Baited remote underwater video sample less site attached fish species along a subsea pipeline compared to a remotely operated vehicle

Keywords: baited remote underwater video, BRUV, caudal fin aspect ratio, fish assemblage, remotely operated vehicle, ROV, site-attachment, subsea pipeline.

When offshore oil and gas (O&G) infrastructure reaches the end of its productive life, the relevant titleholder(s) must ensure it is decommissioned or otherwise managed in accordance with all relevant regulatory requirements. In Australia, current policy favours the full removal of all infrastructure, however, the regulator may accept alternative options if the titleholder can demonstrate that the alternative decommissioning approach delivers equal or better environmental, safety and well integrity outcomes compared to complete removal (Department of Industry, Innovation and Science 2018). Alternative decommissioning options are often evaluated using comparative assessments, which require a sound understanding of the ecosystem’s functioning in the context of habitat (including artificial subsea infrastructure), and necessarily incorporating the associated assemblage of fish. Because much of Australia’s North West Shelf (NWS) infrastructure is located in remote locations and at depths beyond conventional sampling methods, scientists have previously relied on video records collected by industry during routine visual inspections of infrastructure, with video obtained using industry Remote Underwater Vehicles (ROVs). Analyses of such video records has allowed researchers to describe the fish assemblages associated with such infrastructure (see Ajemian et al. 2015; McLean et al. 2017; Bond et al. 2018a; Thomson et al. 2018; Todd et al. 2019).

ROVs fly above the pipeline filming fish that are in close proximity to it, mostly swimming around it or hiding in pipeline spans (McLean et al. 2017). Despite ROVs carrying lights, being noisy due to hydraulic motors and thrusters, and being a large, moving object, the behaviour of some fish appears little affected (Linley et al. 2013). Experiments testing stimuli independently have found that ROV sound (Popper 2007; Stoner et al. 2008), lighting (Trenkel et al. 2004; Ryer et al. 2009), and speed (Trenkel et al. 2004; Stoner et al. 2008) affect fish behaviour in different ways, with different fish species affected to different degrees. However, beyond experimental surveys, such complex interactions among these stimuli make it difficult to determine whether the fish assemblage recorded is representative of that occurring in the absence of the ROV.

A more common and increasingly recognised method to sample fish remotely, independent of fisheries, and without biases introduced by divers, is with baited remote underwater stereo-video systems (stereo-BRUVS; this method is herein after referred to as BRUV). BRUVs sit stationary on the seafloor and use bait to attract fishes into the field of view of the cameras with captured video recordings being used subsequently to identify, count, and measure fish sizes using the stereo aspect of the technique. BRUVs are particularly powerful for sampling carnivorous fish species often targeted by fisheries (Cappo et al. 2003; Harvey et al. 2007) while still effectively sampling herbivorous species (Harvey et al. 2007). Rapidly deploying a fleet of BRUVs allows scientist to sample large areas in a relatively short period of time and provides customised experimental designs which target particular habitat types.

BRUVs have been compared extensively with alternative sampling techniques, including non-baited RUV (Bernard and Götz 2012; Dorman et al. 2012), diver operated video (Watson et al. 2005, 2010; Goetze et al. 2015; Andradi-Brown et al. 2016), underwater visual census (Colton and Swearer 2009; Goetze et al. 2015; Lowry et al. 2011), towed video (Logan et al. 2017), and fishery dependent methods such as trawls (Priede et al. 1994; Cappo et al. 2004), traps (Priede et al. 1994; Harvey et al. 2012b), and longline (Santana-Garcon et al. 2014; McLean et al. 2015). Most relevant is Schramm et al. (2020), who investigated how relative abundance measures (MaxN) of fish collected using BRUVs compare to abundance counts of fish in a strip transect collected with a micro-ROV, diver operated video, and slow-towed video. However, no work has compared fish data collected from a workclass ROV and BRUV on a subsea pipeline.

Bond et al. (2018b, 2018c) used BRUVs to compare fish assemblages on and off subsea pipelines located on the NWS of Australia. Their surveys sampled fish in depths from 14 to 140 m and BRUV footage of the pipeline across this depth range demonstrated that a BRUV could be landed close to a 12-inch (∼30-cm) pipeline. Despite demonstrating that BRUVs can be deployed next to a pipeline in >100-m water depth, it was clear that the fish assemblage recorded using BRUVs deployed along one pipeline was different to the fish assemblage previously described along the same pipeline from ROV video records. Bond et al. (2018c) noted that BRUVs recorded fewer site-attached species, such as Glaucosoma buergeri (pearl perch) and Epinephelus areolatus (areolate grouper), and more individuals of mobile species like Pristipomoides multidens (goldband snapper) and Seriola dumerili (greater amberjack), despite all of these carnivorous species being bait associated and caught in fish traps in the Northern Demersal Scalefish Managed Fishery (Newman et al. 2008). Site attachment is difficult to measure without tracking individuals. Nevertheless, the morphometry of fishes identifies their locomotion and feeding strategies, both of which could infer a level of site attachment (Bridge et al. 2016). For example, aspect ratio of the caudal fin is a functional trait used to infer locomotion strategy and classify fish as periodic swimmers (constantly swimming) or transient swimmers (burst and glide) (Webb 1984). Periodic swimmers have high aspect ratios, are typically fast and constantly swimming, have high stamina and metabolic rate, a fusiform body, and take food which is widely dispersed in space-time (Sharp and Dizon 1979; Palomares and Pauly 1989; Pauly 1989). Transient swimmers have low aspect ratios, tend to be slower but manoeuvrable around complex structures, prey upon locally abundant but evasive items through a burst and glide strategy, and rarely venture far from complex structure resulting in a smaller home range (Webb 1984; Sfakiotakis et al. 1999; Bridge et al. 2016). BRUVs may sample more species with periodic swimming behaviour as they search widely for food and have higher chance of encountering the BRUV during is deployment time. In comparison, ROVs may sample more transient-like swimmers which are less likely to leave the refugia of the pipeline. Furthermore, where experiments investigating stimuli (light, noise, movement, bait, etc.) on different fish species are not possible, comparing morphometry of fishes surveyed by ROV and BRUVs may also provide insight into the biases or advantages of each method.

Limited research has been undertaken to understand fish communities on the NWS of Australia, particularly in depths greater than 100 m. Nearshore waters have exceptionally high species diversity, comparable to that recorded at equivalent latitudes on the NE coast of Australia inshore from the Great Barrier Reef (McLean et al. 2016), whereas NWS offshore waters are fished by professional trap and line fishers, targeting valuable species such as P. multidens, Lutjanus sebae (red emperor), and Epinephelus multinotatus (rankin cod). From 1959 to 1990 these high-value species were targeted by foreign trawlers (Sainsbury et al. 1997) before declining catches and concerns of trawls removing important benthic habitats led to the exclusion of foreign trawlers, and the development of a smaller, more sustainably managed domestic fishery. Despite this reduction in effort, much of the sponge and epibenthic invertebrate communities had been removed by trawling, coinciding with a noted shift in the fish assemblage from high-value lethrinids (emperors) and lutjanids (snappers) to low-value nemipterids (threadin bream) and synodontids (lizardfish) (Sainsbury et al. 1997). The extent to which the epibenthic and fish assemblages have recovered throughout the NWS is unknown; however, Bond et al. (2018c) found the fish assemblage adjacent to the Greater Western Flank 1 (GWF1) and Echo Yodel pipelines was typified by small-bodied nemipterids and synodontids. Regardless, the provision of hard structure in this region, such as that offered by O&G installations, could encourage the re-establishment and growth of benthic macroinvertebrates and boost numbers of high-value fish species. This study compares fish assemblages observed using an industry ROV and deployed BRUVs along a pipeline located in 130-m water depth with findings important for informing future surveys. Results provide insight into the advantages and limitations of both techniques for characterising fish associated with subsea infrastructure, which has broad relevance to science and industry, as more infrastructure is decommissioned and titleholders begin comparative assessments that consider alternative options to full removal. Furthermore, we describe the fish assemblage and abundance of high-value commercial species at GWF1 pipeline and discuss how this compares to natural areas and older pipelines nearby.

Installed over February and March in 2014 and located in 120–130-m water depth on the NWS of Western Australia (starts 115.88953E, 19.76199S and ends 115.92787E, 19.65158S), GWF1 is a 12-inch (∼300-mm) diameter pipeline that runs 14.8 km and links Goodwyn Alpha Platform to Goodwyn and Tidepole wells (Fig. 1). GWF1 is constructed with 13% chromium welded martensitic stainless steel, coated externally with a 13.5-mm four-layer polypropylene insulation system and designed to be inherently stable during gas filled operations.

Fig. 1.  The location of BRUVs deployments on GWF1 pipeline and the section of pipeline with daytime ROV video footage interrogated for this study. The hollow dot represents a BRUV that was not analysed because it over-turned.

This survey was undertaken on 10 April 2017 using BRUVs deployed over the side of the research vessel, Keshi Mer II. BRUVs were deployed from the surface at three sites along the pipeline with each site possessing five replicate deployments. Replicate deployments were separated by at least 250 m to reduce the likelihood of fish swimming between BRUVs during the 1-h sampling period. It is possible that large, mobile species may be observed on multiple deployments, however, we believe that the likelihood of this occurring was low, being reduced by: (1) limited feeding on bait bags and subsequent very low bait dispersal; (2) regionally moderate seabed current speeds of ∼0.128 m s⁻¹; (Leckie et al. 2015); (3) deploying all BRUVs in a single day, thus reducing fish moving big distances overnight, and; (4) rapid BRUV sampling with <10 min between neighbouring deployments, removing the likelihood for fish to depart one bait station for another during the periods of video acquisition. Of the three pipeline sites, one site collected only four successful replicates because one deployment over-turned and filmed vertically towards surface. The GWF1 pipeline was located from coordinates provided by Woodside Energy Ltd and confirmed onsite using the vessel’s echo sounder, with the pipeline being easily identified in echo sounder images providing it was not buried. Where the pipeline could not be located using the echo sounder (two instances), a BRUV was deployed at the location provided.

Five BRUV systems were used concurrently to maximise efficiency in the field. Each BRUV comprised a pair of Canon Legria HGF25, high definition (1920 × 1080 Full HD 50i) video cameras set to record at 25 frames per second and housed in custom built housings. Cameras were ∼700 mm apart with their directions of view inwardly converged by 7° to maximise the overlapping field of view. Both camera housings were mounted on a steel bar within a trapezium-shaped frame and tethered to surface buoys using rope. Further information on the design and calibration of these systems can be found in Harvey and Shortis (1995, 1998). Four steel weight bars were added to each BRUV as ballast to stop frames overturning in strong currents and a single blue LED-array light was added to provide illumination. Blue light was chosen as previous BRUV research has suggested it has a greater maximum illumination range for imaging in Western Australia waters than white light and red lights (Fitzpatrick et al. 2013). Each BRUV was baited with ∼1 kg of crushed pilchards (Sardinops spp.) contained within a plastic-coated wire mesh bait bag, positioned 1.2 m in front of the camera. Sampling was undertaken during daytime hours (between at least 1 h after sunrise and completed at least 1 h before sunset) to record as many fish species and individuals as possible (Harvey et al. 2012a; Myers et al. 2016; Bond et al. 2018a). Each BRUV was left to record on the seafloor for a minimum of 60 min. Resulting video records from left and right cameras were downloaded, converted to .AVI format (Xilisoft Video Convert Ultimate, Xilisoft Corporation) and analysed using the program EventMeasure Stereo (SeaGIS). All BRUVs were calibrated pre- and post-fieldwork using the CAL program (SeaGIS) allowing accurate determination of the sample area.

ROV video records were obtained during routine industry inspection of GWF1 using a work class ROV (Fugro FCV 3083) in ‘pipeline mode’. Three cameras included a central, forward-facing camera mounted high on the ROV, and two boom cameras that extended to the port and starboard side of the ROV and filmed a side-view of the pipeline and underneath spans.

Inspection operations started at ‘kilometre point’ (KP) = 0.000 where the pipeline first receives gas and ended at the Goodwyn Alpha Platform (KP = 14.8). The ROV survey took ∼53 h, commencing at 23:30 on 18 February 2017 and concluding at 04:11 on 21 February 2017. Only video obtained during daytime hours (between 1 h after sunrise and 1 h before sunset) was comparable to BRUVs, because the abundance and species diversity of fish on a pipeline has been shown to change between night and day (Bond et al. 2018a). Consequently, only ROV video obtained between KP 0.680 and 5.540 was used in this study (i.e. a 4.6-km stretch of the pipeline).

ROV video was provided by Woodside Energy Ltd as 15-min video packets in .MPEG file format for three ROV camera views: port, centre, and starboard. Each video was recorded as 25 frames per second and measured 720 × 576. Before importing into EventMeasure Stereo (SeaGIS), video footage was converted to .AVI format in Xilisoft Video Convert Ultimate and files corresponding to the three views were set side by side and merged into a single (2160 × 576), synchronised file using Adobe Premier Pro (ver. CS4, Adobe Systems; Fig. 2). Short, 5-m transects were sampled within the 4.6-km section of pipeline with each transect separated by a 10–15-m gap, following methods described by McLean et al. (2017).

Fig. 2.  Screenshot of the typical ROV video display with port, centre, and starboard screens stitched together from left to right to form an image 2160 × 576 pixels in size.

Relative abundance of fish recorded using BRUVs was determined as the maximum number of individuals present of each species at a single point in time (Priede et al. 1994; Cappo et al. 2004) within 60 min of the BRUV reaching the seafloor. Only fish within 4 m of the cameras were counted as determined by the penetration of artificial blue light to illuminate fish. The distance of each fish from the cameras was checked using the stereo configuration of the cameras and EventMeasure Stereo software. The dispersal of the bait plume from a BRUV is unknown, which results in an undefined area of attraction. As such, abundance is relative to other BRUVs and the absolute density of fish cannot be determined. For ROV video footage, all fish within each 5-m transect were counted and identified to their lowest taxonomic unit. Some individuals could not be identified to species level and were lumped to genus or family, including: Bothidae (flounder), Decapterus spp. (scad), Nemipterus spp. (threadfin bream), and ‘small unidentifiable species’. Small unidentifiable species most likely included apogonids (cardinal fish) and larval fish too small to identify to family.

Examination of differences in fish assemblage composition, species richness, total abundance, and fish caudal fin aspect ratio measured between BRUVs and ROV was undertaken with a two-factor design: Method (‘Method’, fixed, two levels – BRUV and ROV) and Site (‘Site’, random, three levels – S1, S2, and S3). Data were analysed using a permutational analysis of variance (PERMANOVA) in the PRIMER-E statistical software package (ver. 6, PRIMER-E Ltd, Plymouth, UK; Clarke and Gorley 2006) using the PERMANOVA+ add on (ver. 1.0.8, PRIMER-E Ltd, see www.primer-e.com; Anderson et al. 2008). For multivariate abundance data, a Modified Gower log₁₀ dissimilarity matrix was computed using non-transformed abundance data and 9999 permutations were iteratively computed to obtain a P-value for factors Method (fixed), Site (random) and the interaction term Method × Site. A Modified Gower dissimilarity matrix was chosen because it does not count joint absences as similarities and emphasises differences in abundance (Anderson et al. 2011). For univariate analyses on species richness and total fish abundance, the same tests were undertaken using Euclidean distance matrices computed using non-transformed data. Pairwise tests were undertaken where significant results were found. All small unidentified species and Epinephelus sp10 were removed from the data before univariate and multivariate analysis. S. dumerili were removed from all analyses because they followed the ROV and would likely, therefore, have been recounted.

A principal coordinate ordination (PCO) was used to construct an unconstrained ordination and investigate the spatial separation of samples collected by each method. Individual species that were likely responsible for any of the observed difference were identified by examining Pearson correlations of their relative abundance with PCO axes. A Pearson correlation of |R| ≥ 0.5 was used as an arbitrary cut-off to display potential relationships between individual species and the axes. Species accumulation curves were produced using the vegan package (ver. 2.5-6, Oksanen, F. G. Blanchet, M. Friendly, R. Kindt, P. Legendre, D. McGlinn, P. R. Minchin, R. B. O’Hara, G. L. Simpson, P. Solymos, M. H. H. Stevens, E. Szoecs, and H. Wagner, see https://cran.r-project.org/web/packages/vegan/) and the method ‘rarefaction’ which randomly samples the pool of transects multiple times to display the mean number of species found in each transect.

The aspect ratio of the caudal fin is described by the equation:

where h is the height (cm) of the caudal fin and s is its surface area (cm²; Westneat and Wainwright 2001). Aspect ratio data were obtained using the R package ‘rfishbase’ (see https://CRAN.R-project.org/package=rfishbase; Boettiger et al. 2012). Morphometric data for H. dampieriensis, Ophichthus lithinus (Evermann’s snake eel), and Rhynchobatus australiae (whitespotted guitarfish) were absent and data on similar species were also not available. Bodianus perditio was used as a substitute for B. solatus and Nemipterus furcosus was used for Nemipterus spp. Mean AR was calculated when more than one value was available for a species and these data are presented in Supplementary Table S1.

A total of 15 BRUV deployments were made, of which 14 were successful and 1 was not analysed because it overturned. The pipeline was visible on the vessel’s echo sounder for 86% of deployments and, although not seen in any video recordings, BRUV locations are expected to have been in close proximity (<∼50 m) to the pipeline. In total, 171 fish from 22 species were observed using BRUVs. A total of 298 ROV transects were analysed from which 17 027 fish were recorded. Of these, 16 440 (97%) fish were small unidentifiable species and 21 were unidentifiable fish longer than 20 cm. Excluding these unidentifiable individuals, 566 individuals comprising 22 species were identified from ROV video records. In more than 75% of ROV transects (n = 227) no fish were recorded. When considering identifiable fish, fifteen species were recorded only from BRUV deployments and 15 species were unique to ROV transects.

The five most ubiquitous fish recorded using BRUV were: Nemipterus spp. (occurred on 86% of BRUV deployments), A. spinifer (71%), Lagocephalus lunaris (lunar pufferfish; 71%), P. multidens (64%), and Decapterus spp. (57%). The five most ubiquitous fish recorded using ROV included small unidentifiable species (25% of ROV transects; not considered in any further analysis), E. areolatus (7%), L. malabaricus (7%), G. buergeri (6%), and L. russellii (6%).

Commercially important species identified and recorded during the survey included species caught and retained by commercial trap fishers that operate in the vicinity of the pipeline (McLean et al. 2017; Gaughan et al. 2019): Argyrops spinifer (frypan snapper), Bodianus solatus (sunburnt pigfish), Carangoides caeruleopinnatus (onion trevally), Carangoides chrysophrys (longnose trevally), Caranx ignobilis (giant trevally), Caranx papuensis (brassy trevally), Epinephelus amblycephalus (banded grouper), E. areolatus, G. buergeri, Hapalogenys dampieriensis (Australian striped velvetchin), Lethrinus nebulosus (spangled emperor), Lutjanus malabaricus (saddletail snapper), Lutjanus quinquelineatus (five-lined snapper), Lutjanus russellii (Moses’ snapper), L. sebae (red emperor), Lutjanus vitta (brownstripe snapper), P. multidens, and S. dumerili. Multivariate tests returned significant differences for the interaction term Method × Site and single factor Site, but not Method (Table 1). Pairwise tests revealed that Site 2 was significantly different from Sites 1 and 3 in ROV records, this being the most likely cause for significance in the interaction term. Pairwise tests for Method within each Site, revealed significant differences in the fish assemblages (Table 2), suggesting that within a Site each method recorded video of a different fish assemblage.

Table 1.  Results of PERMANOVA based on Modified Gower log₁₀ dissimilarities of multivariate abundance data (fish assemblage) and on Euclidean distance dissimilarities of total abundance of fish (all species summed), species richness, and AR data in response to the factor method and site.

Table 2.  Pairwise test results for the fish assemblage abundance data, species richness, and AR for all pairs within site and method.

Univariate analysis of species richness data revealed significant difference for both factors and the interaction term (Table 1). Pairwise tests returned the same results as the fish assemblage data: different by Method within each Site, with Site 2 being different from the other sites for ROV (Table 2). BRUVs recorded more species at Site 1 (12) and Site 3 (14) compared to ROV (9 and 10 respectively), but fewer at Site 2 (10) compared with the ROV (17). Within the ROV dataset, Site 2 had a higher average species richness per transect (0.90 ± 0.18) compared to Site 1 (0.24 ± 0.07) and Site 3 (0.13 ± 0.07).

Univariate analysis of AR data revealed significant difference for both factors and the interaction term (Table 1). The most common fishes recorded by BRUVs had larger ARs compared to fish recorded using the ROV (Fig. 3). Species recorded using ROV had a range in AR, from 1.10 (Pterois volitans; red lionfish) to 4.52 (C. chrysophrys) compared to BRUV which ranged from 1.47 (L. malabaricus) to 4.52 (C. chrysophrys; Fig. 3). The mean AR of all fish recorded using BRUV was 2.81 ± 0.06 s.e. and higher than that recorded using the ROV (1.87 ± 0.02 s.e.; Fig. 4). The mean AR of species recorded using BRUV was 2.63 ± 0.24 s.e. compared to 2.28 ± 0.24 s.e. using ROV. Pairwise tests returned the same results as the fish assemblage data and species richness at a Site level: different by Method within each Site, with Site 2 being different from the other sites for ROV (Table 2). Aspect ratio of the caudal fin of fish recorded using BRUV at Site 2 and Site 3 were statistically different but not different for other combinations of sites (Table 2). Site 3 had higher abundance of C. chrysophrys (n = 14; AR = 4.52) compared to Site 1 (n = 4) and Site 2 (n = 1). Unlike fish assemblage data and species richness, no difference was recorded by using ROV between sites (Table 2).

A PCO plot of the fish assemblage showed the greatest separation in samples by the factor Method across the PCO2 axis (Fig. 5). That is, there was a much larger distinction between the fish assemblage surveyed by ROV and by BRUV than differences among sites. Separation of samples collected using ROV were largely determined by the abundance of G. buergeri, L. russellii, L. malabaricus, and E. areolatus (Fig. 5). BRUV data were different across the PCO2 axis and their separation correlated with the abundance of four species: C. chrysophrys, L. lunaris, Nemipterus spp., and A. spinifer (Fig. 5).

Fig. 3.  Scaled ubiquity of fish species recorded on GWF1 with ROV and BRUV and each species corresponding caudal fin aspect ratio. The mean aspect ratio of the caudal fins for fishes recorded on BRUV is higher than that recorded on ROV video footage (dashed line). Ubiquity (number of samples that recorded each species) values are rescaled between 0 and 1 using the minimum and maximum ubiquity calculated for each method. ROV data are displayed to the left and BRUV to the right. Aspect Ratio values for each species were obtained using the R package ‘rfishbase’ (see https://CRAN.R-project.org/package=rfishbase; Boettiger et al. 2012) and are the same as those provide on FishBase (see www.fishbase.org). Examples of tails adapted from Yamaguchi et al. (2018).

Species accumulation (rarefaction) curves for each method do not reach asymptotes (Fig. 6) suggesting more sampling on the remaining sections of pipeline would need to occur to document all species interacting with the GWF1 pipeline. Although sample units from each method are not directly comparable, sampling the same length of pipeline with each method returns a similar shaped curve.

Fig. 4.  Comparisons of the mean (diamonds) and median (dashes) caudal fin aspect ratio of all fish recorded using BRUV and ROV video records. The upper and lower hinges represent the first and third inter quartiles (IQR; 25th and 75th percentiles) and the whiskers represent the largest value and smallest values no further than 1.5 times the IQR. Notches extend 1.58 times the IQR ÷ sqrt(n), or ∼95% confidence interval. Notches in each box do not overlap, proving evidence that the medians differ (Chambers et al. 2018 ).

Fig. 5.  PCO plot using the relative abundance of fish. Species with a modular Pearson correlations of |R| > 0.5 to either axis overlaid, with vector length indicating the strength of the correlation and the direction in which each species is shaping the distribution of samples. Samples circled with a dashed line (n = 227) recorded no fish and were all sampled by ROV.

Fig. 6.  Species accumulation (rarefaction) curves for each method show the same shaped curves despite samples units not being comparable. Both curves have not reached asymptote, suggesting more sampling is needed to understand the entire fish assemblage. Interestingly, each method recorded 15 unique species.

Of the 17 commercial fish species identified in this study, 14 were recorded using ROV and 9 were recorded on BRUVs (Table 3). Three commercially targeted species were unique to BRUV (A. spinifer, C. papuensis and P. multidens) and eight were unique to ROV (B. solatus, E. amblycephalus, G. buergeri, H. dampieriensis, L. nebulosus, L. quinquelineatus, L. russellii and L. vitta). Interestingly, 129 G. buergeri and 82 L. russellii were recorded using ROV and no individuals of either species were recorded by BRUV. Conversely, BRUVs recorded 35 P. multidens and 15 A. spinifer and ROV recorded no individuals of either of these two species.

Table 3.  Abundance and commonality (percentage of deployments or transects) of commercial fish species recorded on the GWF pipeline using BRUV and ROV.

This work begins to address one of the suggestions from McLean et al. (2017) and Bond et al. (2018a) by providing the first method comparison in which BRUV and ROV video records were used to understand the fish assemblage associated with a subsea pipeline. Although the total number of species recorded using each method was the same (22 species) and no significant differences were detected in the total abundance of fish, each method recorded a significantly different fish assemblage at each site. BRUVs recorded more mobile species with higher AR like C. chrysophrys, P. multidens, Nemipterus spp., A. spinifer, and puffers, L. lunaris and Lagocephalus sceleratus (silver toadfish), whereas the ROV recorded species that are less mobile with lower AR, including G. buergeri, E. areolatus and lutjanids (L. malabaricus, L. russellii, L. quinquelineatus and L. sebae).

BRUVs have been compared extensively with alternative sampling techniques, however, this is the first study to compare fish data collected using a workclass ROV and BRUV on a subsea pipeline. Here, BRUVs and the ROV sampled a distinctly different fish assemblage and, coincidently, 15 unique species were recorded by each method out of the total 22 species by each method. Many of these unique species included single observations, but others were highly abundant. Nemipterus spp. and P. multidens were the two most abundant species recorded using BRUVs, but none were recorded using ROV. Conversely, G. buergeri and L. russellii were the two most abundant species recorded on ROV video footage and none were recorded using BRUVs. This comparison demonstrates that no single method samples all the species observed and that a combination of BRUV and ROV should be used to sample the most species. This is consistent with work by Schramm et al. (2020), who identified ∼10% increase in the number of species when BRUV is complimented with a transect based survey such as ROV. Although it is not unusual to see differences in the fish assemblage recorded using different visual sampling technique (Watson et al. 2005; Langlois et al. 2010; Logan et al. 2017), such major differences using visual techniques is rare.

BRUVs and ROVs have clear methodological differences, including the use of bait in BRUVs. BRUVs recorded high abundances of P. multidens, A. spinifer, and Nemipterus spp., species that were not recorded on ROV video footage. These species are generalist carnivores with high trophic levels identified in FishBase (R. Froese and D. Pauly, see www.fishbase.org) records (P. multidens = 3.8 and A. spinifer = 4.5) likely attracted to the bait. However, the ROV also recorded high abundance of generalist carnivores, many of which were not recorded using BRUVs or were recorded in relatively low abundance (see Table 3), including; G. buergeri, L malabaricus (trophic level = 4.5), E. areolatus (3.7), and L. russellii (4.1). Traditionally, survey techniques that use bait show significant differences in fish assemblage composition driven by increases in the relative abundance of fish in baited compared to non-baited samples (Cappo et al. 2004; Harvey et al. 2007; Watson et al. 2010; Hardinge et al. 2013). Here, it is clear some species are attracted to the bait on BRUVs, but other attractants or deterrents may also be contributing to differences in abundances of certain species between BRUV and ROV video records.

Fish behaviour is also altered by the lighting (Trenkel et al. 2004; Ryer et al. 2009), sounds (Popper 2007; Stoner et al. 2008), and speeds (Trenkel et al. 2004; Stoner et al. 2008) of underwater vehicles with bias differing among species. The ROV used in this survey had eight 600 Watt lights and two 24 Volt Direct Current light-emitting diodes, a hydraulic motor with seven 15 inch propellers, and was guided along the top of the pipeline by two plastic rollers. This combination of lights, movement, and sound create a complex array of stimuli that would likely elicit a variety of behavioural responses for different fish species. P. multidens and A. spinifer have been recorded on video recordings collected by ROV before (McLean et al. 2017; Bond et al. 2018a), but in low abundance. Bond et al. (2018c) described a disparity in the abundance of these two species recorded on BRUVs compared to previous ROV work at Echo Yodel (EY) pipeline; a pipeline with the same diameter as GWF1 and in close proximity to this study. The results presented here support results from Bond et al. (2018b, 2018c), confirming that P. multidens and A. spinifer interact with pipeline infrastructure on the NWS, despite their absence or low abundance recorded on ROV video records at the same locations (McLean et al. 2017; Bond et al. 2018a). We suggest that the low abundance of these species recorded on ROV video records by Bond et al. (2018c) is likely due to the sound, movement or vibration of the ROV, but targeted experiments testing species-specific responses to these stimuli are needed if the reactions of different fish species to different sensory stimuli is to be understood. Currently, we are unable to ‘normalise’ fish abundance data to compensate for the sensory attributes of different ROVs, or even data from ROVs and BRUVs more broadly.

Although some species may be deterred by the lights, sound or movement of the ROV, others may be attracted to it. S. dumerili, for instance, followed the ROV for much of the survey, making it difficult to determine the true abundance of this species. Fish and sharks circling ROVs were previously noted by Baker et al. (2012), Patterson et al. (2008) and Trenkel et al. (2004) described how cutthroat eels were attracted to a stationary ROV. In this study, S. dumerili appeared to follow the ROV and fed on items disturbed by the ROV. Video analysts reviewing the ROV video footage recorded the first occurrence of S. dumerili and the maximum number of individuals seen at a single point in the same transect. Additional individuals were added in subsequent transects if the total number increased. If three transects passed without seeing any S. dumerili, all individuals observed in the following transect were recorded, the assumption being that these were new individuals rather than previously observed followers. We believe this method did not accurately measure the absolute number of individual S. dumerili associated with GWF1, resulting in the removal of this species from analyses. We suspect the same six S. dumerili continuously followed the ROV for the entire survey but were only filmed in a few instances. We suggest a more conservative, but accurate method to determine their abundance. For example, a MaxN measurement could be adopted for mobile species like sharks and carangids, whereby only one transect with the highest number of individuals is considered in analyses. Alternatively, analysts could follow the methods employed by Baker et al. (2012), who excluded all fish that approached the field of view from behind the ROV.

The small unidentifiable species reported here are common in research projects using ROV to survey oil and gas infrastructure in north Western Australia and most commonly reported as Apogonidae spp. (see McLean et al. 2017, 2018; Bond et al. 2018a, 2018b). Their identity is not known but likely consists of multiple species or families, and their presence and abundance are poorly understood. Bond et al. (2018a), reported diel changes in the abundance of Apogonidae sp. residing on pipeline infrastructure; apogonids reside on pipeline infrastructure during the day and move off it at night. McLean et al. (2017) found no conclusive seasonal changes in the abundance of these small fish, but suggested they could be larval fish. We acknowledge that, on paper, the abundance of small unidentifiable fish reported here is immense and disproportionate to other species. Consequently, we removed them from all statistical analyses because we do not fully understand the ecological or methodological reasons for their presence and abundance. It is important that future research investigates how ROVs and lights affect the presence and abundance of these small fish, and if possible, collect samples to identify the animals to species.

The effectiveness of survey methods that use bait is determined by the behaviour of the target animals, notably their activity rhythms, feeding motivation, and sensory and locomotor abilities (Stoner 2004). As generalist carnivores, the feeding motivation and bait affiliation of G. buergeri, L. malabaricus, L. quinquelineatus, and E. areolatus is considered high, but BRUVs recorded very low abundances of these species relative to the ROV video transects. Despite being generalist carnivores, the feeding motivation of these species may change with their activity rhythm. Bond et al. (2018a) described the activity rhythms of these species at a pipeline nearby to GWF1, noting peak abundance during daytime hours. These peaks in abundance were thought to reflect feeding times, with these species of fish leaving the pipeline at night to feed in the surrounding areas. If this pattern does reflect an activity rhythm of feeding behaviour, it would suggest some species have low feeding motivation during the day, and thus they are less willing to leave the pipeline to explore a feeding opportunity at a BRUV, even if nearby. Additionally, some species might prefer not to leave the pipeline at all, adopting a sit-and-wait approach to prey capture.

In fish, site attachment is difficult to measure without tracking individuals, however, the morphometry of fish can infer their locomotion and feeding strategy (Sharp and Dizon 1979; Webb 1984; Pauly 1989), both of which could infer a likely home range or level of site attachment (Bridge et al. 2016). The mean caudal fin aspect ratio (AR) of fish recorded using BRUVs (2.81 ± 0.06) was statistically different from than that recorded using ROV (1.87 ± 0.02). Fish with higher ARs such as those recorded using BRUV (e.g. C. chrysophrys and Nemipterus spp.) are typically fast and constantly swimming, have high stamina, and known as periodic swimmers. Whereas low AR species such as those recorded using ROV (e.g. E. areolatus and G. buergeri) tend to be slower and adopt a burst-and-glide action, and are known as transient swimmers. Moreover, body form and locomotion are linked to feeding strategy: periodic swimmers take food which is widely dispersed in space–time, and transient swimmers prey upon locally abundant but evasive items (Sfakiotakis et al. 1999; Webb 1984). As previously discussed, the fish assemblage recorded using BRUVs had higher abundance of P. multidens and A. spinifer with few E. areolatus, and no G. buergeri or L. russellii. P. multidens and A. spinifer have AR values of 2.51 and 2.56 respectively and are ranked in the top ten species with the highest AR recorded in this study. Conversely, the ROV video recorded relatively high abundance of E. areolatus and G. buergeri, both species classified as more transient swimmers with AR values of 1.76 and 1.78 respectively. Despite differences in ARs of fishes recorded between survey techniques, it is difficult to decipher whether these differences are due to fish species’ feeding strategies and bait affiliations or to their ability to detect the bait plume and swim to its source.

Ideally, for BRUVs to work most effectively, fish sense the bait plume, decide to follow it and swim into the field of view before the BRUV is retrieved. However, species-specific differences in fish behaviour and locomotion will undoubtedly affect which species are recorded and their estimated abundances. Here, the most mobile species (e.g. C. caeruleopinnatus and C. chrysophrys) would be able to traverse the 250-m separation distance between neighbouring BRUVs and be counted more than once, thus inflating their numbers on BRUVs relative to ROVs. However, for reasons stated previously, this was deemed unlikely in the present study. Collins et al. (1999) and Bailey and Priede (2002) described how the arrival times of fish to bait was reduced for fish that can swim faster. A strong influence on swimming speed is a fish’s caudal fin aspect ratio (Sambilay 1990; Fisher and Hogan 2007). The species recorded using the ROV had a low aspect ratio, compared to those recorded on BRUVs which had a larger range in aspect ratio, but also ARs mostly larger than those recorded on ROV. We do not suggest that fish absent from BRUV data could not swim fast enough to be recorded, rather their vagility, as a consequence of feeding motivation, activity rhythm, and inherited locomotion ability (caudal fin aspect ratio) results in them taking more time to approach the BRUV. To pursue this idea further, we recommend investigating whether longer BRUV soak times result in recordings of more species, particularly those fishes with low AR that seem more closely associated with the pipeline structure.

Finally, using AR to help explain differences between data recorded by each survey technique may also help explain differences recorded at a site level. Within ROV samples, Site 2 was statistically different from the Site 1 and 3 in terms of the composition of fish and species diversity but this difference was not detected using BRUVs. Because BRUVs sample fish with higher AR that roam further, changes in a fish assemblage recorded over a small spatial scale may not be evident with this approach. Here, the ROV recorded fish with smaller AR values, these being species that likely roam less, which allowed the ROV to detect small changes in fish assemblages along the pipeline.

Understanding how commercial species interact with subsea infrastructure, and if this interaction is recorded differently dependent on the method used, is important for fisheries management and in informing decommissioning decisions for the NWS where the value of subsea infrastructure to fishers may be a component of decision making. The GWF1 falls within the Pilbara Trap Fishery and the Pilbara Line Fishery as part of the Northern Demersal Scalefish Managed Fishery and primarily targets high-value species such as P. multidens and L. sebae (Gaughan et al. 2019). L. sebae were recorded on BRUV and ROV video footage but, as discussed, the ROV recorded L. sebae more often than BRUVs. P. multidens was only recorded on BRUV records and was the second most abundant species (35 individuals), after Nemipterus spp. (38 individuals) recorded this way. Both species were recorded on pipelines nearby in studies using ROV (McLean et al. 2017; Bond et al. 2018a) and BRUVs (Bond et al. 2018b, 2018c), although, as previously mentioned, studies using ROV have noted abundances of P. multidens were much lower than expected. Differences in the abundance of P. multidens and L. sebae between methods is likely due to avoidance of the ROV by these species. As discussed, noise, lights, and movement from underwater vehicles have been shown to scare away some species (Trenkel et al. 2004; Popper 2007; Ryer et al. 2009), which is the suspected cause, here. We suggest future work is needed to understand behaviour interactions of commercial species with underwater vehicles used to survey infrastructure, particularly to understand species’ different responses to stimuli such as lighting, noise, and movement.

Although lengths of fish were not used in this comparison, the stereo aspect of BRUVs can provide valuable length data of commercial species, something which is not collected during routine industry inspections of offshore infrastructure using ROVs. Lengths are used to calculate the biomass of each fish and the ‘catch value’ for each BRUV deployment (Bond et al. 2018d). These data can assist decision makers to calculate the ‘value’ of infrastructure fishers, relative to surrounding areas of natural habitat and help inform decisions on decommissioning options. It would, therefore, be useful if industry ROV operators considered the addition of bespoke stereo-video camera systems to enhance ROV capability.

Successfully deploying and retrieving BRUVs in depths greater than 100 m requires a sound understanding of ocean processes and conditions (wind, currents and waves). Unlike Bond et al. (2018b, 2018c), the pipeline was not visible on any BRUV footage including those from rear facing cameras specifically used to identify the presence of the pipeline. It is, therefore, possible that the BRUVs deployed in this study were not close enough to the pipeline to record site-attached species. Bond et al. (2018c) identified the EY pipeline on 43% of recordings from the BRUVs deployed during their study and, interestingly, four out of the five G. buergeri recorded at EY pipeline sites were on BRUVs that had the pipeline in the field of view. Furthermore, all E. areolatus recorded (five individuals) by Bond et al. (2018c) occurred on a single drop which landed next to the pipeline, facing into a pipeline span. A single E. areolatus was recorded using BRUVs in the current study. Furthermore, this BRUV was also the only deployment to record L. sebae (one individual) and L. malabaricus (three individuals); two species which were encountered more often in ROV transects (4 and 22 transects respectively). This deployment was the only BRUV that did not record Nemipterus spp., a species commonly associated with bare sand environments, off-pipeline (Bond et al. 2018b, 2018c). The composition of fish recorded on this deployment suggests it is closer to GWF1 than others. Alternatively, it may be in close proximity to a section of spanning pipeline, which provides increased structural complexity that can be attractive to fish (McLean et al. 2017; Bond et al. 2018a). We suggest that future field work utilising BRUVs to survey offshore infrastructure should consider using an ultra-short baseline (USBL) acoustic tracking system to provide an accurate, real-time position of the BRUV’s on-bottom location and position relative to the pipeline. Finally, considering pipelines like GWF1 can move, we recommend multibeam surveys around the structure to provide an accurate and up-to-date position of the pipeline before sampling with BRUVs.

Industry routinely undertake visual inspections of pipeline infrastructure with work and survey-class ROVs. These operations cost more than A$100 000 per day and are not purposed to survey fish. Consequently, we could consider any information opportunistically gathered (i.e. records of fish) – free. Scientists and O&G operators are realising the scientific potential of historical video records collected during routine operations and using them to understand better the ecology around subsea infrastructure (Macreadie et al. 2018; McLean et al. 2018, 2020; Todd et al. 2020). By contrast, the 14 BRUV deployments used for this study were completed in one field day which cost less than A$10 000. The time taken to analyse video footage was approximately the same between methods. Collecting historical ROV video footage to survey fish may be free, but the design of any experiment utilising this video footage is bound by the limits of available data. A bespoke scientific survey using BRUVs has flexibility in experimental design. Indeed, scientists could employ ROV for targeted surveys (Schramm et al. 2020) and increase the flexibility in their experimental design, but surveys at depths and in remote locations such as those sampled here, would require expensive ROV systems beyond the finances of most scientists. We should iterate that the aim of this study was to compare surveys of fish along a subsea pipeline with ROV and BRUV, not to determine the most appropriate method to survey. As Sward et al. (2019) suggested, true abundance estimates obtained by ROV cannot be used to make reliable comparisons to relative abundance such as MaxN, but studies using these two methods may provide comparisons of assemblage structure and the efficiency of each method. Our study does just that. We recommend that scientists continue to utilise ROV video footage collected by industry to answer questions within the limits of the data and work with their survey team to increase the scientific value of their operations as outlined by McLean et al. (2020). Coupled with historical ROV video footage, BRUVs can be used for surveys of infrastructure and adjacent areas, which require a more complex experimental design financially viable for scientists.

Supplementary material is available online.

The data that support this study cannot be publicly shared due to ethical or privacy reasons and may be shared upon reasonable request to the corresponding author if appropriate.

The authors declare that they have no conflicts of interest.

Woodside Energy Ltd (as the operator of the North West Shelf Project pipelines studied) provided project funding. The Australian Government, Woodside Energy Ltd, and The University of Western Australia, through Research Training Program (RTP) Scholarship, Woodside Top-Top and UWA Safety-Net Top-Up Scholarships, provided stipend for T. Bond to undertake this research as part of a PhD.