Trawl exposure and protection of seabed fauna at large spatial scales

Trawling is the most widespread direct human disturbance on the seabed. Knowledge of the extent and consequences of this disturbance is limited because large‐scale distributions of seabed fauna are not well known. We map faunal distributions in the Australian Exclusive Economic Zone (EEZ) and quantify the proportion of their abundance that occurs in areas 1) that are directly trawled and 2) where legislation permanently prohibits trawling—defined as percentage exposure or protection, respectively. Our approach includes developing a method that integrates data from disparate seabed surveys to spatially expand predicted benthos distributions.

invertebrates (benthos) in nine regions. Our approach combines data from multiple surveys,  Results: Trawling is currently prohibited from more area of Australia's EEZ (58%) than is 53 trawled (<5%). Across 134 benthos-groups, 96% had greater protection of abundance than 54 exposure. The mean trawl exposure of benthos-group abundance was 7%, compared to mean 55 protection of 38%; whereas the mean abundance neither trawled nor protected was 55%. 56 Fishery closures covered 19% less study area than marine reserves, but overlapped with a 57 higher proportion (5% more) of benthos-group abundance. (benthos) help oxygenate the sea floor, break down organic material, provide habitat structure 83 and food sources for other organisms (Tagliapietra & Sigovini, 2010). Accordingly, benthos 84 are often used as indicators for assessing the status and health of marine ecosystems 85 (Rosenberg et al., 2004). From a human perspective, benthos support a range of commercial 86 industries (Hiddink et al., 2011;Choi & Joon Choi, 2012). However, many benthic species 87 are sensitive to disturbance; thus, the extent and intensity of human activity in marine 88 ecosystems can ultimately disrupt the services that benthos provide (Thrush & Dayton, 2002). 89 While the importance of benthos in marine ecosystems is recognized, their distributions and 90 extent of threats on them are largely unknown, particularly across large spatial scales.  Marine reserves and fishery closures are two management tools that are used to protect 105 species and habitats from human disturbance (Rice, 2005). Previously, marine reserve 106 designation was largely opportunistic (Roberts et al., 2003), but now systematic approaches 107 that take account of biota distributions may be used for planning spatial closures ( fishing is displaced to benthos rich areas (Pitcher et al., 2015). Thus, benefits for benthos 113 cannot be assumed, and distributions of benthic habitats and fauna should be assessed and 114 incorporated when planning spatial closures (Hiddink et al., 2006b).

116
Knowledge of the large-scale distribution of benthos is essential for impact assessments,   We aim to quantify, across large spatial scales, the proportion of benthos abundance currently 131 distributed in areas that are trawleddefined as exposureand in marine reserves or 132 fishery closure areas where legislation permanently prohibits trawlingdefined as 133 protection. Our analysis is based on benthos distributions predicted from seabed survey data. 134 We also develop approaches to utilize sparse and disparate datasets with the intention of 135 expanding the spatial extent of distribution mappingan approach that can be widely  Collating large-scale datasets across Australia 156 We collated available data across Australia's continental EEZ (9.14 million km 2 , Fig. 1 Table S1 in Supporting Information). These data were collected from 3200 sites by four gear   Trawl closed areas 180 All available data on the location of marine reserves/parks and fishery closures were collated 181 for the Australian EEZ (Table S3; Table S4). We examined each management and zoning 182 plan to include only spatial areas that permanently prohibit trawl fishing. All areas that 183 prohibit trawling were combined and mapped using ArcGIS (ESRI, 2014). We note that for 184 many Commonwealth Marine Reserves protection is planned but not yet in effect.

185
Environmental data 186 Environmental data for modelling and predicting the distribution of benthos comprised 37 187 environmental variables mapped to the Australian EEZ on a 0.01° grid (Table S5). Predictors 188 that did not vary among surveyed sites (SD=0) or were missing for parts of a region were 189 excluded from individual analyses.

192
Benthos distributions were modelled and predicted using Random Forests (RF), an ensemble 193 of decision trees with binary splits (Breiman, 2001). Analyses were implemented in the R 194 computing environment (R Core Team, 2015) using package 'randomForest' (Liaw & 195 Wiener, 2002). Importance of each predictor was calculated as the increase in Out-of-Bag 196 (OOB) mean squared error (MSE) when the values of the predictor were randomly permuted. 197 We used conditional importance as implemented in 'extendedForest' to take into account   (Table   208   S1). Each study region was bounded by the latitude, longitude and depth-range of surveyed 209 sites. Analyses for each region followed a three-step process: arranging data-matrices, 210 grouping taxa and predicting current benthos distributions using RF ( Fig. 2; Appendix S1 -R 211 code example). Step 1. Arranging data into a matrix 214 The RF analyses required a site-by-taxon matrix (biomass or count data) for each of the nine 215 regions. Three approaches based on the complexity of regional survey datasets were used to 216 produce the matrix (Appendix S1).  were joined together to provide a single hybrid matrix.

273
Step 2: Determining benthos groups 274 We aggregated taxa at taxonomic class level because of: taxonomic inconsistencies among  Shark Bay and lowest for Pilbara Coast (Fig. S1). The most important predictor across all 328 benthos models was sediment sand fraction ( Fig. S3; Fig. S4). Other important variables were Protection and exposure also ranged widely across all 134 benthos groups for which 358 distributions were predicted and mapped by this study (Table S7; Table S8 protection or exposure related to taxonomic classes (Fig. 6b). to trawling (mean=26.7%; Fig. 6a); yet, this region had comparably high protection 371 (mean=43.1%), primarily due to extensive fishery closures. In contrast, benthos in Pilbara

372
Coast had the least protection, but also low exposure to trawling. Benthos groups in the Great

373
Barrier Reef had the highest protection by marine reserves compared to other regions, but its 374 trawl fishery closures have been fully incorporated into its protected areas, so combined 375 protection of its benthos groups (mean=52%) was similar to that of several other study 376 regions (Fig. 6a).

381
The least variation occurred in the South East (min=33%, max=56%), and Pilbara Coast 382 regions (min=3%, max=26%). In all regions, variation in benthos trawl exposure was 383 considerably less than variation in benthos protection. This study provides the most extensive quantitative assessment of the current trawl exposure 399 and protection of Australia's benthic invertebrates. The exposure of most Australian benthos 400 to trawling was relatively low, whereas benthos protection was typically about 6-fold higher.

401
However, for most benthos groups more than half of their benthos abundance was neither  Table S6). Similarly, the national footprint of all Australian trawling as a 416 proportion of the entire EEZ (at ~1% trawled, or 4.7% area of grid-cells with effort) is 417 smaller than the trawl footprints in our study regions (at ~3% trawled, or 14% area of grid-418 cells with effort; Table S6). Thus, the high proportion of area protected at the EEZ scale 419 cannot be assumed at regional scales, where local protection and risks must be quantified to trawl footprints for that region in our study. Second, it is implicit that some details observed 478 by finer-scale studies will not be picked up by a large, cross-regional study. Moreover, we 479 aggregated benthos into groups, which would inherently introduce additional uncertainty 480 compared with species-level analyses (Pearman et al., 2010). Nevertheless, our broad cross-481 regional finding that trawl exposure was low and protection was high, is consistent with ). Third, modelling and predicting regional benthos distributions will always introduce 484 uncertainty due to sampling variability/error in source data, imperfect relationships between 485 benthos and environment and biological/ecological processes among others. For these 486 reasons, we report the OOB prediction performance of benthos models ( Fig. S1; Fig. S2 In conclusion, we discovered greater proportions of benthos abundance in our study regions 493 were distributed in protected and/or closed areas rather than in trawled areas. Our study also 494 highlights the importance of fishery closures in providing protection for benthic invertebrates. organizations for salary support. We are thankful to data contributors (Tables S1, S2, S3, S4   510 and S5). Benthic survey data sources are provided in Table S1. Trawl effort data are confidential and 514 source information is in Table S2. Marine reserve and Fishery closure data are available from 515 sources provided in Table S3 and Table S4. Environmental predictor data are available from 516 sources provided in Table S5. Zone EEZ (see Table S3; Table S4). Step 1: Arrange data into a matrix Step 2: Determine benthos groups

Multivariate Regression Trees
Group sites

DLI metric Assign taxa to site groups
Aggregate abundance Sum taxa abundance per group

Random Forest Models
Predict spatial abundance distributions of benthos groups