Habitat structural complexity metrics improve

A c c e pte d Ar tic le

Habitat structural complexity metrics improve predictions of fish abundance and distribution

Renata Ferrari1,2,7, Hamish A. Malcolm3, Maria Byrne1,4,7, Ariell Friedman2,7, Stefan B. Williams2,7, Arthur Schultz6, Alan R. Jordan5, Will F. Figueira1,7

1

The University of Sydney, School of Life and Environmental Sciences, A11, NSW 2006, Australia

2

The University of Sydney, Australian Centre for Field Robotics, School of Aerospace, Mechanical and

Mechatronic Engineering, J04, NSW 2006, Australia 3

NSW Department of Primary Industries, PO Box 4297 Coffs Harbour, NSW 2450, Australia

4

The University of Sydney, School of Medical Sciences, A11, NSW 2006, Australia

5

NSW Department of Primary Industries, Locked Bag 1, Port Stephens, NSW 2315, Australia

6

Southern Cross University, Southswell Marine, 204 Schnapper Beach Road, Urunga, NSW 2455, Australia

7

Sydney Institute of Marine Science. Building 19, Chowder Bay Road, Mosman NSW 2088, Australia

Corresponding Author: Renata Ferrari, The University of Sydney, School of Life and Environmental Sciences, A11, NSW 2006, Australia. E-mail: [email protected]

Decision date: 30-Aug-2017

This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may

lead to differences between this version and the Version of Record. Please cite this article as

doi: [10.1111/ecog.02580].

‘This article is protected by copyright. All rights reserved.’


ABSTRACT Habitat structural complexity influences biotic diversity and abundance, but its influence on marine ecosystems has not been widely addressed. Recent advances in computer vision and robotics allow quantification of structural complexity at higher-resolutions than previously achieved. This provides an important opportunity to determine the ecological role of habitat structural complexity in marine ecosystems. We used high-resolution three-dimensional (3D) maps to test multiple structural complexity metrics, depth and benthic biota as surrogates of fish assemblages across hundreds of meters on subtropical reefs. Nonparametric multivariate statistics were used to determine the relationship between these surrogates and the entire fish assemblage. Fish were divided into functional groups, which were used to further investigate the relationship between surrogates and fish abundance using generalized linear models. Fish community composition and abundance were strongly related to habitat complexity metrics, benthic biota and depth. Surface rugosity and its variance had a significant positive influence on the abundance of piscivores and sediment infauna predators, and a negative effect on the abundance of predators, herbivores, planktivores and cleaners. Final models for fish functional groups explained up to 68% of the variance. The best metrics to

explain the variance in fish abundance were benthic biota (25 ±7.5% of variance explained, mean ±SE) and complexity metrics (16 ±6.6%, mean ±SE). Our results show that high-resolution 3D maps and derived

metrics can predict a large percentage of variance in fish abundance and potentially serve as useful surrogates of fish abundance across all functional groups in spatially dynamic reefs.

Keywords: 3D topographic maps; benthic biota; species distribution models; autonomous underwater vehicles; surface rugosity; slope; habitat; 3D models.



INTRODUCTION Habitat structural complexity, hereafter referred to as complexity, is the physical three-dimensional

(3D) structure of an ecosystem. Complexity provides a suite of resources to organisms and is linked to species abundance (Graham and Nash 2013), and it influences fish presence and abundance because it provides shelter and physical habitat for benthic assemblages (Harborne et al. 2011, Ferrari 2017). Complexity also has a strong relationship with important ecological processes, such as foraging and predation (Macarthur and Macarthur 1961, Pickett and Cadenasso 1995). The relationship between fish abundance and complexity may be influenced by many factors,

including body size (Harborne et al. 2011), home range, fine-scale mobility (Harborne et al. 2012), trophic preference (Rizzari et al 2014) and life history traits (Bozec et al. 2013). Functional groups differ in their habitat requirements and preferences, and so their presence and abundance can be strongly influenced by habitat at different spatial extents (Curley et al. 2002, Chittaro 2004). For instance, large piscivores are likely to range further than other fishes, but may be attracted to complex habitat due to prey abundance; while herbivores or planktivores may prefer less complex areas where they can forage and monitor their immediate environment for predators (Rilov et al. 2007). In the context of fish, ecologically relevant extents are those that influence key ecological processes,

such as predation. Fish assemblages can change substantially across habitats and spatial extents, even across tens of meters in reefs (Vanderklift et al. 2007, Schultz et al. 2012, Fulton et al. 2016). Important ecological processes, such as herbivory, can vary spatially at sub-meter resolution (Ferrari et al. 2012b). Thus, it is important to encompass broad extents at high-resolution to better understand how habitat features spatially influence fish assemblages (Grabowski 2004). Marine protected areas (MPAs) are used extensively to provide spatial protection of biodiversity

from extractive activities (Guidetti and Sala, 2007, Lester et al. 2009, Babcock et al. 2010). However, effective representation of biota is often constrained by a paucity of spatially-explicit data across the planning area (Williams and Bax 2001, Stevens and Connolly 2004). Habitat features, such as depth, are more readily surveyed than fine-scale biotic features and are mapped at spatial extents suitable to management (Jordan et al. 2005, Ierodiaconou et al. 2007). Thus, the use of habitat features as surrogates for species abundance



enables representation of biodiversity and functional groups in MPAs (Moore et al. 2011, Ferrari et al. 2012a, Guisan et al. 2013), providing that their relationships have been quantified and are strong and predictive (Ferrier 2002). There is evidence for the influence of habitat features on fish abundance and diversity (Friedlander et al. 2003, Gratwicke and Speight 2005a, Gratwicke and Speight 2005b). In marine ecosystems fish abundance is strongly related to benthic biota and depth (Osmundson et al. 2002, Chong-Seng et al. 2012, Fitzpatrick et al. 2012). Yet, precisely measured estimates of habitat structural complexity have only recently been explored as a potential surrogate of fish abundance and richness (Rees et al. 2013). Although airborne laser-scanning and aerial images are used to obtain high-resolution 3D metrics of

complexity on land (Goetz et al. 2010, Zellweger et al. 2013, Eldegard et al. 2014), repeatable and highresolution measurement of complexity in marine and freshwater systems has proven difficult (Graham and Nash, 2013). The use of swath acoustics to derive digital elevation models with horizontal spatial resolutions typically on the order of a few squared meters (Cameron et al. 2014) can allow the estimation of various parameters including slope, aspect, curvature and surface rugosity (the ratio between surface area and planar area) across broad spatial extents (Rattray et al. 2009). Recent advances in computer vision also allow the

quantification of habitat structural complexity of the seafloor across large spatial extents and at centimeter resolution (McKinnon et al. 2011, Williams et al. 2012, Burns et al. 2015, Leon et al. 2015, Figueira et al. 2015, Ferrari et al. 2016a). This provides an important new approach to investigate the role of complexity and related metrics on fish communities and marine ecosystems at very fine scales (Harborne et al. 2012, Ferrari et al. 2016b). The aim of this study was to investigate the relationship between spatial variation in fish abundance

and multiple high-resolution metrics of habitat structural complexity. To do so, we developed 3D terrain reconstructions of 3 x 3 cm-resolution for 625 m2 plots. Four habitat complexity metrics were computed for each plot: variance and maximum value of slope and surface rugosity (SR). We investigated the explanatory power of these metrics along with epibenthic biota and depth as surrogates of fish abundance and community composition on subtropical reefs. These relationships were investigated for the entire fish assemblage and for trophic sub-groups. We anticipated finding a significant relationship between structural complexity and fish abundance for all trophic sub-groups, but not for overall fish abundance (Rees et al. 2013). We hypothesized



that the effect of habitat structural complexity could be positive or negative, depending on the fish trophic sub-group (Alvarez-Filip et al. 2011) and the metric of structural complexity tested (Kelaher et al. 2014). Final models were used to predict fish abundance.

METHODS Study sites This study was conducted in the Solitary Islands region, New South Wales, Australia (30°12’18.10’’S,

153°13’10.71’’E), an area with extensive rocky reef systems in a subtropical-temperate transition zone (Fig.

1); (Malcolm et al. 2011). Depth has a strong influence on dominant sessile epibenthic communities on the studied reefs, with rigid scleractinian coral-dominated communities above 25 m depth, and sponge or macroalgal dominated communities below 25 m (Jordan et al. 2010). Sessile communities in the region are strongly mediated by environmental habitat traits (Sommer et al. 2013). The reef fish community in the study area is diverse, and shows strong assemblage patterns with

depth and distance from shore due to the increasing influence of the warm Eastern Australian Current further offshore (Malcolm et al. 2010). There is a strong separation between shallow fish assemblages above and below 25 m depth and an additional separation in fish assemblages at depths of 50 m (Malcolm et al. 2011) likely associated with physical-environmental factors, including strong upwelling-driven processes (Armbrecht et al. 2014, Rossi et al. 2014). To disentangle the effects of complexity, we have investigated intermediate-depth reefs (25 – 50 m) on the inner continental shelf thereby reducing confounding influences

associated with depth and/or distance from shore. A mix of subtropical and temperate fish species, representing different functional groups, dominate these reefs (Malcolm et al. 2011). Benthic assemblages are dominated by sponges or encrusting organisms, with almost no large-canopy forming macroalgae that could potentially affect the ability to quantify biota using cameras (Fig. 1). Experimental design We surveyed six 625 m2 plots (25 x 25 m) at each of four locations (40 Acres, Split Bommie, The

Patch and South, Fig. 1). Fish relative abundance at each plot was assessed using Baited Remote Underwater Video (BRUV) systems (Harvey et al. 2012). 3D terrain reconstructions and benthic community composition



of each plot were obtained from images captured by an autonomous underwater vehicle (AUV) equipped with stereo-cameras, two strobes and navigation system (Friedman et al. 2012). Depth was measured by the AUV for each image taken and averaged across each plot; and any given individual plot had a maximum depth range of 5 m. Fish might be attracted from areas beyond a 625 m2 plot, and this experimental design assumes that the 6 plots represent the range of structural complexity and associated benthic and fish assemblages in a given location (6 plots per location). Locations in this study were at least 5 km apart and separated by sand. Response and explanatory variables Fish assemblages We surveyed fish assemblages in October 2012 using BRUV, a widely used method suitable for

comparing fish-habitat associations. BRUV systems are effective at detecting a range of species in the overall

community from various functional groups and with different habitat associations (Cappo et al. 2003). BRUV can effectively detect fish species attracted to the bait, species attracted to activity around the bait, species passing through, and species present in the vicinity (Cappo et al. 2003, Cappo et al. 2004). None-the-less, BRUV has limitations; in the context of this study BRUV may fail to detect groups of fish not attracted to the bait and also very small cryptic and benthic associated fish (Harvey et al. 2007). BRUV deployments consisted of a digital video-camera with wide-angle lens (Malcolm et al. 2011).

Plastic mesh bait bags with 1 kg of mashed pilchard, Sardinops neopilchardus (Steindachner 1987), were attached to the end of a bait-pole at 1.5 m from the lens (Hardinge et al. 2013). Deployments were 30 minutes in duration (Watson et al. 2007, Harasti et al. 2015) and video-files were analyzed using EventMeasure (Seager, 2008) for the maximum count of individuals per species (MaxN). The field-of-view was standardized

to a distance of approximately 3 m behind the bait, estimated during analysis by two trained observers. Data were robust to variability in estimating the field-of-view, as the counts reflect relative abundance (Cappo et al. 2003). A variable that is strongly related to fish abundance across functional groups is more useful as a

surrogate than a variable that is only related to one functional group. To test the usefulness of habitat complexity metrics as surrogates of fish abundance we followed Stuart-Smith et al. (2014) and identified four functional traits which describe the feeding (trophic), behaviour (gregariousness) and habitat preference



(mobility and position in water column) of each species (See table S1). Of all possible combinations of traits, we then identified seven functional groups, which were ecologically relevant (Table 1). Fish species were allocated to a sub-group within each functional group according to their functional placement based on detailed knowledge of the species (Malcolm et al. 2011) and a taxonomic reference guide (Kuiter, 2000). The total abundance within each sub-group was then calculated by pooling data according to species memberships. For the purpose of this study invertebrate predators were divided into those that prey on hard substratum reef invertebrates, and those that feed on infauna in sediments amongst or adjacent to reef (hereafter respectively referred to as predators and sediment infauna predators). Sessile Epibenthic assemblages The sessile epibenthic community was categorized from AUV imagery. We used a generalized random

tessellated stratified design in R package spsurvey (Kincaid and Olsen, 2013) to select 50 spatially balanced images (each 1.8 m2) per plot. Each image was analyzed using Coral Point Count with Excel extension (Kohler and Gill, 2006) and the random point count method (25 points per image). Biota were classified to a morpho-family level according to the Collaborative and Automated Tools for Analysis of Marine Imagery classification scheme (CATAMI) version 1.2 (Fig. S1 in Supporting Information); (Althaus et al. 2015). Relative abundance of biota for each plot was quantified as the average percent cover from the 50 images. A total of nine epibenthic categories were used in this study: stony corals, octocorals, hydrocorals, calcareous encrusting macroalgae, large canopy forming macroalgae, other macroalgae, sponges, bryozoans and ascidians. Habitat structural complexity metrics 3D terrain reconstructions were developed using the stereo-cameras and several habitat structural

complexity metrics were computed as in Friedman et al. (2012). In summary, being a visually-based technique, it incorporates all relief into the measure, including biotic and abiotic structure; for example a bare

boulder would have a lower value of habitat structural complexity than a boulder with kelp or sponges growing on it. Surface rugosity (SR) is computed by dividing the surface area of the 3D terrain by the area of the terrain

projected onto a flat plane, which can be the plane of best fit, or the horizontal plane. A high value of SR



indicates a highly complex habitat, while a low value indicates low complexity. If divided by the plane of best fit (SRpbf), the obtained metric is decoupled from the effect of slope, which is important because area of the plane of best fit increases with increasing slope. Alternatively, if one wishes to incorporate the effect of slope and SR into a single metric, then dividing by the horizontal plane might be preferable (SRhp). This means that the SRpbf of a reef on a 10° slope would be the same as the SRpbf of the same reef on a 60° slope, but the SRhp of the same reef would be smaller for the 10° slope example than for the 60° example. This study separately tested the relationship between fish abundance and each of these two SR metrics; and when SRpbf was used, slope was added to the model as a separate explanatory variable. We were interested in the effect of SR on fish abundance and its variation across various fish functional

groups. The rock framework provided the broad (> 100 m2) SR in the studied reefs, while sessile benthic communities contributed to the variance in SR at finer extents (0 - 10 m2). Thus, we also computed and tested two other complexity metrics for each plot, the variance and the maximum value of SR (both types) and slope (only used with SRpbf). The SR (both SRpbf and SRhp) and slope were computed for a central 400 m2 (20 x 20 m) area within each 625 m2 plot. The variance and maximum SR and slope were computed from SR estimates from sixteen non-overlapping 25 m2 (5 x 5 m) windows within each plot.

Statistical analyses The analysis is structured around two fundamental models, each including depth and benthic

assemblages. The two models test either the effect of (1) SRhp alone, or (2) SRpbf and slope. Results were very similar and we only report the results of the best-fitting model in the interest of succinctness (SRpbf and slope). Some of the explanatory variables were collinear (|r| < 0.7, Dormann et al. 2013) and thus we developed a set of nested models for each of the fundamental models. This process resulted in eight models, which were tested during the multivariate analyses for each of the seven functional groups (total of 56 models, Table S2). This approach resulted in nine metrics of structural complexity used as explanatory variables, three tested inter-plot variability of complexity: (1) SRhp, (2) SRpbf and (3) slope; while six tested intra-plot variability of complexity: (4) SRhp variance (VarSRhp), (5) SRhp maximum value (MaxSRhp), (6) SRpbf variance (VarSRpbf), (7) SRpbf maximum value (MaxSRpbf), (8) slope variance (VarSlope) and (9) slope maximum value (MaxSlope).



The relationship between fish abundance and habitat complexity is described by first presenting the

multivariate analyses for both the entire fish assemblage and the functional groups (Table 1). The second section focuses on univariate analyses of six species identified during the multivariate analyses by their strong relationship to complexity and with sufficient abundance to explore further. The second section also reports the univariate analyses results on the relationship between habitat complexity and both total fish abundance and abundance by trophic sub-group (Fig. 2). Multivariate Analyses across functional groups Multivariate generalized linear models (GLM) were run using package mvabund in R (Yi Wang et al.

2013) to investigate the relationship between multivariate fish abundance (MaxN), for each of the seven fish functional groups in Table 1, and three types of explanatory variables: complexity metrics, depth, and benthic biota. All models used a negative binomial distribution to account for over-dispersion, and 999 pit-trap resampling iterations to account for correlation in testing and to calculate model term p-values (Yi Wang et al. 2013). Provided that some of the remaining explanatory variables were still mildly collinear (|r| = 0.1 – 0.6), correlation type “shrink” was used to account for correlation (Warton et al. 2012). Models were simplified following the parsimony principle and simpler models were tested using analyses of variance on loglikelihood ratios to determine the best model. Given the sample size (n=24), first the nine habitat complexity metrics were sequentially tested and the best model was selected according to wald-test statistic (from the 8 possible models listed in Table S2). Second, the nine epibenthic biota categories were sequentially tested and only significant ones were retained. Third, depth was added to the reduced model and tested (Crawley, 2012). Univariate Analyses of trophic sub-groups, fish species and overall fish abundance The relationship between surrogates (complexity metrics, depth and benthic percent cover) and fish

abundance was further investigated using 13 univariate models: one for the overall fish abundance, six for each trophic sub-group, and six for each species highlighted by the multivariate analyses (Prionurus microlepidotus, Parupeneus spilurus, Chaetodon guentheri, Gymnothorax prionodon, Ophthalmolepis

lineolatus, Notolabrus gymnogenis). Models included the interaction between depth and complexity metrics. We used general and generalized hierarchical (GLM/GLMM) models in R packages lme4 (Bates et al. 2013) and mass (Venables and Ripley, 2002).



The choice and type of model was informed by the need to account for (i) random effects of location,

(ii) potential zero inflated or over dispersed data, and (iii) spatial autocorrelation. Location was tested as a random effect in the original models to account for potential spatial autocorrelation; if no significant (p > 0.05) improvement of model fit resulted from including location, random effects were removed from the models (Bates, 2010). Models were simplified using backward selection following the parsimony principle, and tested using Chi-square test on likelihood ratios and AIC to determine the best model (AIC only for GLMs). Assumptions of homoscedasticity and normality (where relevant) were checked using residual plots (package graphics R-Core-Team, 2013) and the Fligner-Kileen non-parametric test (Crawley 2012). Assumptions of independence (to ensure no autocorrelation was present) were checked using semivariograms. Final models were used to predict the abundance of each trophic sub-group as a function of habitat complexity (Fig. 2). Trophic sub-groups were carefully considered, and very abundant species, where present, were removed from sub-groups and models were re-run to ensure that one abundant species was not driving the patterns for a given sub-group.

RESULTS A total of 74 fish species were observed in this study, representative of fish assemblages of

subtropical intermediate-depth reefs of eastern Australia (Malcolm et al. 2011) (Fig. 3). Within trophic subgroups predators were the most species rich (41 species, 55% of surveyed species), followed by piscivores (13

species, 18%), planktivores (9 species, 12%), sediment infauna predators (7 species, 9%), herbivores (3 species, 5%) and cleaners (1 species, 1%). Many of these species were only recorded once or twice, so their individual relationship with habitat complexity could not be determined.

Does habitat structural complexity influence fish assemblage? Multivariate analyses for the entire fish assemblage surveyed indicated that habitat complexity

metrics, depth and several benthic taxa were significantly related to the fish assemblage (Table 2). Eleven species were strongly and significantly related to complexity metrics: one piscivore (Seriola sp), one herbivore (Prionurus microlepidotus), one sediment infauna predator (Parupeneus spilurus), and 8 predators (Chaetodon guentheri, Gymnothorax prionodon, Ophthalmolepis lineolatus, Notolabrus gymnogenis,



Orectolobus halei, Brachaelurus waddi, Bodianus unimaculatus and Bodianus frenchii).The best multivariate model included the slope, VarSlope and MaxSRpbf (Table 2). This model did not differ significantly from

other models including SRpbf (model difference in wald test-statistic < 2), but it was significantly better than any models including SRhp. Thus, SRhp was not investigated further. Patterns were consistent across functional groups with the same complexity metrics being most

important across all 56 multivariate models (Fig. 2c, Table S3). The trophic group had the smallest confidence interval among all models (Fig. 2d), hence it was selected to perform the univariate analyses. When adding depth and benthic sessile biota to the final model for each fish functional group, the increase in model fit varied from 1 (fish spp) to 16 (Gregarious-mobility group) wald-units (Fig. 2d), suggesting that the

importance of benthic biota and depth varied across fish funtional groups (Fig 2, Table 2, Table S3). The univariate analyses revealed that total fish abundance significantly decreased with depth, but

increased with the relative abundance of stony corals; these two variables alone explained 28% of its variance (Table 4). Yet, total fish abundance was not significantly related to any of the nine metrics of structural complexity investigated (Table 4). Does habitat structural complexity influence fish species occurrence and abundance? Habitat structural complexity significantly influenced the abundance of individual fish species (Table

3, Fig. 4). Only six of the 11 species that were strongly related to habitat structural complexity were sufficiently abundant to model individually. The probability of occurrence of both P. microlepidotus (herbivore) and C. guentheri (predator) significantly decreased with variance in SRpbf but was not influenced by other predictors. In contrast, the probability of the occurrence of G. prionodon (predator) increased with the variance in SRpbf, but decreased with overall slope. Slope influenced the abundance of P. spilurus (sediment infauna predator) negatively, yet the abundance of this species increased with VarSlope. The MaxSlope was negatively and significantly related to O. lineolatus (predator). Finally, the abundance of N. gymnogenis (predator) significantly decreased with depth (Table 3, Fig. 4b). Does habitat structural complexity influence trophic sub-groups occurrence and abundance? Piscivore abundance increased with SRpbf, and was influenced by three benthic taxa (stony corals,

octocorals and macroalgae) (Table 4, Fig. 4a, Fig. 5). Predator abundance was negatively related to the



interaction between maximum SRpbf and depth, indicating that the negative effect of depth changed with

habitat structural complexity (Table 4, Fig. 4a, Fig. 5). Interestingly, sediment infauna predator abundance increased with SRpbf, but decreased with slope and depth. Five benthic categories also influenced their abundance significantly, three positively (octocorals, hydrocorals and ascidians) and two negatively (sponges and unknown benthic biota) (Table 4, Fig, 4a, Fig. 5). Herbivore abundance decreased significantly with the variance in slope, depth, large canopy macroalgae and bryozoans (Table 4, Fig. 4a, Fig. 5). Patterns in planktivore abundance were driven by mid-water planktivores, and significantly decreased with the VarSRpbf, ascidians and bryozoans (Table 4, Fig. 4a, Fig. 5). Cleaners’ abundance was positively related to the VarSlope but negatively related to SRpbf and sponge abundance. Interestingly, 32% of the variance in cleaner fish

abundance was positively associated to sponge abundance (Table 4, Fig. 5). Cleaner wrasse often associate with tall structures (unpublished data), in the surveyed reefs sponges and corals provide height on top of the rocky reef relief. However, sponges were more abundant than corals and thus we hypothesize that cleaner fishes had a strong association with sponges given the shelter these benthic organisms might provide.

DISCUSSION Habitat complexity strongly influenced fish community composition and fish abundance. Habitat

complexity metrics explained a large proportion of the variance in abundance of six individual species and in the abundance of fish in all trophic sub-groups. Fish community composition and abundance was also significantly related to benthic biota and depth. The final models for the trophic sub-groups explained a larger proportion of the variance (28 – 68%) in fish abundance than the models for individual species (18 - 32%). A mosaic of reef habitats with different complexities will contribute to a more complex fish community and their abundance than one that is not complex or has uniform high-complexity. The influence of habitat complexity metrics varied in direction and strength by sub-group and by

species, complementing findings from other studies (Rees et al. 2014). Therefore, it is not surprising that overall fish abundance was not significantly related to structural complexity metrics, and that it was necessary to investigate this relationship through the trophic sub-groups. For instance, the abundance of higher trophic sub-groups (i.e. piscivores) increased with surface rugosity and slope. The abundance of lower trophic sub-



groups (i.e. herbivores) decreased with the variance and/or maximum value of surface rugosity across the reef. While other functional groups were investigated during the multivariate analyses (Table 1), the effect of habitat complexity metrics was consistently important across functional groups, highlighting the value of habitat complexity as a surrogate of fish abundance. The influence of benthic biota on fish abundance Benthic biotic microhabitat has been found to influence fish abundance (Fulton et al. 2016). In this

study, benthic biota explained 9% more of the variance in fish abundance than complexity metrics (25 % and 16% respectively), but did not influence all trophic sub-groups. An important consideration is the significant post-processing required to manually quantify benthic assemblages from imagery. However, when data on benthic biota are available, we strongly suggest that these data are incorporated into distribution models as the resulting explanatory power increases substantially and thus the reliability of model predictions is strengthened. The influence of depth on fish abundance Depth was significantly related to total fish abundance, and explained 28% and 36% of the total

variance in herbivores and predators respectively, but only 0 - 3% for all other trophic sub-groups. This study examined fish assemblages between 25 and 50 m depth, it is probable that depth has a stronger influence on fish abundance across a wider depth range (Malcolm et al. 2011). Depth is a metric that can be incorporated into species distribution models easily, so has considerable utility in spatial planning. Depth and habitat structural complexity are not mutually exclusive and the application of both is likely more useful than each individually for management planning purposes. Habitat structural complexity complements depth, especially for those species or functional groups less influenced by depth or where depth and habitat complexity interact to influence fish abundance (i.e. predators).

The influence of habitat structural complexity on fish abundance Habitat complexity is a key driver of species abundance (Pickett and Cadenasso, 1995, Harborne et al.

2011, Rees et al. 2014). The broad importance of habitat complexity was supported by this study, where abundance of all trophic sub-groups was influenced by surface rugosity or slope. The positive influence of surface rugosity on the abundance of piscivores and sediment infauna predators may relate to the influence on



their prey. Piscivores may benefit from surface rugosity because it increases the availability of their prey; while sediment infauna predators may find a habitat with more patches of sand around reef edges has higher abundance of their prey, compared to sandy areas further away from reefs, and also provides quicker access to shelter (Schultz et al. 2012). The negative influence of surface rugosity on herbivores likely relates to advantages of grazing in more open flatter terrain, both for turf algal growth and detection of predators (Rilov et al. 2007). Worthy of mention is the opposing directions of the effect of slope and surface rugosity (or their

variance) on two trophic sub-groups; and thus the importance of separating both metrics (surface rugosity and slope) rather than incorporating them into one (SRhp). For example, the abundance of sediment infauna predators increased with surface rugosity but decreased with slope, while cleaner fish abundance had the opposite relationship to these habitat metrics. Planktivores decreased with the variance in surface rugosity; perhaps they are unable to detect their predators in time to escape successfully in highly complex environments (Rilov et al. 2007). These results reveal the importance of considering trophic sub-groups and multiple metrics of habitat complexity to better understand their relationship. Five of the six species investigated were also significantly influenced by habitat complexity.

Crimson-banded wrasse, the sixth species, was only influenced by depth. The variance and maximum value of complexity metrics across a plot seemed more relevant, explaining 18 – 32% of variance in fish abundance/presence, than surface rugosity or slope, which explained only 5 – 10%. Although the findings were constrained by the BRUV method used to survey fishes, which has a bias towards fishes that are attracted to bait or commotion, a variety of species across a broad range of functional groups are typically represented by BRUV sampling (Cappo et al. 2003, Harvey et al. 2007). Fish operate on a range of spatial extents and there are multiple factors that affect fish-habitat

associations (Fulton et al. 2016). Johnson et al. (2013) found that most studies of demersal fish habitat have used large spatial extents, and suggest that there is little or no justification for this. They conclude that studies need to apply a greater focus to smaller spatial extents when considering abiotic and biotic habitat variables, to advance our understanding and management of demersal fish communities. Our results demonstrate that



surface rugosity measured at high-resolutions is meaningful for all the trophic sub-groups and across the entire fish assemblage surveyed. Previous studies examining the relationship between habitat complexity and fish abundance have

quantified complexity using remote sensing (Roelfsema et al. 2013) and/or hydro-acoustic mapping (Brown et al. 2011, Bejarano et al. 2011, Cameron et al. 2014), and some surveyed fish using BRUV (Moore et al. 2011, Rees et al. 2014). Their findings with respect to the relationship between surface rugosity and fish abundance are equivocal; perhaps because metrics related to surface rugosity at cm-resolution were not estimated, or because fish were not divided into functional groups as discussed by Rees et al. (2013). To the best of our knowledge, this is the first study to quantify habitat complexity at this resolution (3 cm cell) at spatial extents covering 100s of meters, and to test it as a surrogate of fish abundance with this level of confidence (27% of total variance explained for fish species and 11% for trophic sub-groups). Management implications and conclusions Planning effective MPAs can be improved through knowledge of the relationships between species and

habitat attributes, especially in the absence of detailed ecological data (Ward et al. 1999, Malcolm et al. 2016). Understanding relationships between habitat complexity and fish assemblages, as surrogates for biodiversity patterns, will benefit management planning, especially as high-resolution seabed terrain data become available (Ierodiaconou et al. 2007, Todd and Kostylev, 2011). High-resolution data is capable of capturing heterogeneity across multiple spatial extents and therefore should be incorporated into spatial ecology and conservation planning, as suggested in other studies (Vanderklift et al. 2007, Valentine et al. 2007, Schultz et al. 2012). This study demonstrated that high-resolution surface rugosity is a useful measure that explained on average 16% of the variance in fish abundance, whereas the mean amount of variance explained by the main predictor of interest in ecological studies is typically between 2.5 – 5.4% (Møller and Jennions, 2002). Thus, in combination with other important predictors, such as depth and benthic biota, habitat complexity metrics can improve predictions of fish abundance, making it a potential surrogate of fish abundance. Currently, the cost of AUV deployments is too high to achieve high-resolution 3D maps of entire MPAs.

However, statistical extrapolation methods (i.e. species distribution models) can be used to take advantage of



small areas with high-resolution data (Johnston et al. 2015). In addition, ongoing advances in computer-vision 3D mapping, including photogrammetry and off-the-shelf diver-operated cameras, means that high-resolution habitat complexity metrics can now be quantified across spatial extents useful for conservation planning at a lower cost than previously possible (Hu He et al. 2012, Figueira et al. 2015, Burns et al. 2015, Lavy et al. 2015, Ferrari et al. 2016a). Importantly, metrics derived from techniques that operate at different spatial resolutions and extents, such as computer-vision and swath-acoustic mapping, can be combined; creating a hybrid solution applicable to management, making high-resolution surface rugosity available as a potential surrogate of fish abundance (Cameron et al. 2014). This study advances our knowledge of the relationship between fish abundance and multiple habitat structural complexity metrics, paving the way for future studies that might have access to high-resolution 3D data. It also informs research using alternate technologies (i.e. acoustic mapping) to investigate the relationship between habitat complexity and fish abundance at larger spatial extents and considering functional groups. A thorough understanding of how different metrics of complexity influence fish abundance is key for

developing appropriate predictive models and guiding future spatial management. Ultimately, spatial planning could benefit from incorporation of habitat structural complexity (and derived metrics) as a complementary surrogate of fish abundance because: (1) habitat complexity is strongly and significantly related to fish abundance across the entire assemblage (depth and, if possible, benthic biota data should also be incorporated into species distribution models), but (2) an increase in complexity does not translate into increased fish abundance, and thus (3) spatial planning should incorporate a mosaic of multiple habitat structural complexity levels. Information on complexity will also allow an improved understanding of the importance of this factor on reef fish community composition and abundance compared to MPAs zone effects.

ACKNOWLEDGEMENTS This project was funded by New South Wales Department of Primary Industries and Office of Environment and Heritage, the University of Sydney, the Sydney Institute of Marine Sciences (SIMS) and the Integrated Marine Observation System Autonomous Underwater Vehicle Facility. We also thank the Marine



Biodiversity Hub, a collaborative partnership supported through funding from the Australian Government’s National Environmental Research Program (NERP). SIMS contribution number XXXX.



REFERENCES Althaus, F., et al. 2015. A Standardised vocabulary for identifying benthic biota and substrata from underwater imagery: The CATAMI Classification Scheme. - PLoS One 10: e0141039. Alvarez-Filip, L., et al. 2011. Region-wide temporal and spatial variation in Caribbean reef architecture: is coral cover the whole story? - Global Change Biol. 17: 2470-2477. Armbrecht, L. H., et al. 2014. Phytoplankton composition under contrasting oceanographic conditions: Upwelling and downwelling (Eastern Australia). – Cont. Sh. Res. 75: 54-67. Babcock, R. C., et al. 2010. Decadal trends in marine reserves reveal differential rates of change in direct and indirect effects. – Proc. Nat. Ac. Sciences 107: 18256-18261.

Bates, D., et al. 2013. Lme4: Linear mixed-effects models using S4 classes - In: R package Bates, D. M. 2010. lme4: Mixed-effects modeling with R. - URL http://lme4. r-forge. r-project. org/book

Bejarano, S., et al. 2011. Predicting structural complexity of reefs and fish abundance using acoustic remote sensing (RoxAnn). – Mar. Biol. 158: 489-504. Bozec, Y. M., et al. 2013. Reciprocal facilitation and non‐ linearity maintain habitat engineering on coral reefs. - Oikos 122: 428-440. Brown, C. J., et al. 2011. Benthic habitat mapping: A review of progress towards improved understanding of the spatial ecology of the seafloor using acoustic techniques. – Est., Coast. and Shelf Science 92: 502-520. Burns, J. H. R., et al. 2015. Integrating structure-from-motion photogrammetry with geospatial software as a novel technique for quantifying 3D ecological characteristics of coral reefs. – Peerj:3. Cameron, M. J., et al. 2014. Understanding community-habitat associations of temperate reef fishes using fine-resolution bathymetric measures of physical structure. - Mar. Ecol. Prog. Ser. 506: 213-229. Cappo, M., et al. 2003. Potential of video techniques to monitor diversity, abundance and size of fish in studies of Marine Protected Areas. - In: J. P. Beumer, et al. (eds), Aquatic Protected Areas - what works best and how do we know? World Congress on Aquatic Protected Areas proceedings, pp. 455-464. Cappo, M., et al. 2004. Comparison of baited remote underwater video stations (BRUVS) and prawn (shrimp) trawls for assessments of fish biodiversity in inter-reefal areas of the Great Barrier Reef Marine Park. – J. Exp. Mar. Biol. Ecol. 302: 123-152.

Chevan, A. and Sutherland, M. 1991. Hierarchical partitioning. - The Am. Stat. 45: 90-96. ‘This article is protected by copyright. All rights reserved.’


Chittaro, P. M. 2004. Fish-habitat associations across multiple spatial scales. - Coral Reefs 23.2: 235-244. Chong-Seng, K. M., et al. 2012. The influence of coral reef benthic condition on associated fish assemblages. - Plos One 7: e42167. Crawley, M. J. 2012. The R book. - John Wiley & Sons. Curley, B. G., et al. 2002. Spatial and habitat-related patterns of temperate reef fish assemblages: implications for the design of Marine Protected Areas. – Mar. Freshwater Res. 53: 1197-1210.

Dormann, C. F., et al. 2013. Collinearity: a review of methods to deal with it and a simulation study evaluating their performance. - Ecography 36: 27-46. Eldegard, K., et al. 2014. Modelling bird richness and bird species presence in a boreal forest reserve using airborne laser-scanning and aerial images. - Bird Study 1-16.

Ferrari, R., et al. 2012a. Size matters in competition between corals and macroalgae. - Mar. Ecol. Prog. Ser. 467: 77-88. Ferrari, R., et al. 2012b. Interaction of herbivory and seasonality on the dynamics of Caribbean macroalgae. Coral Reefs 31: 683-692. Ferrari, R., et al. 2016a. Quantifying multiscale habitat structural complexity: a cost-effective framework for underwater 3D modelling. – Rem. Sensing 8: 113. Ferrari, R., et al. 2016b. Quantifying the response of structural complexity and community composition to environmental change in marine communities. - Global Change Biol. 22: 1965-1975.

Ferrari 2017. The hidden structure in coral reefs – Coral Reefs doi:10.1007/s00338-017-1540-6 Ferrier, S. 2002. Mapping spatial pattern in biodiversity for regional conservation planning: where to from here? – Syst. Biol. 51: 331-363. Figueira, W. F., et al. 2015. Accuracy and precision of habitat structural complexity metrics derived from underwater photogrammetry. – Rem. Sensing 7(12): 16883 – 16900. Fitzpatrick, B. M., et al. 2012. Habitat specialization in tropical continental shelf demersal fish assemblages. PLoS One 7: e39634. Friedlander, A. M., et al. 2003. Effects of habitat, wave exposure, and marine protected area status on coral reef fish assemblages in the Hawaiian archipelago. - Coral Reefs 22: 291-305.



Friedman, A., et al. 2012. Multi-scale measures of rugosity, slope and aspect from benthic stereo image reconstructions. - PLoS One 7: e50440. Fulton, C.J., et al. 2016. Microhabitat selectivity underpins regional indicators of fish abundance and replenishment. Ecological Indicators 70: 222-231 Goetz, S. J., et al. 2010. Lidar remote sensing variables predict breeding habitat of a Neotropical migrant bird. - Ecology 91: 1569-1576. Grabowski, J. H. 2004. Habitat complexity disrupts predator–prey interactions but not the trophic cascade on oyster reefs. - Ecology 85.4: 995-1004. Graham, N. and Nash, K. 2013. The importance of structural complexity in coral reef ecosystems. - Coral Reefs 32: 315-326. Gratwicke, B. and Speight, M. R. 2005a. The relationship between fish species richness, abundance and habitat complexity in a range of shallow tropical marine habitats. – J. Fish Biol. 66: 650-667.

Gratwicke, B. and Speight, M. R. 2005b. Effects of habitat complexity on Caribbean marine fish assemblages. - Mar. Ecol. Prog. Ser. 292: 301-310.

Guidetti, P. and Sala, E. 2007. Community-wide effects of marine reserves in the Mediterranean Sea. - Mar. Ecol. Prog. Ser. 335: 43-56.

Guisan, A., et al. 2013. Predicting species distributions for conservation decisions. – Ecol. Let. 16: 1424-

1435. Harasti, D., et al. 2015. Appropriate set times to represent patterns of rocky reef fishes using baited video. – J. Exp. Mar. Biol. Ecol. 463: 173-180.

Harborne, A. R., et al. 2011. Biotic and multi-scale abiotic controls of habitat quality: their effect on coralreef fishes. - Mar. Ecol. Prog. Ser. 437: 201-214. Harborne, A. R., et al. 2012. The effectiveness of different meso-scale rugosity metrics for predicting intrahabitat variation in coral-reef fish assemblages. – Env. Biol. of Fish. 94: 431-442. Hardinge, J., et al. 2013. A little bait goes a long way: The influence of bait quantity on a temperate fish assemblage sampled using stereo-BRUVs. - J. Exp. Mar. Biol. Ecol. 449 250-260.



Harvey, E. S., et al. 2012. Comparison of the relative efficiencies of stereo-BRUVs and traps for sampling tropical continental shelf demersal fishes. – Fish. Res. 125-126: : 108- 120. He, H., et al. 2012. Measuring reef complexity and rugosity from monocular video bathymetric reconstruction. – Proc. 12th International Coral Reef Symposium, 2012 Jul 13 (pp. 1-5). James Cook

University. Ierodiaconou, D., et al. 2007. Marine benthic habitat mapping using multibeam data, georeferencedvideo and image classification techniques in Victoria, Australia. – J. of Spat. Science 52: 93-104. Johnson, A. F., et al. 2013. Linking temperate demersal fish species to habitat: scales, patterns and future directions. - Fish and Fish. 14: 256-280. Johnston, A., et al. 2015. Abundance models improve spatial and temporal prioritization of conservation resources. – Ecol. Appl. 25: 1749-1756. Jordan, A., et al. 2005. Seabed habitat mapping in the Kent Group of islands and its role in marine protected area planning. – Aquat. Conserv. 15: 51-70. Jordan, A., et al. 2010. Seabed habitat mapping of continental shelf waters of NSW. - In: NSW Department of Environment Climate Change and Water Occasional Paper Series 209. Kelaher, B. P., et al. 2014. Changes in Fish Assemblages following the Establishment of a Network of NoTake Marine Reserves and Partially-Protected Areas. - PLoS One 9: e85825. doi:10.1371/journal.pone.0085825. Kincaid, T. M. and Olsen, A. R. 2013. Spsurvey: Spatial Survey Design and Analysis. . - In: R package version 2.6. Kohler, K. E. and Gill, S. M. 2006. Coral Point Count with Excel extensions (CPCe): A Visual Basic program for the determination of coral and substrate coverage using random point count methodology -Computers & Geosciences 1259-1269. Kuiter, R. H. 2000. Coastal fishes of south-eastern Australia. - In: Gary Allen Pty Ltd. Lavy, A., et al. 2015. A quick, easy and non-intrusive method for underwater volume and surface area evaluation of benthic organisms by 3D computer modelling. – Meth. in Ecol. and Evol. 6: 521-531.



Leon, J. X., et al. 2015. Measuring coral reef terrain roughness using 'Structure-from-Motion' close-range photogrammetry. - Geomorphology 242: 21-28. Lester, S. E., et al. 2009. Biological effects within no-take marine reserves:a global synthesis. - Mar. Ecol. Prog. Ser. 384: 33-46. Macarthur, R. and Macarthur, J. W. 1961. On Bird Species-Diversity. - Ecology 42: 594. Malcolm, H. A., et al. 2010. Biogeographical and cross-shelf patterns of reef fish assemblages in a transition zone. – Mar. Biod. 40: 181-193. Malcolm, H. A., et al. 2011. Testing a depth-based Habitat Classification System against the pattern of reef fish assemblages (15 - 75m) in a subtropical marine park. – Aqua. Conserv.: Mar. and Freshwater Ecosys. 21: 173-185. Malcolm, H.A., et al. 2016. Integrating seaflor habitat mapping and fish assemblage patterns improves spatial management planning in a marine park. Journal of Coastal Research 75: 1292-1296 McKinnon, D., et al. 2011. Towards automated and in-situ, near-real time 3-D reconstruction of coral reef environments. - In: OCEANS 2011. pp. 1-10. Møller, A. and Jennions, M. D. 2002. How much variance can be explained by ecologists and evolutionary biologists? - Oecologia 132: 492-500. Moore, C. H., et al. 2011. The effect of landscape composition and configuration on the spatial distribution of temperate demersal fish. - Ecography 34: 425-435.

Osmundson, D., et al. 2002. Flow-sediment-biota relations: implications for river regulation effects on native fish abundance. – Ecol. Appl. 12: 1719-1739.

Pickett, S. T. A. and Cadenasso, M. L. 1995. Landscape Ecology - Spatial Heterogeneity in EcologicalSystems. - Science 269: 331-334. Pinheiro, J., et al. 2013. Nlme: Linear and Nonlinear Mixed Effects Models. . - In: R package. R-Core-Team 2013. R: A language and environment for statistical computing. . - In: R Foundation for Statistical Computing, Vienna, Austria. Rattray, A., et al. 2009. Hydro-acoustic remote sensing of benthic biological communities on the shallow South East Australian continental shelf. – Est., Coast. and Shelf Science 84: 237-245.



Rees, M. J., et al. 2014. Abiotic surrogates for temperate rocky reef biodiversity: implications for marine protected areas. – Div. and Distr. 20: 284-296. Rilov, G., et al. 2007. Complex habitats may not always benefit prey: linking visual field with reef fish behavior and distribution. - Mar. Ecol. Prog. Ser. 329: 225-238. Risk, M. J. 1972. Fish diversity on a coral reef in the Virgin Islands. - Atoll Res. Bull. 153: 1-6. Rizzari, Justin R., et al. 2014. Not worth the risk: apex predators suppress herbivory on coral reefs. - Oikos 123.7: 829-836. Roelfsema, C., et al. 2013. Mapping coral reefs at reef to reef-system scales, 10s–1000s km2, using objectbased image analysis. – Int. J. of Rem. Sensing 34: 6367-6388. Rossi, V., et al. 2014. Seasonality of sporadic physical processes driving temperature and nutrient high‐ frequency variability in the coastal ocean off southeast Australia. – J. of Geoph. Res.: Oceans 119: 445-460. Schultz, A. L., et al. 2012. Effects of reef proximity on the structure of fish assemblages of unconsolidated substrata. - PLoS One 7: e49437. Seager, J. 2008. EventMeasure Version 2.04. - In: SeaGIS Pty. Ltd., Bacchus Marsh, Victoria, Australia. Available at http://www.seagis.com.au. Sommer, B., et al. 2014. Trait-mediated environmental filtering drives assembly at biogeographic transition zones. - Ecology 95: 1000-1009. Stevens, T. and Connolly, R. M. 2004. Testing the utility of abiotic surrogates for marine habitat mapping at scales relevant to management. – Biol. Conserv. 119: 351-362. Stuart-Smith, R. D., et al. 2014. Integrating abundance and functional traits reveals new global hotspots of fish diversity. - Nature 501: 539-542. Todd, B. J. and Kostylev, V. E. 2011. Surficial geology and benthic habitat of the German Bank seabed, Scotian Shelf, Canada. – Cont. Shelf Res. 31: S54-S68. Valentine, J. F., et al. 2007. Food web interactions along seagrass-coral reef boundaries: effects of piscivore reductions on cross-habitat energy exchange. - Mar. Ecol. Prog. Ser. 333: 37-50. Vanderklift, M. A., et al. 2007. Proximity to reef influences density of small predatory fishes, while type of seagrass influences intensity of their predation on crabs. - Mar. Ecol. Prog. Ser. 340: 235-243.



Venables, W. N. and Ripley, B. D. 2002. Modern Applied Statistics with S. Ward, T. J., et al. 1999. Selecting marine reserves using habitats and species assemblages as surrogates for biological diversity. – Ecol. Appl. 9: 691-698. Warton, D. I., et al. 2012. Distance‐ based multivariate analyses confound location and dispersion effects. Meth. in Ecol. and Evol. 3: 89-101. Watson, D. L., et al. 2007. Protection from fishing alters the species composition of fish assemblages in a temperate-tropical transition zone. – Mar. Biol. 152: 1197-1206.

Williams, A. and Bax, N. J. 2001. Delineating fish-habitat associations for spatially based management: an example from the south-eastern Australian continental shelf. – Mar. Freshwater Res. 52: 513-536. Williams, S. B., et al. 2012. Autonomous underwater vehicle monitoring of benthic reference sites. - IEEE Robotics and Automation Magazine 19: 73-84. Yi Wang, et al. 2013. Mvabund: statistical methods for analysing multivariate abundance data. - In: R package. Zellweger, F., et al. 2013. Remotely sensed forest structural complexity predicts multi species occurrence at the landscape scale. - Forest Ecol. and Management 307: 303-312. Zuur, A. F., et al. 2009. A protocol for data exploration to avoid common statistical problems. - Meth. in Ecol. and Evol. 1: 3-14.



Table Legends

Table 1. Fish functional groups used in this study. A total of four different functional traits were used to characterize each species based on behavioral, feeding and habitat-related characteristics. Each species was assigned a value for each functional trait (see Table S1 in Supporting Information for full list of species and assignments to sub-groups). Species were then grouped based on one or multiple combinations of these traits. Note that only combinations of traits that were ecologically relevant were explored in this study (not all possible combinations). The third column indicates the total number of species/sub-groups (i.e. piscivore is a sub-group within the trophic group) within each group and gives a few examples of member sub-groups.

Fish Functional groups

Description

Grouping within sub-groups (no. of species/groupings) Fish species (n = 74)

Species

Individual fish species

Trophic

Feeding preferences

Piscivore; predator; sediment infauna predator; herbivore; planktivore; cleaners (n = 6)

Mobility

Mobility and positioning in the water column

Mobility-Trophic

Combination of the mobility/positional and the trophic traits

Midwater; mobile bottom dwellers; sedentary bottom dwellers (n = 3) Piscivore-midwater; piscivore-mobile bottom dweller (n = 16)

Trophic-Gregarious

Combination of the gregariousness and the trophic traits

Piscivore-schooling, piscivore-small groups, piscivore-solitary (n = 14)

Mobility -Gregarious

Combination of the gregariousness and the mobility traits

Midwater-schooling (n = 10)

Mobility-TrophicGregarious

Combination of the mobility, the trophic and the gregariousness traits

Piscivore-midwater-school (n = 24)



Table 2. Summary of results for the permutational multivariate generalized linear model (GLM) with negative binomial distribution for fish species abundance (MaxN) at 24 plots in Solitary Islands region, New South Wales, Australia.

Coefficient

Model test-statistic: 9.114, p-value: 0.001 Wald value

Pr(>wald)

Intercept

6.484

0.018

SRhp

6.287

0.030

VarSRhp

7.173

0.006

Depth

9.591

0.001

Corals

7.519

0.004

Ascidians

7.405

0.003

Notes: the wald-value is the test statistic used for model selection, the higher it is the more variance explained by the model. SRhp = surface rugosity (horizontal plane) at 400 m2, and VarSRhp = variance in surface rugosity (horizontal plane), calculated as the variance among 25 m2 quadrats within a 400 m2 plot. See Table S3 for a summary of results of the 56 multivariate GLMs.



Table 3. Summary of final models for fish species abundance or presence. Deviance explained is the R2 contribution averaged over orderings among terms (Chevan and Sutherland, 1991).

Species and model type P. microlepidotus

C. guentheri

G. prionodon

N. gymnogenis

O. lineolatus

P. spirulus Overdispersion: 1.2 Theta: 2.207

Coefficient

Estimate

Pr(>|t|)

Deviance explained

Intercept

0.084

0.863

-

VarSRpbf

-1.520

0.04

22 %

Intercept

-1.585

0.03

-

VarSRpbf

-2.194

0.05

26 %

Intercept

-54.36

0.001

-

VarSRpbf

85.52

0.001

32 %

Slope

-25.98

0.023

5%

Intercept

0.197

0.339

-

Depth

-0.681

0.006

51 %

Intercept

0.472

0.004

-

MaxSlope

-0.383

0.04

18 %

Intercept

0.431

0.065

-

Slope

-1.139

0.004

10 %

VarSlope

1.224

0.004

24 %

Notes: the first three are presence models with a binomial distribution; the last three are abundance models with either poisson (N. gymnogenis and O. lineatus) or negative binomial (P. spirulus) distributions. VarSRpbf: variance in surface rugosity, VarSlope: variance in slope, MaxSlope: maximum slope.



Table 4. Summary of final models for fish trophic sub-groups abundance. Deviance explained is the R2 contribution averaged over orderings among terms (Chevan and Sutherland, 1991). Coefficient Estimate Pr(>|t|) Deviance explained Intercept 3.338 0.001 Total (28 %) Stony corals 0.135 0.02 16 % Depth -0.132 0.04 12 % Intercept -1.493 0.02 Total (29 %) Piscivores Poisson SRpbf 1.135 0.005 9% OD: 0.95 Stony corals 0.887 0.02 4% Octocorals 1.231 0.005 7% Macroalgae 1.207 0.005 9% Intercept 2.661 0.001 Total (48 %) Predators Poisson MaxSlope 0.004 0.94 2% OD: 0.84 Depth -0.198 0.002 36 % MaxSlope:Depth -0.152 0.03 10 % Intercept 1.906 0.001 Total (68 %) Sediment infauna predators SRpbf 0.356 0.001 4% Poisson Slope -0.482 0.001 13 % OD: 1.02 Octocorals 0.534 0.001 4% Hydrocorals 0.318 0.001 9% Sponges -0.418 0.001 15 % Ascidians 0.612 0.001 16 % Unk biota -0.355 0.003 4% Depth -0.351 0.03 3% Intercept 0.913 0.001 Total (48 %) Herbivores Negbin VarSlope -0.386 0.02 10 % θ: 9.56 Depth -0.819 0.001 28 % Large canopy macroalgae -0.436 0.02 8% Bryozoans -0.309 0.05 2% Intercept 2.845 0.001 Total (42 %) Planktivore Negbin SRpbfVar -0.588 0.038 1% θ: 0.65 Bryozoans 1.168 0.001 34 % Ascidians -0.722 0.009 6% Intercept -0.053 0.872 Total (51 %) Cleaners Negbin SRpbf -0.521 0.018 15 % θ: 11003 VarSlope 0.469 0.025 4% Sponges -1.512 0.001 32 % Notes: SRpbf: surface rugosity (calculated by projecting surface area onto projected area of best fit), VarSR pbf: variance in surface rugosity, MaxSRpbf: maximum value of surface rugosity, VarSlope: variance in slope and MaxSlope: maximum value of slope, OD: over dispersion parameter, θ theta parameter. All Fish Negbin θ: 19.05



Figure Legends

Figure 1. Study area Solitary Islands region. Land is colored black and shaded areas denote the distribution of rocky reef habitats mapped using swath acoustic sonar. Black dots represent the Autonomous Underwater Vehicle and Baited Remote Underwater Video deployment locations. Gray lines indicate the boundary of the Solitary Islands Marine Park. The inset shows the location within Australia, NSW: New South Wales.





Figure 2. Flowchart explaining the workflow of this study from (a) data collection to (e) predicting fish abundance. (a) Two spatial extents of SR measured in this study, finer

extents are embedded in broader extents and represented by the variance and maximum value of (b) complexity metrics, which were quantified from images along with benthic community composition and fish relative abundance. (c) Summary of 56 multivariate negative binomial models ranked by their wald-values (higher value means more variance explained compared to the null model). In all models fish relative abundance from seven

different functional sub-groups was the response variable, for each sub-group eight different models were run to avoid collinearity of predictor variables. Models 1.1 and 1.2 included SRhp and related variables, while models 2.1 – 2.6 included SRpbf and related variables. SRpbf decouples SR from slope, while SRhp includes the effect of slope. In (c) the red rectangle highlights the best model, this was consistent across sub-groups. (d) Shows

the difference in model fit, assessed by the wald-value, as benthic biota relative abundance and depth are added to the model as predictors. Pbf: plane of best fit, MvGLM: multivariate generalized linear models, GLMMs: generalized linear hierarchical models, GLM: general linear models, M: mobile, T: trophic, G: gregarious, and combinations there in, for instance MTG: mobile-trophic-gregarious, see Table A1 for details of functional groups allocation per fish species.





Figure 3. (a) Mean trophic fish relative abundance and (b) mean benthic percent cover by location, error bars denote standard error. Macroalgae was the most abundant class across locations, thus this category is broken up into turf, calcareous coralline algae (CCA) and fleshy macroalgae. Ascidians and Bryozoans were pooled for visualization only given their low cover. Similarly, stony corals, soft corals and anemones were pooled for visualization into the category of Cnidarians. There were no significant differences in benthic percent cover or fish abundance across locations.



Figure 4. Summary of the effect of complexity metrics, depth and benthic community composition on fish abundance (MaxN): (a) trophic groups and (b) fish species. The symbol denotes a positive (+) or negative (-) effect. (a) Only shows significant predictors for final models of each of the trophic sub-group, while (b) shows all habitat complexity predictors to highlight the most and least important ones. All values are standardized variance explained. Benthic biota variables were tested but not significant. SR: surface rugosity, VarSR: variance in surface rugosity, MaxSR: maximum value of surface rugosity, VarSlope: variance in slope and MaxSlope: maximum value of slope.



Figure 5. Predicted relationships between fish relative abundance (MaxN) and different habitat complexity metrics. Solid dots denote the predicted values of fish abundance based on the full final models holding all other predictors constant and allowing the habitat metric (i.e. SR) to vary across its range, while open circles

denote raw values. The dotted line shows the predicted increase/decrease in fish relative abundance in relation to the habitat complexity metric, black solid lines show the standard error of the prediction, and gray solid lines the 95% confidence interval. Each panel shows a different trophic group: (a) piscivores, (b) predators, (c) sediment infauna predators, (d) herbivores, (e) planktivores, and (f) cleaner fish. The total variance explained in the response variable is denoted by “Total”, while the “Partial” refers to the variance explained by the plotted predictor.


Habitat structural complexity metrics improve

Habitat structural complexity metrics improve

Suggest Documents

habitat structural complexity alters the refuge value of ...

New dimensionless indices of structural habitat complexity: predicted ...

Quantifying Multiscale Habitat Structural Complexity: A Cost-Effective ...

Accuracy and Precision of Habitat Structural Complexity ... - MDPI

Structural complexity in relation to the habitat preferences ... - CiteSeerX

Habitat structural complexity and morphological diversity of fish ...

Complexity metrics for design (simplicity + simplicity = complexity)

GUI Structural Metrics

Web Structural Metrics Evaluation

Metrics for Characterizing Complexity of Network Traffic

Cyclomatic complexity metrics revisited - DSpace@MIT

Software Complexity Analysis Using Halstead Metrics

Product-service system (PSS) complexity metrics

Understanding Website Complexity: Measurements, Metrics, and ... [PDF]

Program Complexity Metrics and Programmer ... - Semantic Scholar

Exploring Complexity Metrics for Artifact-Centric

Understanding Website Complexity: Measurements, Metrics, and ...

Complexity metrics for formally specified business

Advanced Program Complexity Metrics and ... - Semantic Scholar

Structural complexity and dynamical systems

Structural optimization complexity - Springer Link

Using Metrics to Improve Sales Results - RainToday

Habitat complexity: approaches and future directions

Freshwater Fish Habitat Complexity Mapping Using