Offshore platforms as novel ecosystems: A case study from Australia’s Northwest Shelf

Abstract The decommissioning of offshore oil and gas platforms typically involves removing some or all of the associated infrastructure and the consequent destruction of the associated marine ecosystem that has developed over decades. There is increasing evidence of the important ecological role played by offshore platforms. Concepts such as novel ecosystems allow stakeholders to consider the ecological role played by each platform in the decommissioning process. This study focused on the Wandoo field in Northwest Australia as a case study for the application of the novel ecosystem concept to the decommissioning of offshore platforms. Stereo‐baited remote underwater video systems were used to assess the habitat composition and fish communities at Wandoo, as well as two control sites: a sandy one that resembled the Wandoo site pre‐installation, and one characterized by a natural reef as a control for natural hard substrate and vertical relief. We found denser macrobenthos habitat at the Wandoo site than at either of the control sites, which we attributed to the exclusion of seabed trawling around the Wandoo infrastructure. We also found that the demersal and pelagic taxonomic assemblages at Wandoo more closely resemble those at a natural reef than those which would likely have been present pre‐installation, but these assemblages are still unique in a regional context. The demersal assemblage is characterized by reef‐associated species with higher diversity than those at the sand control and natural reef control sites, with the pelagic community characterized by species associated with oil platforms in other regions. These findings suggest that a novel ecosystem has emerged in the Wandoo field. It is likely that many of the novel qualities of this ecosystem would be lost under decommissioning scenarios that involve partial or complete removal. This study provides an example for classifying offshore platforms as novel ecosystems.


| INTRODUC TI ON
Offshore oil and gas platforms (hereafter offshore platforms) have been a feature of continental shelf waters for over 70 years, with nearly 12,000 of these structures currently installed around the world (Aagard & Besse, 1973;Ars & Rios, 2017). When an offshore platform is no longer economically viable, a decision is made on the fate of the structure through a process referred to as decommissioning. In most cases, decommissioning involves complete removal of the platform from the marine environment for scrapping or recycling on land (Schroeder & Love, 2004). Complete removal is legislated as the default decommissioning method in many countries and regions, including Australia and the North Sea, as well as internationally under the United Nations Convention on the Law of the Sea (UNCLOS) and the 1996 Protocol to the London (Dumping) Convention (Chandler et al., 2017;Elizabeth, 1996;Techera & Chandler, 2015). However, the London Convention does permit in situ decommissioning for purposes other than disposal, and some regions have legislated such methods. In the Gulf of Mexico, platforms can be left either wholly or partially in place, or towed to a new location, under a program known as Rigs-to-Reefs (RTR, Reggio, 1987).
Offshore platforms have been shown to form highly complex artificial reefs (Shinn, 1974), and RTR programs represent a method for preserving and maintaining these artificial reef communities that are established around offshore platforms over the decades they spend in the ocean, similar to the reefs formed by shipwrecks (Dauterive, 2000;Leewis et al., 2000). In situ decommissioning is a financially beneficial option for energy companies due to the excessive costs associated with complete removal (Dauterive, 2000), and this motivation is often used as an argument against rigs-to-reefs, particularly by environmental groups (Löfstedt & Renn, 1997).
Offshore platforms play various ecological roles, including acting as aggregation sites for marine megafauna (Haugen & Papastamatiou, 2019;Robinson et al., 2013), nurseries for juvenile fishes Nishimoto et al., 2019), and providing habitat for economically important and overfished species Love et al., 2006). The presence of these offshore platforms creates new habitat, which can have a significant impact on fish production; platforms in California are some of the most productive fish habitats in the world, and platforms in Gabon have higher fish biomass than pristine reefs in the Pacific (Claisse et al., 2014;Friedlander et al., 2014). Fishing is excluded around offshore platforms in many countries, either by law as is the current case in Australia (Commonwealth of Australia, 2010), or by the presence of subsea infrastructure which can damage fishing equipment (de Groot, 1982). The partial or complete exclusion of fishing effectively creates de facto marine protected areas (MPAs) around offshore platforms (de Groot, 1982;Friedlander et al., 2014).
The exclusion of fishing is particularly important in areas which are overfished, or where hard substrate is limited and infrastructure may be some of the only obstacles to trawling (de Groot, 1982;Fujii & Jamieson, 2016;Love et al., 2006;Schroeder & Love, 2002).
There is an increasing research focus around the world on the potential ecological importance of offshore platforms, and particularly on ensuring that the role of these platforms as ecosystems is considered in the decommissioning process (Bull & Love, 2019;Fowler et al., 2014Fowler et al., , 2018Macreadie et al., 2012;Meyer-Gutbrod et al., 2020). An ecological perspective of offshore platforms allows scientists to apply restoration principles to the decommissioning process, in a similar way to terrestrial restoration of abandoned mine sites (Koch & Hobbs, 2007). The presence of offshore platforms modifies communities and habitats to such an extent that returning the site to its pre-installation state may no longer be feasible or preferable (Sommer et al., 2019), and as such, the benefits of in situ decommissioning must be evaluated.
This assertion is congruent with the concept of novel ecosystems, which is intended to complement existing restoration practices. A novel ecosystem is one which has been altered by human activity and where restoration is not feasible or would result in the loss of ecosystem value (Hobbs et al., 2013). Recently, there have been attempts to apply restoration management concepts to offshore platforms in terms of: establishing ecological baselines for restoring the ecosystem post-decommissioning (Fortune & Paterson, 2020); the potential for restoration paradigms to shift the discourse surrounding RTR decommissioning (Ounanian et al., 2019); and direct application of novel ecosystems criteria to offshore platforms (Schläppy & Hobbs, 2019; van Elden et al., 2019).
There is still a significant knowledge gap around the ecology of these platforms, particularly outside of the major northern hemisphere oil and gas producing regions. In Australia, only a limited number of studies exist on the fish and shark communities around offshore infrastructure McLean et al., 2019;Pradella et al., 2014;Thomson et al., 2021). Information on how ecological value is retained under varying decommissioning scenarios is needed at a time when the Australian government is reviewing legislation to potentially allow in situ decommissioning options (Offshore Resources Branch, 2018;Taylor, 2020). It is critical that we understand the ecological role platforms play in a regional context before the associated ecosystems are potentially lost due to decommissioning and restoration activity.
The offshore oil and gas producing region of northwest Australia, the Northwest Shelf (NWS), is comprised of over 40 production facilities and over 2000 km of subsea pipelines Geoscience Australia, 2009). This is not a large number of platforms when compared with other locations around the world.
However, the NWS is largely devoid of any significant natural hard substrate, and therefore offshore platforms contribute a significant portion of such habitat regionally, along with its associated fishes.
This area was historically characterized by established macrobenthos communities made up of sponges, gorgonians, and soft corals on flat, sand inundated pavement (Evans et al., 2014). These macrobenthos communities were largely removed by pair-trawling operations in the 1960s and 1970s (Fromont et al., 2016;Sainsbury et al., 1997). Previous studies on both platforms and pipelines on the NWS have found significant macrobenthos habitat associated with these structures, and abundance and richness of fish was | 3 of 29 van ELDEn Et aL. higher on pipelines than on nearby natural habitats McLean et al., 2018McLean et al., , 2019.
These results suggest that the hard substrate provided by oil and gas infrastructure may modify the habitat and associated communities from their previously trawled state.
We investigate whether the presence of active offshore infrastructure at a site on the NWS has resulted in the emergence of a novel ecosystem, characterized by a shift in the structure of marine communities. Demersal and pelagic taxonomic assemblages, as well as macrobenthos communities, were documented around the infrastructure in the Wandoo oil field (Wandoo) over 3 years and six surveys and in contrast to two control sites: a sandy site (Control Sand) and a natural reef (Control Reef). Baseline (pre-installation) ecological information for the Wandoo site was not collected, as has been the case for many older offshore platforms (Fortune & Paterson, 2020).
As such, the Control Sand site acts as a proxy for the historical state of the Wandoo site. We determined historical state as the state of the environment immediately prior to the installation of the Wandoo infrastructure, and as such, this site would have been subject to trawling. Anthropogenic disturbance creates challenges in selecting historical baselines, and our baseline selection is congruent with the Anthropocene baseline concept (Kopf et al., 2015). The Control Reef site is characterized by a rocky substrate with significant physical relief, and similar in spatial extent to the infrastructure in the Wandoo field. Control Reef provides contrast to the Wandoo site in the form of a natural reef that is comparable in size (area) and depth (m). These two sites allowed us to both assess Wandoo as a novel ecosystem and predict how the marine communities would be altered under two different decommissioning scenarios. Specifically, complete removal may see the Wandoo site revert to a state more similar to the Control Sand site, and partial removal (topping) may lead to something more similar to the Control Reef site, due to the loss of the mid-water hard substrate. We chose to use the post-trawling state of the Northwest Shelf as our historical baseline, as if the Wandoo infrastructure were to be removed, this area would likely be exposed to trawling again.
We used baited remote underwater video systems (BRUVS) to determine how taxonomic richness, abundance, biomass, fork length, and community assemblage structure varied between these sites, as well as intra-and inter-annually. We hypothesize that the demersal and pelagic assemblages at Wandoo would more closely resemble those at the control reef site than those at the control sand site with respect to diversity, abundance, and size. We then evaluated our findings on the Wandoo field against the criteria for testing whether an offshore platform can be classified as a novel ecosystem (van Elden et al., 2019).

| Study sites
The three sites sampled are located in the NWS region of north- Unlike the WN site, the CS and CR sites are accessible to commercial and recreational fishing.
Seabed stereo-BRUVS have been adapted to mid-water environments, making them a useful tool for documenting highly mobile and elusive species (Bouchet et al., 2018;Letessier et al., 2013;Thompson et al., 2019). BRUVS-derived data should be interpreted recognizing the potential impact of variable bait plumes (Whitmarsh et al., 2017), the potential higher representation of piscivores, and the relative nature of abundance estimates in contrast with density estimates generated by, for instance, underwater visual census (UVC, Langlois et al., 2010). Despite these constraints, BRUVS can be used to document clear signals in marine communities relative to other methods (Cappo et al., 2006;Lowry et al., 2012). Mid-water stereo-BRUVS are deployed for a minimum of 120 min, and in this study, are anchored to prevent entanglement with subsea infrastructure.

| Data processing and treatment
Prior to each survey, individual stereo-BRUVS were calibrated in an enclosed pool, according to standard protocols, using the CAL software (Harvey & Shortis, 1998;SeaGIS Pty Ltd, 2020). All video samples collected in the field were converted to AVI format using Xilisoft Video Converter Ultimate (Xilisoft Corporation, 2016) and videos were processed using the Eventmeasure software package (SeaGIS Pty Ltd, 2020). Processing commenced either once seabed stereo-BRUVS had settled on the seabed, for a period of 60 min, or when the mid-water stereo-BRUVS had stabilized at 10 m depth following deployment, for a period of 120 min. All animals entering the field of view were identified to the lowest possible taxonomic level, and abundance was estimated using the conservative abundance metric MaxN, which is the maximum number of individuals of a given taxon in a single frame (Cappo et al., 2006). The appropriate length metric (e.g., fork length FL, disc width DW, or carapace length CL) was measured in stereo with individuals measured where they were well positioned relative to the camera and not occluded by other individuals. For seabed stereo-BRUVS, the habitat visible in the field of view was broadly categorized into three groups: sand (bare substrate with no visible macrobenthos or other marine growth); sparse macrobenthos (predominantly bare substrate with <50% biotic cover); and dense macrobenthos (the visible substrate was dominated by more than 50% biotic cover).
For seabed stereo-BRUVS, a sample was an individual rig deployment. For mid-water stereo-BRUVS, samples consisted of each set of five BRUVS deployed in a zone. This method mitigates the potential effect of highly mobile pelagic species being observed on multiple mid-water deployments.
The video analysis yielded identification, abundance, and length data for each stereo-BRUVS deployment. These data were analyzed as taxonomic richness (TR), total abundance (TA), and fork length (FL), respectively. Total biomass (TB) was calculated as the sum of mean weight of a given taxa on a given sample. Weight was calculated based on FL using taxon-specific length weight relationships (LWR) sourced from Fishbase (Froese et al., 2019). Where the LWR was not available for a particular taxon, the LWR based on total length (TL) for that taxon was used, in combination with taxon-specific TL:FL conversions. Where an animal was identified to genus or family, the Bayesian LWR was sourced from Fishbase (Froese et al., 2014).
Taxon-specific biomass estimates were calculated by multiplying the abundance of each taxon by the mean weight of that taxon. Marine mammals were excluded from the biomass estimates as they were multiple orders of magnitude heavier than the largest observed fish and heavily skewed the estimates. These four univariate metrics, TR, F I G U R E 2 Wandoo oil field schematic adapted from Vermilion Oil and Gas Australia (2014). The infrastructure at the Wandoo field includes the unmanned monopod Wandoo A, the concrete gravity structure Wandoo B, the pipeline end manifold (PLEM), and the catenary anchored leg mooring (CALM) Buoy. Not to scale TA, TB, and FL, were analyzed separately for each survey in order to ensure like-for-like comparisons between sites. Annual and seasonal variability were also assessed for each site to determine the variability in the demersal and pelagic communities at each site over time.
These analyses were also carried out at the level of survey, comparing annual variability separately for autumn and spring at each site, and seasonal variability (i.e., between spring and autumn) for each year at each site.
The prevalence of each taxon at each site was calculated by determining the percentage of seabed deployments or mid-water zones on which the particular taxa were observed of the total for that site.
The prevalence data were then used to determine the number of unique demersal and pelagic taxa for each site, by extracting taxa that were only recorded at one site. We did not count taxa which were recorded on only one mid-water zone or seabed deployment per site, in order to eliminate chance sightings and possible incorrect identifications. Within the lists of unique taxa, any taxon that was only identified to genus or family was removed if there was a record from that genus or family at another site.

| Statistical analyses
The categorized habitat data were analyzed using a Chi-square contingency test to determine whether habitat varied significantly by site (Zar, 1999). Variation in the fish assemblage was tested using PERMANOVA as it is robust to data heterogeneity (Anderson, 2017). The linear variables of TA, TB, and FL were log 10 transformed to stabilize variance (Zar, 1999). For each of these univariate measures, a Euclidean distance resemblance matrix was calculated and a PERMANOVA was applied based on unrestricted permutations (Anderson, 2017) with Site and Survey as fixed factors. Site was defined as Wandoo, Control Reef, or Control Sand, while Survey was defined as each of the six BRUVS surveys conducted in a particular season and year (e.g., Autumn 2017). Our main hypothesis was whether sites differed in their fish assemblages and the degree to which such differences varied temporally. To first determine whether sites differed, one-way pairwise PERMANOVAs were applied within each survey period. We also similarly tested for differences between years and between seasons within sites. Repeated measures ANOVA was not used as the sampling through space and time varied randomly within the zones and seasons (Zar, 1999).
The assemblage composition data were treated differently from the univariate metrics. Species composition data were pooled across all surveys in preparation for the multivariate analyses for each sampling method. The data were analyzed by survey to ensure likefor-like comparisons between sites. Multivariate analyses were completed on the pelagic and demersal taxonomic assemblage data in terms of abundance and biomass to understand variations in species composition between sites as well as which variables explained this variation. We log(x + 1) transformed the assemblage data and calculated Bray-Curtis resemblance matrices for abundance and biomass of each species. Pairwise PERMANOVAs were applied to determine the differences between the demersal and pelagic species compositions of the three sites, across all surveys, in terms of both abundance and biomass. Canonical analysis of principal coordinates (CAP) was used in order to visualize a constrained ordination of the data on the basis of distance or dissimilarity.
A database of physical, chemical, and biological variables was also compiled in order to understand the potential environmental effects on taxonomic assemblages. Distances to marine features (e.g., coral reefs and seamounts) were calculated using bathymetry data following Yesson et al. (2021). Environmental data were derived from the following datasets: • Geoscience Australia (GA) 250 m bathymetry (Whiteway, 2009); • GA Australian submarine canyons (Huang et al., 2014); • CSIRO Atlas of Regional Seas (CARS) (Ridgway et al., 2002); and A number of anthropogenic variables, such as time to market and distance from nearest human population, were also calculated based on human accessibility calculations undertaken by Maire et al. (2016). However, the three sites are almost exactly the same distance from the coast, so distance-based variables were similar for all sites, and fishing effort data were not fine scale enough to separate the three sites. As such, the anthropogenic variables were excluded.
A Pearson's correlation was run to identify highly correlated independent variables with a correlation coefficient >0.6 (Havlicek & Peterson, 1976). Analyses included only one of any highly correlated variables in a given test. A distance-based linear model (DistLM) was used to determine the relationship between these variables and the assemblage data across all surveys. All analyses were completed using the Primer 7 software package with the PERMANOVA + add-on (Anderson et al., 2015).

| RE SULTS
In the six surveys across 3 years, we counted 35,070 individuals from 358 taxa, representing 85 families (Appendices 4 and 5). The total biomass of these animals was 42.5 tons, excluding marine mammals.

| Environment
Observed habitats across the three sites, included sand, and macrobenthos which consisted of sponges, sea whips, crinoids, soft corals, and gorgonians. Macrobenthos coverage was both sparse (<50%) and dense (>50%). Habitat differed significantly across the three sites with the WN site characterized by a higher percentage of samples dominated by dense and sparse macrobenthos relative to the other two sites (X 2 (2,N=417) = 91.1, p < .001). Macrobenthos was present on 57% of the deployments at WN, with sand dominating deployments at CR and CS (60% and 99%, respectively; Figure 3).
The highest percentage of dense macrobenthos also occurred at WN (22%), compared with 15% at CR and none at CS.
There was limited environmental variability between the sites (Appendix 1). As expected, based on sampling design, depth was not significantly different between WN and CR (t 468 = 1.87, p = .06), although CS was significantly but only marginally deeper than WN and CR (t 418 = 16.8, p < .001 and t 298 = 7.87, p < .001, respectively).
Mean sea surface temperature (SST) in autumn was similar at WN and CR (t 218 = 1.18, p = .26), but was approximately one degree higher at CS than at WN and CR (t 218 = 6.15, p < .001 and t 148 = 6.19, p < .001, respectively; Appendix 1).

| Demersal richness, abundance, biomass, and fork length
The mean demersal richness was 13.1 ± 0.90 SE and ranged between 7.7 and 17.5 taxa per sample. There was significant variation in richness between sites in four of the six surveys, where richness was higher at WN than at the control site sampled in the same survey ( Figure 4a; Table 1). The only seasonal variation was in 2018 at WN, when richness was higher in autumn than spring, and the only annual variation was at CS, where richness was higher in 2018 than 2019 (Appendix 2).
Abundance ranged from 17.9 to 77.4 individuals per sample, with a mean of 43.5 ± 4.99 SE. Abundance was consistent between sites in most surveys, only differing in Autumn 2019 when abundance at WN was higher than at CS (Figure 4b; Table 1). In terms of annual variability at WN, abundance in autumn was higher in both 2017 and 2018 than in 2019. In spring, abundance was also higher in 2017 than both 2018 and 2019. There was more annual variability in abundance at WN than at CR or CS, and seasonal abundance followed the same pattern at all sites, with abundance generally being higher in autumn than spring (Appendix 2).
Mean biomass was 44.5 kg ± 3.31 SE, and ranged from 28.2 kg to 70.2 kg per sample. Similar to abundance, biomass was consistent between sites for all surveys except Autumn 2019, when biomass was higher at WN than CS (Figure 4c; Table 1). The only annual variation was at CS, where biomass was higher in 2018 than 2019.
Biomass was consistent between seasons at WN but was higher in autumn than spring at the control sites (Appendix 2).
Fork length ranged from 24.8 cm to 38.5 cm per sample, with a mean of 32.6 cm ± 1.05 SE. Fork length was consistent between sites in most surveys, but was higher at WN in Autumn 2017 and Autumn 2018 ( Figure 4d; Table 1). Fork length was generally higher in spring than autumn, and higher in 2019 at WN and CS (Appendix 2). F I G U R E 3 Percentage habitat composition for each of the three sites. The habitat types are sand (yellow), sparse macrobenthos (light green) and dense macrobenthos (dark green)

| Pelagic richness, abundance, biomass, and fork length
Mean pelagic richness was 3.9 ± 0.15 SE, with a range of 1.3 to 8.4 taxa per zone. Richness in the Autumn 2018 survey was significantly higher at WN and CS than at CR, but was consistent between sites in all other surveys ( Figure 4e; Table 1). Annual and seasonal richness was consistent in most surveys at both WN and CS, but there was significant annual and seasonal variation at CR (Appendix 3).
Abundance ranged from 1.3 to 271 individuals per zone, with a mean of 29.6 ± 4.7 SE. Abundance was consistent between sites in four of the six surveys, and was higher at WN than at CR in the other two surveys (Figure 4f; Table 1). Annual variability occurred at WN in spring, but at the control sites in autumn, and there was seasonal variability in abundance at WN and CS (Appendix 3).
Mean biomass was 48.5 kg ± 5.7 SE and ranged from 7.5 g to 429 kg per zone. Biomass was significantly lower at CR than CS in Autumn 2018, but was consistent between sites across all other surveys ( Figure 4g; Table 1). The only annual or seasonal variation in biomass was at CR, where biomass was higher in Autumn 2017 than Autumn 2018 (Appendix 3).
Fork length ranged from 3.8 cm to 182 cm per zone, with a mean of 37.5 cm ± 3.3 SE. Fork length was higher in Autumn 2017 at WN than CR, and higher in Autumn 2018 at CS than WN. Fork length was consistent between sites in all other surveys ( Figure 4h; Table 2). There was no annual or seasonal variability in fork length at the control sites, with annual variability in three of the six surveys at WN (Appendix 3).

| Community assemblages
There was strong separation of both demersal and pelagic taxonomic assemblages between sites, with abundance and biomass at each site characterized by unique species assemblages. Demersal F I G U R E 4 Mean values with standard errors (SE) for taxonomic richness (TR), and logged values of total abundance (TA), total biomass (TB) fork length (FL) by survey for demersal (left) and pelagic (right) communities at the three sites: Wandoo (green); Control Reef (dark blue) and Control Sand (light blue). Solid bars indicate autumn surveys while broken bars indicate spring surveys TA B L E 1 Pairwise PERMANOVA tests comparing demersal and pelagic variation between sites for each survey, for taxonomic richness (TR), log total abundance (log 10 TA), log total biomass (log 10 TB), and log fork length (log 10 FL)  Figure 5) and biomass ( Figure 6) ( Table 2). The DistLM analysis showed that the three environmental variables, depth, SST, and Chl-a, did not explain a sufficient proportion of the variance in the assemblage data, and as such, these analyses were excluded.
Demersal abundance (Figure 5a) and biomass (Figure 6a)  Pelagic assemblages followed similar patterns in terms of abundance ( Figure 5b) and biomass (Figure 6b) to those observed in the demersal assemblages. Abundance and biomass at WN were driven by great barracuda Spyhraena barracuda and rainbow runner Elegatis bipinnulata. Great barracuda were usually solitary, but frequently observed at WN (60% of zones, Appendix 5), while rainbow runner was observed less frequently (15% of zones) but in large schools. There was some overlap in abundance between WN and CS, characterized by herrings (Clupeidae spp.), which were observed on 25% of zones at WN and 41% at CS. Abundance and biomass at CS was driven by silky sharks Carcharhinus falciformis and live sharksuckers Echeneis naucrates, and biomass was also characterized by cobia Rachycentron canadum.
Abundance at CR was not strongly characterized by any particular species, while biomass was driven by great hammerheads Sphyrna mokarran, which was always solitary and only observed on 16% of zones. WN was the only site where any unique pelagic taxa were recorded, with rainbow runner not observed at either of the control sites (Table 3).

| DISCUSS ION
The demersal and pelagic community assemblages in the Wandoo field are distinct from those that would have existed prior to the installation of the infrastructure. The habitat around Wandoo is TA B L E 3 Abundance, biomass, and prevalence of taxa observed at a single site at WN (16 demersal and 1 pelagic species), CR (5 demersal and 0 pelagic species), and CS (4 demersal and 0 pelagic species) based on demersal and pelagic sampling records. Species marked with an asterisk are commonly caught commercially and/or recreationally in the North Coast Bioregion (Rome & Newman, 2010 (Sainsbury et al., 1993). As a result, the Wandoo demersal assemblage is characterized by reef-associated rather than sand-associated taxa. The pelagic assemblage at Wandoo is different from the other two sites, driven by species associated with offshore platforms on the NWS as well as in other regions around the world (Friedlander et al., 2014;McLean et al., 2019;Reynolds et al., 2018). Overall, the demersal and pelagic assemblages more closely resemble a natural reef than the assemblages that would have existed pre-installation, which is congruent with our hypothesis. However, the composition of these assemblages is still unique to Wandoo, suggesting the emergence of a novel ecosystem.
While the focus of our study was on the fish assemblages, we saw clear differences in the habitat at the three sites. The proliferation of macrobenthos at the otherwise flat WN site, in contrast to the barren sand habitat at CS, likely reflects the exclusion of seabed trawling at WN. The WN site also had higher demersal fish richness than the control sites in most surveys which suggests that habitat composition is a driver of diversity in these demersal communities, as has been found elsewhere on the NWS (Anon, 2019). The Pilbara Offshore mesoscale region, within which the study sites are located, is a biodiversity hotspot for sponges (Fromont et al., 2016). However, as much of the macrobenthos biomass was removed by seabed trawling (Sainsbury et al., 1993), The demersal community at WN was more diverse and reef associated than the communities at the control sites. The higher demersal richness at WN is congruent with studies from Brazil, the Persian Gulf, and Gabon, which describe offshore platforms as diversity hotspots (Fonseca et al., 2017;Friedlander et al., 2014;Torquato et al., 2017). High diversity is often associated with structural complexity of hard substrate (Friedlander & Parrish, 1998), and this association was observed in ROV surveys of the Wandoo infrastructure (Tothill, 2019 The similarity in pelagic communities across sites in terms of all four metrics was expected, given the three sites are located relatively close to each other and the highly mobile nature of pelagic species. For example, great barracuda have been shown to travel 12 km in a day and can migrate over 100 km, while silky sharks can travel up to 60 km a day (Bonfil, 2008;O'Toole et al., 2011). While these species are highly mobile, there was still strong distinction in the taxonomic assemblages between the three sites. The two species which characterized the taxonomic assemblage at WN, great barracuda and rainbow runner, are often associated with offshore platforms. Great barracuda is a commonly recorded species around offshore platforms in the Gulf of Mexico (Reynolds et al., 2018;Wetz et al., 2020), accounting for 33.2% of the biomass at offshore platforms in Gabon (Friedlander et al., 2014), and was recorded in Brunei (Chou et al., 1992;Friedlander et al., 2014;Reynolds et al., 2018 , 2020). This population of silky sharks is thought to be opportunistically targeting the discards from the commercial fishing vessels as a food source, which would explain their high abundance at a site otherwise scarce in the typical prey of this species, which includes scombrids, carangids, snappers, and groupers (Compagno, 1984 (Frederick, 1997). While it is possible that some species recruit to WN from natural sites, and certainly would have done when the platform was first installed, the high number of species unique to WN suggests that fish are being produced at the platform, rather than simply being attracted from natural habitats.
Tothill (2019)  There was only one pelagic species unique to a single site, which may reflect the relatively mobile nature of pelagic animals. Rainbow runner were only observed at the WN site, which could be attributed to the association of this species with offshore platforms around the world (Chou et al., 1992;Reynolds et al., 2018). Offshore platforms can function as fish aggregation devices (FADs), aggregating fish by facilitating foraging and school formation (Dagorn et al., 2000;Haugen & Papastamatiou, 2019). Rainbow runner are thought to primarily aggregate around FADs to prey on small FAD-associated pelagic fishes (Xuefang et al., 2013), and it is possible that the vertical hard structure at WN is providing enhanced foraging opportunity for this species.
The exclusion of fishing around WN has created a de facto MPA, as has been reported at other offshore platforms (Friedlander et al., 2014;Fujii & Jamieson, 2016;Love et al., 2006).
Seabed trawling on the NWS in the 1970s not only removed much of the macrobenthos habitat but also resulted in a significant shift in fish composition (Sainsbury et al., 1993 (Sainsbury et al., 1993;Thresher et al., 1986). This relationship between habitat and species composition was also observed in this study: macrobenthos habitat was present at WN and CR, both of which were characterized by emperors. In contrast, at CS, the habitat was almost completely devoid of macrobenthos, and the species composition was characterized by brushtooth lizardfish. Lizardfishes feed on benthic fishes, particularly on juveniles of other species, and are estimated to collectively consume 4 × 10 7 fishes per day on the NWS (Thresher et al., 1986). Demersal communities dominated by lizardfish, such as CS, would therefore have been significantly impacted by the proliferation of this genus. The de facto MPA has also resulted in a large ecological halo around the WN infrastructure. The ecological halo around offshore platforms and artificial reefs is usually around 15-34 m, with abundance and diversity similar to natural habitats beyond this distance (Reeds et al., 2018;Scarcella et al., 2011;Stanley & Wilson, 1996). In contrast, diversity at WN was higher than natural habitats at more than 50 m from the infrastructure. It is likely that the larger ecological halo at WN is due to the 500 m exclusion zone which was not present around the infrastructure in other ecological halo studies. The WN ecological halo is driven by recovery of macrobenthos habitat due to the exclusion of trawling activity.

| Wandoo as a novel ecosystem
The ecosystem in the Wandoo field clearly has novel attributes when compared with natural systems in the region; however, this assertion is not, on its own, sufficient to warrant labelling 3. Novel ecosystems are prevented from returning to their historical states by practical limitations, in the form of ecological, environmental, and social considerations. Wandoo is due to remain operational for at least a further 10 years, which is a significant social consideration as the presence of the infrastructure is central to this ecosystem. When Wandoo is decommissioned, it is possible that complete removal will allow the ecosystem to return to its pre-installation state due to exposure to trawling, but the evidence presented here on the unique ecology of Wandoo should provide an ecological consideration against complete removal, thereby preventing a return to the historical state of the site.
Based on these criteria, Wandoo may be classified as a novel ecosystem. The environment and ecology of the site have been altered, a self-organizing ecosystem with novel qualities has emerged, and the presence of the platform prevents the ecosystem from returning to its post-trawling state.

| Implications for decommissioning
We have used proxies for different decommissioning scenarios, which can provide a broad idea of how the Wandoo ecosystem might look post-decommissioning. We suggest that the Control Sand site is a proxy for complete removal, as this site is already a proxy for the Wandoo site without infrastructure. If the Wandoo infrastructure was completely removed, there would be a significant loss in diversity, particularly in terms of reef-associated species. Pelagic species associated with mid-water structure, such as great barracuda and rainbow runner, are also likely to no longer be present at this site.
Commercial and recreational fishing activity would likely recommence in the field post-decommissioning, as the petroleum safety zone would no longer be in effect and there would be no significant hard structure to prevent seabed trawling.
Topping, a second decommissioning scenario, would result in partial removal of Wandoo down to around 25 m below the surface. This method has been applied to shallow-water platforms in the United States (Ajemian et al., 2015). The reef at the Control Reef site rises to around 30 m below the surface, making this a close approximation to a topped Wandoo. This scenario would also result in the loss of pelagic species associated with structure, but would result in the retention of more of the demersal community than complete removal. There would be some losses: the shallower portions of Wandoo are important for juveniles, exhibit higher richness and abundance than deeper portions, and are characterized by small reef fish such as damselfishes (Tothill, 2019).
Indeed larval-stage juveniles were absent from the Control Reef site, and abundance of small demersal species such as damselfishes was generally lower than at the Wandoo site. It is likely that even under a topping scenario there would no longer be any exclusion of fishing activity around the remaining part of the platform. Seabed trawling could still occur in the areas surrounding the infrastructure that were previously protected by the petroleum safety zone.
Partial or complete removal of the Wandoo platform will likely have adverse impacts on a number of taxa and alter the role of the infrastructure as a novel ecosystem, specifically in terms of the artificial reef and associated ecological halo. Partial removal would be less detrimental in that it would also still afford protection to the macrobenthos from seabed trawling. However, there is significant ecological benefit in retaining the mid-water sections of the infrastructure, for both pelagic species and juvenile reef-associated species, and leaving the platform standing in place would maintain these benefits. Additional aspects that should also be considered include the role of the infrastructure for seabirds, marine megafauna, and macrobenthos communities attached to the infrastructure, as have been reported from other offshore platforms around the world Ronconi et al., 2015;Thomson et al., 2021;Todd et al., 2019). The exclusion of fishing is a critical component of the large ecological halo present at Wandoo; however, the petroleum safety zone would likely cease to exist post-decommissioning.
We would recommend that post-decommissioning protection from fishing, in the form of a no-take MPA, should be considered.
The installation of infrastructure in the Wandoo field has resulted in the emergence of a novel ecosystem with distinct ecological characteristics not found at natural sites in the region. The demersal and pelagic communities more closely resemble reef communities than those present pre-installation, but are still unique from those found at natural habitats in the region. The novel ecosystem at Wandoo also acts as a refuge for these communities, functioning as a de facto MPA in a region impacted by historical and current fishing activity.
This MPA not only protects fish communities but also has allowed the macrobenthos to recover from the impacts of seabed trawling.
Many of the novel characteristics of the Wandoo ecosystem would be lost under decommissioning scenarios that involve partial or complete removal, and the impact of decommissioning on fauna such as seabirds is still unknown. Recognizing the Wandoo field as a novel ecosystem provides a mechanism for recognizing the ecological role played by the Wandoo infrastructure, and underlines the need to consider the ecological role of each offshore platform prior to decommissioning.

ACK N OWLED G M ENTS
Our thanks to Vermilion Oil and Gas Australia ( with funds donated by Vermilion Oil and Gas Australia (Pty) Ltd. We thank two anonymous reviewers whose comments helped improve and clarify this manuscript.

CO N FLI C T O F I NTE R E S T
This project is funded by Vermilion Oil and Gas Australia (VOGA).
VOGA has 100% operating interest in the Wandoo field where this work was conducted. VOGA did not participate in the design of the study, analysis of the data, or development of the manuscript. Environmental data for each survey used in the DistLM analyses, including start and end dates, depth, sea surface temperature (SST), and chlorophyll concentration (Chl-a). Data were derived from: Geoscience Australia 250 m bathymetry (Whiteway, 2009)

A PPE N D I X 2
Pairwise PERMANOVA comparing demersal variation between years for autumn and spring at each site for taxonomic richness (TR), log total abundance (log 10 TA), log total biomass (log 10 TB), and log fork length (log 10 FL

A PPE N D I X 3
Pairwise permanova comparing pelagic variation between years for autumn and spring at each site, in terms of taxonomic richness (TR), log total abundance (log 10 TA), log total biomass (log 10 TB), and log fork length (log 10 FL. Carangoides ferdau --5 Carangoides fulvoguttatus 3.8 --