcolm, Parker, & Sivyer, 2001;Hiddink et al., 2006) and can cause reductions in the prey abundance of commercial fish species (Collie et al., 2017).

Reductions in faunal biomass, numbers and species richness
When protection of habitats and their associated biota are the management objectives, the implementation of ecosystem-based fisheries management requires information on the distribution and impact of bottom trawling, and status of biota and habitats (Rijnsdorp et al., 2016). This information enables assessment of the intensity of potential impacts which can be used to help society achieve an accepted balance between fisheries production and environmental protection (Rice, 2005(Rice, , 2011. Evaluating the consequences of management interventions requires indicators of the state of seabed environment. Furthermore, commitment to marine policies such as the European Marine Strategy Framework Directive and evaluation of descriptors therein such as 'seafloor integrity', requires the development of indicators of trawling impacts that capture changes in the structure and function of benthic ecosystems (Rice et al., 2012).
Here we define 'state' as the condition of the ecosystem, while impact is the change in this state in response to trawling pressure relative to its untrawled reference level. State indicators to support the management of bottom trawling impacts on benthic ecosystems should satisfy a range of requirements (Jennings, 2005;Rice & Rochet, 2005).
The theoretical basis for the cause-and-effect between trawling and the indicator should be easily understood and intuitive, as this would facilitate acceptance and support among stakeholders and the wider public. Effective indicators should quantify ecologically important parameters that relate to changes in the structure and functioning of the benthic ecosystem, both of which correlate closely to benthic biomass (Hiddink et al., 2006;Queiros et al., 2013). The parameter should be easily measured, sensitive to fishing impacts and provide rapid and reliable feedback on the efficacy of management actions. Changes in the indicator should be specific to the effect of trawling rather than confounded by environmental variation, unless other sources of variation are understood, quantifiable and can be accounted for. Attribution of causality for changes in ecosystem properties is challenging, given that all the changes in trawled communities are not necessarily responses to trawling. Finally, indicators for which (historical) data are available and that are cost-effective to generate are preferable. In practice, the best indicators will exhibit a strong response with a low variance, indicating a high specificity of the response, and will include only small effects of other environmental variation (Maxwell & Jennings, 2005).
A number of indicators of the impact of trawling on benthic ecosystems have been proposed, including numbers, biomass, species richness, measures of diversity and trait-based community descriptors of benthic biota (e.g. Rijnsdorp et al., 2016;van Loon et al., 2018). However, the utility of many commonly used indicators, such as species richness, has not been tested and no systematic comparison of the sensitivity nor specificity of different indicators has been performed. Such tests are needed given that some of the currently used indicators are in fact insensitive to trawling, and respond instead to environmental gradients (e.g. Gislason, Bastardie, Dinesen, Egekvist, & Eigaard, 2017).

Performance of indicators can be assessed by comparing their
responses to a known pressure. Searches of the literature revealed many trawling impact studies where the benthic community is compared in two or more areas with contrasting, although not always quantified, trawling intensity (e.g. Engel & Kvitek, 1998;Sciberras et al., 2013). These control-impact studies provide an opportunity to compare the sensitivity, responsiveness and specificity of different indicators. Here, we perform a systematic evaluation of potential state indicators of bottom trawl impacts by testing each indicator against the criteria defined by ICES (2005) and Rice and Rochet (2005): concreteness, theoretical basis, public awareness, cost, measurement, historical data, sensitivity, responsiveness and specificity (Table 1). To test sensitivity, responsiveness and specificity, we perform a meta-analysis of comparative control-impact studies well on all criteria. Strengths of these indicators are that they respond strongly to trawling, relate directly to ecosystem functioning and are straightforward to measure. Evenness and Shannon-Wiener diversity are not responsive to trawling and unsuitable for the monitoring and assessment of bottom trawl impacts.

K E Y W O R D S
beam trawl, ecosystem approach to fisheries management, hydraulic dredge, meta-analysis, otter trawl, scallop dredge, systematic review to compare the effect of different trawl gears, in different habitats and on different indicators (numbers and biomass for both individual taxa and whole communities, and three measures of diversity), while the other criteria are assessed using judgement by the authors.

| MATERIAL S AND ME THODS
Data were collated from published comparative studies of the effects of bottom trawling on seabed habitat and biota following a systematic review protocol, thereby including all available studies and avoiding selection bias (Hughes et al., 2014). The methods were designed to identify and collate evidence from comparative controlimpact studies to identify changes in state of benthic biota resulting from mobile bottom fishing. The search strategy is documented in Hughes et al. (2014), which specifies the databases searched and search terms used in detail. Our literature search period finished in 2014 and no studies beyond that date are included here. Studies were only included in the meta-analysis when they compared benthic invertebrates in two comparable areas, where one area was commercially trawled and the other was not trawled or was only lightly trawled. This excludes studies where areas were experimentally trawled, and comparative gradient studies where many different levels of quantified trawling effort were sampled. Included studies were restricted to those performed on the continental shelf and upper slope (0-400 m) and to those reporting numbers, biomass or diversity of benthic communities, species, genera or families of infauna or epifauna. Studies needed to report the mean and a measure of variation, such as a standard deviation or confidence interval, to be included in the meta-analysis. Our analysis of comparative studies assumed that other environmental covariates did not correlate or vary with trawling intensity at the scale of the experiments. Studies where this assumption was apparently violated in our assessment of study quality, such as in Hixon and Tissot (2007) where the depths of trawled and control areas diverged by up to 180 m and the species composition in the two areas diverged greatly, were not included in the analysis as this would confound environmental with trawling effects (see Text S1). The meta-data extracted for each study (including location, depth, trawl gear type, habitat) are provided in Table S1.
Gear types in the studies were classified as OT, beam trawls (BT), towed dredges (TD) and HD. Further details of the methodology are available in Hughes et al. (2014). Although this approach may underestimate the effect of trawling, excluding these studies would have removed almost all studies on biogenic habitats. We address the extent of this underestimate in our interpretation of results. Trawling intensity was not quantified in most studies, but where trawling frequency was quantified, the mean swept-area-ratio was 3.36 year −1 in the trawled area (range: 0.2-12.9) and 0.1 year −1 in the control area (range: 0.0-0.4).

| Analysis
Studies were analysed using weighted meta-analysis via linear mixed-effects models (a standard approach for meta-analysis, using rma.uni function in r package metafor, Viechtbauer, 2010) with the log response-ratio (lnRR) for the candidate indicator (I) as the response variable, calculated as ln(I trawled /I control ), where the logtransformation helps to homogenize and normalize the residuals.
Studies were weighted by the inverse of variance of the original study, where the combined variance per study was calculated as in Borenstein, Hedges, Higgins, and Rothstein (2009). A significant effect of trawling is present when the 95% confidence intervals of lnRR do not overlap with lnRR = 0. TA B L E 1 Criteria for the selection of state indicators from ICES (2005) and Rice and Rochet (2005) Criteria

Concreteness
Directly observable and measurable property of physical/biological world rather than reflecting abstract properties which can only be estimated indirectly

Theoretical basis Link between pressure and indicator based on well-defined and validated theoretical links
Public awareness Public understanding consistent with its technical meaning. Nature of what constitutes 'serious harm' is widely shared

Cost
Uses measurement tools that are widely available and inexpensive to use

Measurement
Measurable in practice and in theory, using existing instruments, monitoring programmes and analytical tools, and on the time-scales needed to support management. Minimum or known bias, and signal should be distinguishable from noise Historical data Supported by a body or time series of data to aid interpretation of trends and to allow a realistic setting of objectives Sensitivity Trends should be sensitive to changes in the ecosystem state, pressure or response that the indicator is intended to measure

Responsiveness
Responsive to effective management and provides rapid and reliable feedback on the consequences of management Specificity Responds to the properties they are intended to measure, rather than to other factors and/or it should be possible to disentangle the effects of other factors from the observed response

| Response measures (I) for calculating lnRR
Studies reported many different metrics for benthic fauna, including numbers and biomass for individual taxa (at different levels ranging from species to phylum) and for whole communities. Candidate indicators examined were: numbers and biomass by taxa and for the wholecommunity, species richness, Shannon-Wiener diversity Hʹ, Margalef's d and Simpson's dominance D and evenness Jʹ. Other potential indicators were reported in a few studies but not included in this analysis because fewer than five studies reporting their effects were available and they fell outside the scope of the systematic review (e.g. ABC plots in Vergnon &Blanchard, 2006 andTDI in de Juan &Demestre, 2012).
All indicators were used as reported in the studies. Responses for 'taxa' indicate the responses of the abundance of all individual taxa that were reported in the studies (rather than the response of the summed abundance of taxa, which is already reported as numbers or biomass for the whole community). Taxon

| Environmental covariates determining the effect of trawling in comparative studies
Environmental factors play a role in determining the magnitude of effect of trawling on seabed biota. Thus we evaluated the influence of a number of environmental variables, at the between-study level, by including them individually as covariates in the mixed-effects metaanalysis. The significance of each of the covariates in isolation was assessed using the p-value of the Q M test statistic (Borenstein et al., 2009). The effect of trawling on benthos is likely to increase when the fraction of animals depleted (d) by a trawl pass is high, and is likely to decrease when recovery from trawling (r) is fast. Pitcher et al. (2017) showed that the effect of trawling on benthic biomass is proportional to the d/r ratio when population growth is determined by logistic population dynamics. Therefore, we examined a number of environmental covariates that are related to d and r, and may thus influence both the magnitude of depletion and the rate of recovery following trawling.
Values of r are expected to depend on variables that affect growth rates of individuals and populations. Thus, the following covariates for r were examined: primary production (PP) estimated from the vertically generalized productivity model ( (Folk, 1954) where needed, and then converting the Folk classification to percentages based on the means in each Folk category.
In addition to analyses using covariates of r, we also conducted analyses using covariates of the d/r ratio, using gear-specific d es-  Table 2). The effect of trawling is expected to increase with water depth due to the lower levels of natural disturbance in deeper water and the corresponding increase in the relative abundance of individuals with slower life histories (low r), so d × depth was examined as a covariate for d/r, with depth expressed as a negative number. Some of these covariates are ad hoc approximations of relationships that are likely to be more complex.
Habitat categories and gear type (OT, BT, TD and HD) were also examined as categorical variables, but a category was only included in the analysis when the number of studies was >2. The effects of environmental covariates on Hʹ and Jʹ were not examined because of the limited number of replicate studies that reported these response variables.

| RE SULTS
In total we found 41 control-impact studies with 18 studies reporting the effect of otter trawling, 20 studies of TD and three studies of hydraulic dredging ( Europe and NE USA ( Figure 1). All studies were carried out on the continental shelf, with only three studies at depths > 100 m.
Significant effects of trawling were detected on the indicators 'numbers of individuals in individual taxa' (mean: −35%, for confidence intervals see Figure 2), 'numbers of individuals in whole communities' (−43%) and for whole-community biomass (−59%, with the lowest upper confidence limit), but not for the biomass of individual taxa (−14%, Figure 2, although using a less conservative 90% confidence interval also results in a significant effect for taxa biomass).
The effect on species richness was smaller but significant (−21%), while the effects on the other measures of community diversity (Jʹ and Hʹ) were small and not significant, with evenness Jʹ increasing with trawling ( Figure 2).
Several environmental covariates explained a significant amount of variation in the response of indicators, although most did not.
There was a significant negative relationship between the ratio of the depletion to primary production ratio (log 10 d/PP) and the lnRR of the number of individuals in the community (p = 0.014, Figure 3c) and species richness (p = 0.043, Figure 3e), with the effect of trawling being stronger for fishing gears that cause a higher depletion and the effect being weaker in areas of high primary production (Table S2).
This means that the impact of trawling is larger for gears that deplete a larger fraction of fauna, such as dredges, and in areas with a lower food supply to the benthos where recovery is likely to be slower.
Although the effects of gear and habitat on lnRR were not significant for most outcomes (Table S2) Table 2), PP = primary production (mg C m −2 day −1 ) taxa, the effect of habitat was significant, with strong effects on biogenic habitats (which are mostly dredged, Table 2), smaller effects on muddy sediments (which are mostly otter trawled, Table 2) and the weakest effects on sand (Figure 4a). We could not disentangle the gear-habitat interaction in our analysis because of a lack of studies.

| D ISCUSS I ON
Community numbers and biomass met all performance criteria (9/9), taxa numbers (8/9) and biomass (4/9) and species richness (8/9) met many criteria. Whole-community numbers and biomass satisfied most criteria and are, therefore, the most suitable for monitoring the effect of bottom trawling on seabed biota. Evenness (2/9) and Shannon-Wiener (1/9) diversity did not respond to trawling and met few criteria, and hence are not suitable state indicators for monitoring and assessing the effect of bottom trawling on the seabed biota (Table 3).

Strengths of whole community numbers and biomass as indicators
are that they respond strongly to trawling, reflect aspects of ecosystem functioning and are straightforward to measure. All of these indicators are expensive to measure as they require sampling of the seabed from vessels, but benthic sample processing is substantially cheaper for whole-community biomass and numbers than for the other indicators because no identification of fauna is required. Taxa numbers and biomass and species richness also met most criteria. The whole-community biomass-based indicator also has the particular advantage that it is likely to correlate more closely to ecosystem functioning than numbers and richness, because it incorporates the effects on body size and age structure, as well as numbers and energy flow through food webs and other ecosystem processes that are linked closely to biomass (Hiddink et al., 2006;Queiros et al., 2013). However, these separate properties are confounded in the whole-community biomass primary production. This means that similar amounts of fishing will affect less productive communities more relative to more productive communities, as previously observed for trawl impact studies . The effects of trawling were particularly strong in biogenic habitats, in coarse sediment habitats trawled by dredges, and weaker in finer sediment habitats trawled by OT. Similar effects were found in a meta-analysis of comparative gradient studies .
The responses of the indicators to trawling may be correlated.
Bottom trawling reduces the number of individuals for many spe- The significance of the differences is presented in Table S2, and can be inferred from whether confidence intervals overlap indicator therefore correlates to changes in community and taxa abundance (Gislason et al., 2017). The responses of evenness Jʹ and Shannon-Wiener diversity Hʹ are driven by relative changes in abundance between different taxa, which depend on competitive and predatory interactions of the species in the community and differ between regions and environments (Svensson, Lindegarth, Jonsson, & Pavia, 2012), and this explains why the observed responses in our meta-analysis are not significant. Differences in the depletion per trawl pass between taxa (Sciberras et al., 2018) can also play a role, and are not easy to predict (Sciberras et al., 2018). Arctica islandica) can be very high and make up >50% of all biomass in some untrawled areas . When trawled or disturbed, such large species are often strongly reduced in biomass, thereby increasing evenness of the community (Kimbro & Grosholz, 2006) and substantially reducing total community biomass, while the magnitude of the reduction in numbers is modest. Other, generally smaller, taxa may benefit to some extent and increase due to a reduction in competition and/or predation, without fully compensating for the decrease in biomass of the (larger) sensitive species.
The observation that community numbers and biomass responded more strongly than mean taxa numbers and biomass suggests that such compensatory responses are weak. In the taxa level analysis, the effect of trawling on each taxon is equally weighted regardless of its contribution to community biomass, and this results in a smaller overall effect because the decrease in high-biomass sensitive species has a much smaller effect on the value of the indicator.
Biomass-based indicators capture effects on body size and age structure as well as numbers. These properties of the community affect the energy flow through food webs and other ecosystem processes, meaning that they are likely to correlate to the functioning of the ecosystem. Biomass-based indicators are also less likely to show sudden jumps in response to recruitment pulses that are unrelated to trawling, because even though recruits may be numerically abundant, they usually contribute very little to total biomass.
An unimpacted, and naturally functioning, benthic community in a stable environment has a size-, age-and longevity-distribution that is normally characterized by a large biomass of old and large biota (Hiddink et al., 2019;Rijnsdorp et al., 2018). Of the indicators considered, whole-community biomass is most likely to reflect the difference between this type of unimpacted community and one that is trawled.
The studies upon which our conclusions are based were obtained using a systematic review approach and therefore represent all globally available studies that satisfied the selection criteria, but they do not necessarily provide a balanced sample of all habitat × gear combinations. The conclusions drawn here are most applicable to the habitats represented in the underlying sources, although because of the general nature of the indicators examined, and the generality of the responses of seabed biota to trawling , there is no reason to assume that the general ranking of indicator performance would vary substantially among geographies and habitats. For these reasons we recommend community biomass as a globally applicable state indicator for monitoring and assessing the status of seabed biota impacted by trawling.
Applications would include measuring and reporting comparative seabed status in trawled and untrawled areas or across gradients of trawling intensity, and describing temporal changes in seabed status (e.g. rates of depletion or recovery following initiation or cessation of trawling).
A substantial amount of the observed variation in benthic states was explained in our analysis, and much of the remaining variation is likely to be due to variation in the actual trawling intensity at both control and impact locations, as well as variations in gear size, weight, selectivity and rigging. Other reasons for the large variation (indicated by 95% CI) around the observed means are the substantial spatial variation in abundance of benthic invertebrates at the scale of the sampling gear, and differences in environmental conditions between trawled and control areas that were not reported or may not have been appropriately controlled for in some studies. As a result, the statistical power to detect effects was low, and the environmental covariates that we tested only explained a significant amount of variation in three out of 70 covariate-indicator combinations. Some low-intensity trawling occurred at control locations in some studies, although at much lower intensities than at impact locations, and will potentially lead us to underestimate of the effect of trawling. Other factors contributing to a potential underestimate of the trawling effect include the history of fishing disturbance as depletion of community abundance will be higher in unfished areas relative to previously fished areas (Sciberras et al., 2018)

| Synthesis and applications
We Index are not responding to trawling (Gislason et al., 2017) and likewise should not be used as state indicators to describe the effects of trawling pressure.

TA B L E 3
Scoring of the candidate state indicators against each of the criteria described in Table 1. √ = 'meets criterion'. × = 'does not meets criterion'. Scoring of measurement, historical data, sensitivity, responsiveness and specificity are entirely or partly based on analyses presented in this paper. Other criteria as scored by consensus of the authors based on existing knowledge from the literature Theoretical basis √. Relevant to quantity of biota and links to pressure supported by models of trawl impacts (Hobday et al., 2011). Numbers of individuals by taxa to some extent linked to ecosystem function √. Relevant quantity of biota and links to pressure supported by models of trawl impacts (Pitcher et al., 2017). Biomass by taxa positively linked to functional role √. Relevant to quantity of biota and links to pressure supported by models of trawl impacts (Blanchard et al., 2009). Community numbers may be linked to functional role, but typically less strongly than biomass  (Sciberras et al., 2018). Present results and others show signal can be distinguished from noise and environment (e.g. Atkinson, Field, & Hutchings, 2011) ×. Less widely recorded than numbers (Sciberras et al., 2018). Some results (Link et al., 2005) (Gislason et al., 2017). Models not well established (Hiddink et al., 2006). Theory links species richness to functioning (Gamfeldt et al., 2015) √

DATA AVA I L A B I L I T Y S TAT E M E N T
The systematic review database is available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.bzkh1 895k .