Enhanced ROV Survey of Tropical Fish and Benthic Communities Associated With Shallow Oil and Gas Platforms

Understanding the ecology of oil and gas infrastructure in our oceans is required to inform decommissioning such that environmental impacts are minimised, and benets maximised. This study equipped an industrial remotely operated vehicle (ROV) with a stereo-video system and collected seven hours of high-denition imagery of two platform jackets, for assessments of benthic (type, density, complexity) and sh assemblages (richness, abundance, length) present. Harriet Alpha (25 m depth) and Gibson/South Plato (8 m depth) are located ca. 120 km offshore of north-west Australia, adjacent to Varanus Island. Twenty-one ‘classes’ of benthic biota were observed in high density, with little bare structure on either platform. Encrusting sponges (average 19% cover per virtual quadrat), barnacles (17%) and sponges (16%) were common on Gibson/South Plato while encrusting sponges (16%) and macroalgae (18%) were ubiquitous on quadrats on Harriet Alpha. The sun coral Tubastraea faulkneri was common on Harriet Alpha (38% of quadrats), though coverage was low (<6%). A total of 102 sh species from 27 families were observed. This included 33 shery target species with an estimated relative mass of 116 kg on Gibson/South Plato and 299 kg on Harriet Alpha, with these species typically concentrated near the seaoor-structure interface. Differences in benthic biota communities present on each platform likely reect spatial variations in oceanographic processes, proximity to natural habitats and the age, conguration, cleaning regime and depth of the structures themselves. This study demonstrates the value of adding stereo-video systems to ROVs to undertake rapid scientic surveys of oil and gas infrastructure.


Introduction
There are presently ca. 12 The present study assessed marine communities associated with two shallow-water platform jacket structures located adjacent to the Lowendal Islands, 120 km offshore on Australia's north-west shelf. A 500 m petroleum safety zone surrounds HA and GSP, meaning no shing has occurred in the immediate vicinity of the platforms since 1986 in the case of HA (34 years) and 2002 for GSP (18 years).

Enhanced ROV surveys of platforms
A small stereo-video system (< 10 kg) was a xed to a Seaeye Panther-XT ROV (PXT 949) operated by Mermaid Marine. The system comprised a pair of high-de nition GoPro Hero5 (black models) inwardly converged at 7° to provide an overlapping eld of view. The stereo-video system uses the same camera components as those currently used in many baited remote underwater stereo-video systems (stereo-BRUVs). Cameras were set to record in 1080p video format at 30 frames per second with a medium eld of view. Video cameras were calibrated pre-and post-mobilisation to ensure accurate stereo-video length measurements using the CAL software (Harvey and Shortis 1998; SeaGIS 2019). To assist in sh and habitat identi cation, an additional GoPro Hero5 camera was mounted in the centre of the system and programmed to take high resolution 12 Mega Pixel still photographs at 10 second intervals throughout the sampling period. Video analysis to identify, count and measure sh was performed using the program EventMeasure Stereo (SeaGIS 2020).
No researchers could access the HA or GSP platforms during this ROV program, therefore a scienti c sampling protocol and camera operation manual were developed in consultation with Industry and ROV operators to ensure that their staff were able to independently perform scienti c surveys. Imagery was collected for the majority of each platform jacket from the surface to the seabed. The ROV maintained a consistent speed of ca. 0.5 m/s and used continuous lighting (40 W, 110 VAC, 2520 lumens, White 6500 K, 130-degree beam angle). All surveys were performed between the hours of 0700 to 1700 AWST to avoid crepuscular changes in the recorded sh assemblage.
To examine the in uence of depth and platform position on sh and benthic communities for HA, we used the ROV imagery of horizontal beams near the surface, at 10 m depth and near the sea oor at 25 m depth on both sides of the platform (widest sides; Fig. 2A). Due to its small size, GSP could be surveyed in its entirety with a vertical transect that enabled viewing from one side of the structure through to the other (Fig. 2B). For comparisons between HA and GSP, we used all imagery collected during the survey which was a combination of the dedicated transects and opportunistic roaming surveys providing a more comprehensive assessment of the structure.

Stereo-video imagery analysis
Benthic community assessment Virtual quadrats were analysed using TransectMeasure™ software (SeaGIS 2020), to assess benthic cover according to modi ed CATAMI categories (Collaborative and Annotation Tools for Analysis of Marine Imagery; Althaus et al. 2015). For each quadrat, 20 points were randomly allocated on the image, and benthos directly underneath the point identi ed to biota categories (e.g., macroalgae, sponges, hard corals) and morphological characteristics (e.g., encrusting, complex or taller erect forms) (see A- Table 1). For each quadrat, a measure of epibiota complexity was recorded by estimating benthic biota height according to four categories: 0: negligible, 1: low (0-20 cm), 2: medium (20-40 cm) and 3: high (> 40 cm) using the known dimensions of structural elements of each platform as a reference. Finally, for each quadrat we also estimated benthic biota density (percent cover) as 0: none, 1: < 25%, 2: 25-75%, 3:>75%.
Fish community assessment For HA, as the ROV often backtracked and covered sections already surveyed, it was not possible to analyse the imagery in a single continuous transect without re-counting individuals. For this platform, we therefore assessed sh communities in two ways: 1) MaxN -we recorded the maximum abundance of each sh species encountered at any point in time throughout the ROV video. This provided a single conservative measure of relative abundance for each species for the entire platform, avoided recounting individuals, but likely underestimated the total abundance of each species present.
2) Horizontal transects -all sh observed on horizontal transects were counted with these transects performed just below the surface (~ 2 m depth), at 10 m depth and at the sea oor (25 m) ( Fig. 2A). These horizontal transects were performed on each of the wide faces of HA giving n = 2 transects per depth.
For GSP, we used the MaxN method (1) and as the platform was small, which likely more accurately represented abundances of each species for this structure. Those individuals that could not be con dently or consistently identi ed to species level were identi ed to the next nest taxonomic level possible. All species were assigned to a feeding guild based on dietary information obtained from Fishbase (Froese and Pauly 2021) and shery-target species noted as those likely to be caught and retained by recreational or commercial sheries operating in this region (Fletcher et al. 2017).
The stereo-con guration of cameras, and the calibration procedure, allows accurate and precise lengths and distances to be measured (Harvey et al. 2001). The lengths of all sh observed at the time of 'MaxN' (HA and GSP), and on each of the six horizontal transects on HA, were measured. The fork length (snout to tail fork; FL) of each sh was measured to the nearest mm while the sh was straight and orientated at an angle 45-90° to the cameras principal optical axes to maximise accuracy. For all species, the mass (M; kg) of sh was calculated using the equation where a is a coe cient describing body shape and condition, L is length, and b the coe cient of geometric growth in body proportions, isometric growth being indicated when b = 3. Values for a and b were obtained from FishBase (Froese and Pauly 2021) for all species where possible; for some species a genus average or values from a congeneric species were used.
The lengths of all sh could not be measured from video imagery as many were only visible on one camera. Approximately 12% of sh were measured as we measured only ca. 20% of individuals where they made up large schools (e.g., Alepes vari, herring scad). Bait sh were also not measured. Therefore, we derived an 'estimated mass' for all sh recorded on transects by assigning the mean size of measured conspeci cs to the additional sh that were observed but for which accurate length measurements were not possible. Where necessary, the average size from congeneric species was used but when unavailable these species were excluded from mass estimates. In assigning a mean size to those that could not be measured we are assuming that conspeci c schools comprise individuals of similar size, a schooling behaviour that has been theorised to reduce predation risk (Ranta et al. 1994).

Data analysis
Analyses and plots were performed using Primer-E (Clarke and Gorley 2006) with the PERMANOVA + add on (Anderson et al. 2008), and the R language for statistical computing (R Development Core Team 2019) with the plotting package ggplot2 (Wickham 2009).
The multivariate sh and habitat data sets were compared across the Harriet depth bands and for benthic habitats, also between structures in Primer-7 (Clarke and Gorley 2015) with the PERMANOVA + add on (Anderson et al. 2008). For the sh comparison, a Bray Curtis dissimilarity matrix (Anderson et al. 2006) was constructed on log(x + 1) transformed multivariate sh abundance datasets. This was used as it does not treat the absences of species as similarities and emphasizes the composition of the assemblage rather than the relative abundance of individual species (Anderson et al. 2008). The multivariate habitat data was normalised, and Euclidean distance was used to produce resemblance matrices. Both sh and benthic habitat analyses used a single factor design with unrestricted permutations of raw data with either depth (three levels) or platform (two levels -benthic data only) as xed factors. Principle coordinate analysis (PCO) were used to determine unconstrained similarities and patterns among groups (depths, transects/quadrats). Individual species or benthic habitat types that were likely responsible for any of the observed differences were identi ed using Pearson correlations of their abundance with the canonical axes. A Pearson correlation of |R| ≥ 0.6 for sh and |R| ≥ 0.5 for benthos was used as an arbitrary cut-off to display potential relationships between individual species and the PCO axes. These relationships are graphically illustrated using vectors that are superimposed onto the PCO plot.
We compared length distributions between platforms and across the different depths of HA using kernel density estimates (KDEs) in the "gg-plot2" package in R to estimate the probability density function of the length-frequency data (Wickham, 2016). The algorithm used disperses the mass of the empirical distribution function over a regular grid of > 512 points and then uses the fast Fourier transform to combine this approximation with a discretised version of the kernel. The model then uses linear approximation to evaluate the density at the speci ed points with bandwidths as the standard deviation of the kernel.

Benthic communities
Ten broad categories of benthic biota were assessed from imagery representing a high but unknown diversity of soft corals, sponges, hard corals, macroalgae, crustaceans, molluscs and hydroids (Supplementary Table 1). Examples of biota observed are presented in Fig. 3. The density (percent cover) of benthic biota was consistently high on both platforms ranging from 25-75%. Benthic biota height was lower on GSP (0-20 cm) than on HA (< 20 ->40 cm), where height was greatest on quadrats assessed in 20-25 m depth.
The composition of benthic communities differed among the platforms (d.f. = 1,55, MS = 4.26, Pseudo-F = 4.26, P(perm) < 0.01; Supplementary Table 1). The hard corals, Tubastraea faulkneri and T. micranthus were only observed on HA in depths > 10 m where they were present in low coverage (< 6%; Supplementary Table 1; Fig. 3, 4). The soft coral Dendronephthya sp1 was also only observed on HA with an average cover of 1.5%. Generally, however, soft corals were more abundant on GSP (~ 10%; Fig. 4) and included one class unique to this platform (Melithaeidae sp1; Supplementary Table 1; Fig. 3). Also only observed on GSP were coralline algae, hydroids, branching and laminar sponges, each in very low cover. Barnacles were evident on both platforms, but in particularly high cover on GSP (mean 16.7% per quadrat) where they were observed on quadrats nearest the surface ( Supplementary Table 1; Fig. 3, 4). A variety of sponge morphologies were observed on both platforms, the most common being encrusting varieties (Supplementary Table 1). The HA platform had a much higher proportion of biota that could not be identi ed from imagery ('other' 33%; Fig. 4). Macroalgae was evident on both platforms, but in higher cover on HA (19% on HA vs. ~3% GSP; Supplementary Table 1; Fig. 3, 4). Benthic communities on HA did not differ according to depths surveyed (P(perm) = 0.21).

Fish communities
A description of sh communities at Harriet Alpha and Gibson/South Plato platforms Across both platforms, and from MaxN data, we observed 4716 sh from 102 species and 27 families ( Table 1, Supplementary Table 2). Forty-nine species were unique to HA and 22 to GSP with 31 species observed at both platforms. Total sh abundance was higher on HA than on GSP, however small bait sh comprised 60% of total sh abundance observed on this platform (n = 2000). Bait sh were also abundant on GSP (n = 750) (Fig. 5). The top ve most abundant sh observed on each platform (excluding bait sh) are presented in Table 1. These abundant species are typically small (< 20 cm FL), schooling, and considered non-target shery species (with the exception of stripey snapper which is shed). No endangered (International Union for the Conservation of Nature; IUCN listed) or protected (Environmental Protection and Biodiversity Conservation; EPBC Act) species were observed on either platform.  (21) painted sweetlips The proportion of the sh assemblage comprised of different families varied between the two platforms, re ecting schools of species within families such as Glaucosoma magni cum (thread n pearl perch) at GSP and of pomacentrids (N. cyanomos, C. fumea), caesionids (P. chrysozona) and carangids (A. vari) at HA (Fig. 6, Table 1). The most abundant shery targeted species were similar on both platforms (Fig. 6) and included various snapper and trevally species (Fig. 5). The most abundant shery target species on GSP was Lutjanus carponotatus (stripey snapper, n = 100), followed by Lutjanus russellii (Moses' snapper, n = 15), Lethrinus atkinsoni (yellowtail emperor, n = 14), Gnathanodon speciosus (golden trevally, n = 13) and Carangoides gymnostethus (bludger trevally, n = 9) (Fig. 5, 6). Three of the top ve shery target species on GSP, were also among the top ve on HA. The most abundant shery target species on HA were Carangoides fulvoguttatus (turrum, n = 30), L. carponotatus (n = 29), G. speciosus (n = 17), C. gymnostethus (n = 16) and Carangoides sp. (unidenti ed trevally, n = 15) (Fig. 5, 6).
The size distribution of sh communities differed among the two platforms with a bimodal pattern observed at HA re ecting a high proportion of both small (< 200 mm FL) and of larger (> 350 mm FL) individuals with fewer between these sizes (Fig. 7). The mean length of sh at HA was 328 ± 12 SE mm and the total mass of measured sh at MaxN was 645 kg. Conversely, a high proportion of the sh assemblage at GSP were of sizes 150-300 mm, where the mean length for the assemblage was 248 ± 8 SE mm (Fig. 7) and total measured mass 287 kg (Fig. 7). This re ects schools of different sized individuals at each platform and the extended depth-range of HA to include larger, typically snapper and emperor species, at the base of this jacket. The total mass of shery target species on HA was 299 kg while on GSP it was 116 kg. Values for HA are lower than those obtained for transect measures as indicated in Sect. 4.2.2 (Fig. 9).

Fish communities across depths on Harriet Alpha
A total of 80 sh species from 24 families were observed around HA from MaxN data. Examination of the separately assessed horizontal transects revealed that total abundance of sh observed on horizontal transects was much higher at 8589 (compared to 3344 from summed MaxN), suggesting that the MaxN measurement for the entire platform was indeed a conservative estimate of total abundance, but accurate for species richness. Fish communities differed across depths (d.f. = 2,5; MS = 5054.5, Pseudo-F = 3.87; P(MC) = 0.045). Of these 8589 sh observed on horizontal transects, most (84%) were observed along transects performed at the sea oor. Species richness was also highest near the sea oor (n = 50 spp) and lowest at surface transects (n = 9 spp). High abundances and diversity near the sea oor re ect, primarily, the high abundance of algae/invertivore and carnivorous species observed there. Examples of these species include Heniochus acuminatus (long n banner sh), Epinephelus coioides (goldspotted rockcod), Lutjanus argentimaculatus (mangrove jack), with others presented in Fig. 8.
The distinction between sh communities observed near the surface, mid-water and at the sea oor was signi cant across this 25 m depth range, illustrated by the varying assemblage composition and abundance (Fig. 8) and size structure (Fig. 9). Abudefduf vaigensis (Indo-paci c sergeant) characterised the shallow sections of the platform (Fig. 8). Abundant in mid-water sections were several moon wrasse species (Thalassoma spp), damsel sh (N. cyanomos, Dascyllus trimaculatus), the roundface bat sh (Platax teira) and schools of Alepes vari (herring scad) (Fig. 8). At the sea oor, snappers (Lutjanus spp) and groupers (Epinephelus spp.) were abundant, in addition to schools of bait sh (unknown sp) and short n bat sh (Zabidius novemaculeatus) (Fig. 8).
While bait sh were most abundant near the sea oor, as their lengths were not measured, they are not included in the length distribution modelling, resulting in the shallower and mid-water depths comprising a higher proportion of smaller-bodied sh (Fig. 9). Most sh species measured from transects at the top and middle sections of HA were < 250 mm FL, with 9 kg total measured mass near the surface and 456.7 kg mid-water. Unlike the top sections, the middle section also comprised sh species of larger sizes (250-500 mm). Length distributions at the sea oor were skewed to primarily larger-bodied species 300-700 mm FL (Figs. 8, 10), typically shery important snappers and groupers (Fig. 8) with a total measured mass in this section of 781.8 kg from transects. Across the depth bands, total mass was 1247 kg for all sh species and 685 kg for shery target species, a less conservative estimate than the mass derived from the MaxN data used to compare structures (Fig. 7).

Discussion
This is the rst study to describe benthic and sh communities associated with two oil and gas platforms within (GSP) and immediately adjacent to (HA) the BIMMA in an area of conservation signi cance. The distinct benthic and sh communities observed on the two structures likely relates to their size, con guration, age, the depth of water in which they are located and their proximity to natural habitat features.
Relative to the size of the two platform structures, high sh abundance and diversity was observed. The platforms were characterised by schools of small caesionids (fusiliers) and carangids (trevallies) that swam within and around the support beams. Attraction of pelagic species such as these to physical structures that span the water column is a well-known phenomenon where such structures are often referred to as sh aggregating devices (FADs; Castro et al. 2002). These behaviours are linked to species obtaining bene ts in terms of nding food, predation success, avoiding predation and increasing mating opportunities (Deudero et al. 1999;Sancho 2000). Additional behaviours were noted that suggest that platforms provide a variety of speci c niche habitats that bene t different species depending upon their ecology. For example, L. carponotatus (stripey snapper) would 'line up' along vertical beams (Fig. 5G) while G. magni cum (thread n pearl perch) would exhibit a similar behaviour but usually in the shadows beneath horizontal beams (Fig. 5E). L. carponotatus may form this vertical con guration to remain as close to the structure as possible, likely for shelter while the behaviour of the Australian endemic G. magni cum suggests it has an a nity for caves/overhang environments.
Herbivorous sh were observed in low numbers (e.g., parrot sh, surgeon sh), perhaps due to a limitation in preferred algal species present on these structures. We did not separately include 'corallivores' as a feeding guild (butter y shes), for we saw very few, likely due to the low amount of hard coral cover on these platforms. Conversely, invertebrate and generalist carnivores were prevalent (A- Table 2), suggesting the presence of abundant invertebrate communities within and around the platforms. A high abundance of bait sh, particularly near to the sea oor likely provides a food source supporting the higher abundance of many other predatory demersal species observed in this part of the platform. An a nity of these larger, typically predatory sh species (groupers, snappers) with the sea oor-structure interface is expected as they are predominantly reef-associated species. Although the structures observed here remain relatively consistent in their structural complexity throughout the water column, species preferring to remain near the bottom may suggest defensive behaviour against open eater predation (Rilov and Benayahu 1998).
An ability to utilise the structure for shelter and to ambush prey would bene t these species. The dominance of carnivorous sh species and relatively low abundance of herbivores/corallivores is likely linked to an absence of suitable benthic biota to support high abundances of these feeding guilds on platforms. While it is unlikely that shery target populations on these platforms have a signi cant contribution to sheries, they do provide sanctuary for these species through an enforced exclusion zone. Research that seeks to monitor shery target species inside and outside no-take sanctuary zones around the Montebello and Barrow Islands should seek to include oil and gas platforms.
In this way, their role as de facto marine reserves may be assessed in addition to providing an understanding of how sh communities on these structures compare to those in surrounding natural ecosystems.
The cover of benthic biota on both platforms was high, yet diversity is unknown given that it is very di cult to identify many corals and sponges to genus or species level without physical examination of specimens. Thus, while a high diversity of biota is known for this region (Richards and Rosser 2012;Fromont et al. 2016), few distinct groups could be identi ed on platforms. This highlights the challenges of utilising imagery, no matter the quality, to assess the diversity of benthic communities where specimen collection or physical assessment is required. However, imagery is particularly useful for the development of reference libraries that become especially valuable when paired with specimen identi cations. On GSP, some of the beams nearest the surface were dominated by barnacles (Fig. 3G). A high diversity of barnacle species exist in shallow waters of north-west Australia (Jones 2012) and their growth and settlement is positively associated with productivity (Menge et al. 2003). Presence of particular biota on various sections of platforms can be related to shading or sheering effects of currents (Love et al. 2019b). Here, high cover on GSP platform beams near the surface likely provides them with a vantage point to utilise their cirri (feathery feeding appendages) to lter feed and trap food such as plankton that may pass them by. Barnacle production on platforms has been shown to serve important ecological functions in the Gulf of Mexico where they are prey for many important commercially shed species (Reeves et al. 2018). While imagery suggests large monospeci c mats of barnacles on the structures, some species of barnacle only associate with particular organisms (Jeffries and Voris 1996), with the latter not observable from imagery. Instead of barnacles, HA had a higher prevalence of macroalgae cover (up to 62% of quadrats possessed an undifferentiated form). Macroalgae genera could not be distinguished from imagery but was of low complexity/height. The presence of macroalgae on HA likely limits space for barnacles to colonise. The ubiquity of two different colonising groups of benthic biota on two relatively close platforms may re ect varying processes of succession (linked to structure age), the presence of different predatory sh communities or even cleaning regimes.
Sun corals (Tubastraea spp) were reasonably ubiquitous (5-38% of quadrats) but present in low cover (< 6%) on HA and were not observed at all on GSP. Unlike other jackets that have been surveyed in Australia (> 50% cover  (Sammarco et al. 2014). This is the third study on offshore platforms of north-west Australia that has documented the presence and abundance of Tubastraea spp. While they are present within their natural range here, further research is required to understand the distribution and extent of sun coral species on arti cial and natural reefs in Australia and the ecological implications of anthropogenic impacts (climate change, arti cial structures) on its distribution and abundance.
Sea fans (Melithaeidae sp1) were observed only on GSP but identi cation to genus level of this family is not possible without physical examination of the sclerites. Observed only in two quadrats, the 'tree-like' soft coral was light pink in appearance (Fig. 3E). The deep blue coloured soft coral, Dendronephthya sp1 (Nephtheidae), was only observed on HA while Carijoa sp1 (Alcyonacae) was observed on both platforms. Each of the soft coral taxa observed occur in moderate to high current ow environments where they can lter feed on zooplankton and other food particles (e.g., phytoplankton) that come within In summary, the two relatively small platform structures surveyed here are within the BIMMA and possess marine communities that share species with those known to exist in surrounding natural ecosystems.
The diversity of sh communities was high, although likely higher than that we were able to quantitatively assess using ROV video imagery. The platforms provide a unique habitat that a subset of sh species utilise to serve their speci c ecological niche. While all species present on platforms occur in natural surrounding ecosystems, the platform structures facilitate their co-occurrence in a single place with habitat spanning the water column from the surface to the sea oor. The signi cance of marine communities present on these, and other platform structures in this conservation signi cant region, requires investigation. This can occur by comparing communities present on platforms to those in natural ecosystems, both within and outside sanctuary zones (to better understand their role as de facto marine reserves). Further, and perhaps most importantly, research should examine the processes that facilitate species establishment and dispersal in addition to the movement of organisms between these structures and surrounding ecosystems -to understand the ecological role these structures play in the Tables Table 2 is not available with this version.