Angolan licence block 31 covers a 5,349 km² area of seafloor and became operational in 2012. To the north of the block is the Congo submarine canyon, which directly connects with the Congo River (Droz et al., 2003; Savoye et al., 2009). An estimated average of 55 x10⁶ tonnes of suspended sediment, typically laden with particulate and dissolved organic matter, is released annually from the Congo River (Wetzel, 1993; Vangriesheim et al., 2009) and deposited on the seabed from surface-derived flux or diverted through the Congo canyon in episodic turbidity flows (Vangriesheim et al., 2009), with sedimentation rates within the canyon of around 36 cm ka^-1 near Block 31 (Savoye et al., 2009). In the area of Block 31 the sediments are mostly hemi-pelagic in origin and sedimentation rates are lower at 20 cm ka^-1 (Hill et al., 2010). The surface waters of Block 31 are characterised by high primary productivity, estimated to be over 200 gCm^-2 yr^-1 (Behrenfeld and Falkowski, 1997), driven by river-induced and coastal upwelling (van Bennekom and Berger, 1984; Schneider et al., 1994). The near-seabed water in the area of Block 31 appears to have typical temperatures (~5°C) and salinities (34.9 PSU), as well as high dissolved oxygen concentrations (230 µmol kg^-1) (Vangriesheim et al., 2009). Current speeds in Block 31 are low (0.02 to 0.12 m/s) (Hill et al., 2010; Jamieson et al., 2017). Regular and large turbidity currents in the nearby Congo canyon may suspend sediment and organic material into the water column, where they may be transported along the slope (Azpiroz-Zabala et al., 2017).

The benthic biology of the deep waters off Angola has received increasing attention (Sibuet et al., 1989; Vinogradova et al., 1990; Bianchi, 1992; Kröncke and Turkay, 2003; Arbizu and Schminke, 2005; Van Gaever et al., 2009; Jones et al., 2014) and has been the focus for a number of large international programmes with biological components including Latitudinal Gradients of Deep-Sea BioDIVersity in the Atlantic Ocean [DIVA] (Arbizu and Schminke, 2005), ZaiAngo (Savoye et al., 2009), BioZaire (Galéron et al., 2009; Vangriesheim et al., 2009), and Congolobe (Rabouille et al., 2017). These studies have mostly focussed on chemosynthetic systems in the deep bathyal (Sibuet and Olu-LeRoy, 2002; Sibuet and Vangriesheim, 2009) and abyssal (Sen et al., 2017) or deep-water coral reefs on the upper slope (Le Guilloux et al., 2009). Less scientific attention has been directed towards the bathyal (200 – 2000 m) depths of the margin, although this area has been the subject of many environmental assessments by the oil industry (Cazes et al., 2012). Access to the region through oil company activities has permitted some scientific investigations on the scavenging fishes in Block 18 and 31 (Jamieson et al., 2017) and on communities associated with asphalt mounds in Block 31 (Jones et al., 2014). Two permanent scientific observatory platforms were installed at Block 18 to the South East of Block 31 (Vardaro et al., 2013), which are bringing further insights into the deep-water biology of the region (Milligan et al., 2020).

In February 2014 BP Angola installed a gas export pipeline off the coast of Angola, which extends a total of 56 km from a deep-water manifold in BPs Concession Block 31 (approximately 1800 m water depth, 6°15’S, 10°49’E) to a tie-in manifold in Block 15 (700 m water depth, 6°14’S, 11°17’E). Remotely operated vehicle (ROV) video surveys were carried out prior to installation in December 2013 and after installation in May 2014 ( Figure 1 ).

A routine pre-lay inspection of the 56 km long swath of seafloor was recorded by ROV Hercules 18 operated from the pipelay support vessel (PLSV) Seven Pacific in December 2013. The ROV was outfitted with a single colour camera and a single black and white camera on a pan and tilt mounting enabling oblique angle viewing. Pre-pipeline installation video survey commenced on December 9^th 2013; the last video collected on December 12^th 2013. Transect videos were recorded in 30 minute files including position in Easting and Northing (coordinate system Camacupa_TM_11_30_SE), depth (meters), altitude (meters), heading, date, and time. All video was standard definition (720 x 576 pixels). The altitude varied as the ROV recorded above the seafloor and images were scaled using known acceptance angles of the camera (32°) following Jones et al. (2006). The average image width for the pre-pipeline survey was calculated at 2.90 m. Image width was also verified by travelling over objects of known size, such as a drink can, on the seafloor. The video was divided into sections representing 100 m depth intervals. Video from each depth band was segmented at random into three sample units each with 500 m transect length. The sample area was calculated from the average image width and transect length, averaging 1429.21 m² in the pre-pipeline survey.

After the installation of the pipeline (pipeline installation was completed on 23 February 2014), an initial inspection survey was done three months after completion in May of 2014 from the same vessel. The ROV travelled along the 304.8 mm (12 inch) diameter pipeline. The ROV remained at a consistent height from the seafloor as it moved on wheels in direct contact with the pipeline. The ROV had three cameras viewing obliquely: one was mounted on the front of the ROV viewing the pipeline ahead, while two were on either side of the ROV recording the sides of the pipeline. The video commenced on May 12^th and was completed on May 26^th. Transect videos were recorded in the same video format as the 2013 pre-pipeline survey. Images were scaled with reference to the known width of the pipeline. The average image width for the post-pipeline survey was calculated at 2.02 m. The same 500 m samples were selected where possible to the pre-pipeline survey. The area imaged was smaller (reduced altitude survey) averaging 939.76 m².

Pipelines were surveyed in a predominantly westerly direction (except in the 700 m depth zone, which were surveyed in an easterly direction) during the pre-installation surveys. The post-installation surveys were surveyed in the opposite direction ( Table 1 ).

Image annotation was standardized between the two surveys. The order of assessment of individual 500 m transects was randomized to reduce the impact of sequence bias (Durden et al., 2016).

Video files were viewed using VLC Media Player (version 2.2.6, VideoLan Community). For the post-pipeline survey only video from the central camera was used for annotation, ensuring a similar approach to the pre-pipeline survey. The video from the side cameras were used on occasion to confirm identification of organisms. All organisms that were entirely within the field of view were counted and identified within each sample (Durden et al., 2016). Organisms were counted at a constant position near the base of the screen (at a line 150 pixels above the bottom of the frame in 2013 and at the level of laser line in 2014) to ensure observations were made with the highest resolution and best lighting.

An organism only partly within the frame was excluded. Every observation was noted with the time, position, depth, identification, and necessary measurements for area, as well as relative position of the organism from the centre of the field of view or pipeline. Measurements were made from the centre of the pipeline, or screen, to the centre of the organism using ImageJ software (version 1.51s 64-bit, National Institutes of Health). In 2014, when the pipeline was installed and visible in video, it was noted whether the pipeline was buried or exposed. It was also noted if organisms counted were directly on or under the pipeline.

Faunal identifications (morphotypes) were made following a standardised catalogue developed for the region (Pfeifer, 2021) that aligns with existing image-based literature from the area (Vardaro et al., 2013; Jones et al., 2014; Jamieson et al., 2017). Organisms were classified to lowest taxonomic level and named according to WoRMS (Horton et al., 2019). Open nomenclature identifiers (Horton et al., 2021) were used to indicate identifier confidence following the protocol set out for physical specimens (Sigovini et al., 2016).

Video resolution varied along transects (particularly prior to pipeline installation) because of suspended sediment, debris, or altitude variation. Poor quality sections of video were identified, removed from analysis and additional video was analysed at the end of each transect to bring the total length to 500 m. In all cases, any changes in transect location did not bring the data outside of the identified depth zone. A single annotator was used for all video analysis to minimize annotator bias (Durden et al., 2016). Only megafauna greater than 80 mm in maximum diameter were included in analysis to ensure consistent identifications could be made between surveys. Smaller fauna, including amphipods, squat lobsters and other small decapods were very abundant but were not included in analysis. Wholly benthopelagic organisms, such as scyphozoans and ctenophores were commonly observed, but were not included in analysis.

A range of ecological parameters were calculated for each replicate transect (used as the sampling unit for all analysis). Numerical density (individuals m⁻²) was calculated for the whole area examined as well as split between the area on the pipe and on the adjacent sediment. To assess the range of diversity characteristics, Hill’s diversity numbers of order 0, 1, and 2 (Jost, 2006) standardised by individuals were calculated as morphospecies richness (S), the exponential form of the Shannon index (exp H´), and the inverse form of Simpson’s index (1/D), using the ‘vegan’ package implemented in R (Oksanen et al., 2019). Hill numbers are a mathematically unified family of diversity indices that incorporate species richness and species relative abundances (Chao et al., 2014).

Generalized linear models (GLM) (Dobson and Barnett, 2008) were built to test whether statistically significant variation in biological parameters was apparent between years and 100 m depth bands, using the ‘car’ package (Fox and Weisberg, 2011) implemented in R (R Core Team, 2020). Models were fitted with quasi-Poisson errors in non-negative integer metrics (i.e. density, S) with over-dispersion (Gardner et al., 1995), and with normal errors applied to non-integer variables (i.e. exp Hí, 1/D) (Freund and Littell, 1981). Homogeneity of variance and normality assumptions were verified by visual inspection of model histograms and QQ plots. Statistical significance was reported for p < 0.05.

Differences in community composition between years and depth bands were tested using permutational multivariate analysis of variance (PERMANOVA) based on Bray-Curtis dissimilarity matrix of fourth-root transformed faunal densities, using the ‘vegan’ package implemented in R. Similarities were visualised using non-metric multi-dimensional scaling (nMDS).

The seabed observed prior to pipeline installation comprised soft sediments with abundant bioturbation (<100 mm width pits, tracks and mounds). No hard substratum was observed, apart from the pipeline and anthropogenic litter. Most litter ( Figure 2 ) was observed after pipeline installation and over 1500 items were recorded in the 2014 survey (16 items total in the 2013 survey) comprising plastic bags, bottles, buckets and aluminium cans. Litter was relatively abundant (>0.1 litter items m^-2) in 2014 between 700 and 1400 m depth (max 0.24 litter items m^-2 at 1300 -1400 m depth) and the amounts reduced below 1500 m depth (< 0.07 litter items m^-2; < 0.03 litter items m^-2 at >1700 m). Terrigenous vascular plant and algal material was observed very occasionally (3 occurrences) prior to pipeline installation, but regularly observed near the pipeline after installation (>500 occurrences). Drifting particles in the water column showed that currents were predominantly from the north to the south. Most (79%) of the litter accumulated on the downstream (southern) side of the pipeline. However, organic material was approximately evenly distributed (56% on southern side) either side of the pipe.

In both 2013 and 2014 megafaunal density varied significantly with depth ( Table 2 ). Before the pipeline was installed megafaunal density was lowest at 1800 m (0.023 ind m^-2) and highest at 800 m (0.215 ind m^-2) with an additional peak in density at 1400-1500 m ( Figure 3 ). There were significant differences between years (GLM: L-ratio = 12.03, d.f. = 1, p < 0.001), depths (L-ratio = 56.74, d.f. = 11, p < 0.001) and a significant interaction (L-ratio = 60.62, d.f. = 11, p < 0.05) – indicating that the pattern with depth changed between years. In 2014, when the pipeline was present, megafaunal density was significantly higher overall than the 2013 survey particularly at 900 – 1200 m and 1800 m depth.

In 2013, the benthic megafauna ( Figure 4 ) were evenly distributed across the width of the transect ( Figure 5 ). After the installation of the pipeline, the distribution of animals changed. In 2014 a total of 132 individuals were observed on the pipeline itself (overall density = 0.02 ind m^-2) but there were not sufficient individuals to address patterns with depth. On the adjacent sediment a total of 5496 individuals were observed (overall density = 0.17 ind m^-2). The density was significantly higher than on the pipeline itself and the 2013 seabed once the depth variation had been accounted for (GLM: L-ratio = 87.95, d.f. = 2, p<0.001) but organisms were distributed close to the margin of the pipe ( Figures 4 , 5 ).

The most abundant organism was the echinoid Phormosoma sp. indet., almost all the individuals of this morphotype occurred between 800 and 1100 m depth and it was around an order of magnitude more abundant in 2014 than in 2013. Phormosoma sp. indet. (example image in Figure 4D ) was particularly abundant at the edge of the pipeline. In 2014, densities tended to be several times higher on the north side of the pipeline, except at 1000 - 1100 m where abundances were nearly an order of magnitude higher on the south side. Although it was harder to detect an elevation in abundances close to the pipe in less abundant taxa, there is some evidence of increased abundances (approximately double the average frequency) of two large holothurians Paelopatides sp. indet. (most abundant >1200 m) and Benthothuria sp. indet. 1 (most abundant <1000 m, Figure 4I ) adjacent to the pipe. Other echinoderms, such as ophiuroids and asteroids, were more evenly distributed with distance from the pipeline. Many taxa showed increased abundances at particular depth ranges, for example ophiuroids (second most abundant taxon - comprising several species) were almost exclusively found at 700 – 800 m and 1700 – 1800 m depth (see Supplementary Table ). Many, but not all, taxa, for example asteroids, Bythitidae stet. and Zoarcids (Pachycara sp. indet.), had higher densities in 2014 than 2013 ( Figures 4M, P ). Notable exceptions, with higher densities in 2013, include ophiuroids, Psychropotidae indet. and Peniagone sp. indet.

Only 30 individual sessile animals were observed on the pipeline, these comprised Actinoscyphia sp. indet. 1 (12 individuals; 36% total observations; Figure 4B ), Actinaria indet. 5 (9 individuals; 41% total observations), Anthozoa indet. 2 (4 individuals; 44% total observations), Actinoscyphia sp. indet. 2 (4 individuals; 21% total observations; Figure 4C ) and Actinaria indet. 1 (1 individual; from a total of 53 individuals). These animals were large (Actinaria indet. 1, the largest, was up to 200 mm diameter) and were visibly attached to the pipe. When the anemones came into contact with the wheels that directed the ROV along the pipeline they were detached easily at the base of the column.

There were significant differences in species richness between years (GLM: L-ratio = 10.64, d.f. = 1, p<0.01), depths (GLM: L-ratio = 61.33, d.f. = 11, p<0.001) and the interaction (GLM: L-ratio = 28.87, d.f. = 11, p<0.001). Species richness was generally higher in 2014, after the pipeline was installed ( Figure 6 ), particularly at depths between 700 and 1100-1200 m. Deeper than 1200 m species richness was similar between the years. In 2013, species richness varied with depth with peaks at around 800 m and 1300 m depth. In 2014 species richness was approximately constant between 700 and 1200 m depth, decreasing at deeper depths.

Exponential Shannon diversity was not significantly different between years (GLM: L-ratio = 0.47, d.f. = 1, p = 0.5) but was different between depths (GLM: L-ratio: 68.73, d.f. = 11, p < 0.001). There was broadly higher diversity, as measured by exponential Shannon, at 1200 – 1800 m depth than shallower, although there was an exception at 1400 m in 2013 ( Figure 6 ). Inverse Simpson diversity shows a very similar pattern to exponential Shannon diversity.

Overall, the community composition varied significantly between years (PERMANOVA: Pseudo-F=21.41, df = 1, p < 0.001), depths (Pseudo-F=8.12, df = 11, p < 0.001) and the interaction between year and depth (Pseudo-F = 2.40, df = 11, p < 0.001). The difference between the years is particularly striking, with clear dissimilarity between years ( Figure 7 ). In both years, the multivariate community similarity shows a gradient of change with depth ( Figure 7 ). Although the gradient is similar before and after pipeline installation the significant interaction indicates differences between the depth patterns in 2013 and 2014. The most distinct communities (by Bray-Curtis dissimilarity) were in the shallowest sites in 2013 (700-900 m) and the deepest in 2014 (1600-1800 m), although there was a gradual increase in dissimilarity with increasing depth difference between sites.