Isolated reefs support stable fish communities with high abundances of regionally fished species

Abstract Anthropogenic impacts at isolated and inaccessible reefs are often minimal, offering rare opportunities to observe fish assemblages in a relatively undisturbed state. The remote Rowley Shoals are regarded as one of the healthiest reef systems in the Indian Ocean with demonstrated resilience to natural disturbance, no permanent human population nearby, low visitation rates, and large protected areas where fishing prohibitions are enforced. We used baited remote underwater video systems (BRUVS) to quantify fish assemblages and the relative abundance of regionally fished species within the lagoon, on the slope and in the mesophotic habitat at the Rowley Shoals at three times spanning 14 years and compared abundances of regionally fished species and the length distributions of predatory species to other isolated reefs in the northeast Indian Ocean. Fish assemblage composition and the relative abundance of regionally fished species were remarkably stable through time. We recorded high abundances of regionally fished species relative to other isolated reefs, including globally threatened humphead Maori wrasse (Cheilinus undulatus) and bumphead parrotfish (Bolbometopon muricatum). Length distributions of fish differed among habitats at the Rowley Shoals, suggesting differences in ontogenetic shifts among species. The Cocos (Keeling) Islands typically had larger‐bodied predatory species than at the Rowley Shoals. Differences in geomorphology, lagoonal habitats, and fishing history likely contribute to the differences among remote reefs. Rowley Shoals is a rare example of a reef system demonstrating ecological stability in reef fish assemblages during a time of unprecedented degradation of coral reefs.


| INTRODUC TI ON
Coral reef fisheries provide sustenance and livelihoods for millions of people (Pauly et al., 2005), yet uncontrolled fishing pressure on many of the world's reefs is threatening the stability of these fisheries (MacNeil et al., 2015;Newton et al., 2007). Managing reef resilience and multispecies coral reef fisheries are often premised on an understanding of fish assemblages in the absence of other anthropogenic pressures, particularly fishing. Remote reefs, with minimal human interaction and low historical fishing pressure, offer an ecologically meaningful baseline to compare with regions facing more severe human disturbances (Knowlton & Jackson, 2008).
The low fishing activity often associated with remote coral reefs can result in fish biomass higher than accessible reefs, even those with long-standing and well-enforced no-take areas (McClanahan et al., 2019). High biomass estimates are often associated with a high abundance of larger-bodied predatory fish (Quimbayo et al., 2017;Stevenson et al., 2007) that are typically targeted by fishers (Pauly et al., 1998). The ecological benefits of high fish biomass and abundant predators may flow through the system, improving overall ecosystem health and resilience (Friedlander & DeMartini, 2002;Jackson et al., 2001;Link & Watson, 2019). The high biomass often found at isolated coral reefs is also typically associated with high functional diversity of fishes, which maintains vital ecosystem processes (Mora et al., 2011). This includes rare species, whose ecological roles can be important despite their low abundance (Graham & McClanahan, 2013;Mouillot et al., 2013). However, remoteness may not always result in diverse and abundant fish assemblages and it is important to identify variation within and among isolated reefs (McClanahan et al., 2019). The combination of isolation and protection from fishing is rare; a large proportion of remote oceanic atolls lack fisheries protection, even when they are hot spots for fish and shark populations (Cinner et al., 2016;Letessier et al., 2019).
Remoteness, while an advantage in terms of impacts from human disturbances, can also increase vulnerability. Fish recruitment at isolated reefs is often dependent on locally produced larvae (Green et al., 2015;Underwood et al., 2012), with large-scale oceanic dispersal generally reduced by geographical and physical barriers associated with large stretches of deep oceanic water (Luiz et al., 2012).
Population persistence therefore depends on maintaining levels of spawning stock biomass, as external sources of recruitment are limited. Low connectivity among oceanic atolls can also make isolated fish assemblages more susceptible to inbreeding, which promotes low genetic diversity, reduces the capacity of local populations to respond and adapt to change (Almany et al., 2009;Frankham, 1996), and may increase local extinction risk (Dulvy et al., 2003). In a rapidly changing world, isolated reefs may therefore be vulnerable to the combined effects of warming waters, coral bleaching, altered oceanographic patterns, and increased storm frequency (Hughes et al., 2018;Puotinen et al., 2020).
Reefs naturally differ among habitats and depth gradients, due to differences in physical and oceanographic factors such as temperature, water movement, primary productivity, light availability, and physical orientation (Hamner et al., 2007;Ke et al., 2018;Moore & Morrison, 2009). These environmental factors shape the composition of benthic biota and associated fish assemblages. For example, reef predators often show a preference for outer reef slopes (Dale et al., 2011;Friedlander et al., 2010), where planktivorous fish prey are abundant due to enhanced primary productivity driven by oceanic currents (Skinner et al., 2019). On deeper reefs (>30 m), fish assemblages are also often ecologically distinct, with low abundance of herbivores, a concentration of predatory fish biomass, and high abundance of planktivorous fishes (Rocha et al., 2018;Stefanoudis et al., 2019). Furthermore, although pressures such as pollution may extend throughout a coral reef (Rocha et al., 2018), the impact may vary among habitats. The impacts from heat stress and damaging waves are typically greatest on shallow reefs (Frade et al., 2018).
Deeper reefs (>30 m) are thought to provide areas where coral reef taxa can survive during periods of adverse conditions elsewhere (Bongaerts et al., 2015;Smith et al., 2014), and may assist recovery of shallow-water reefs by restocking them with larvae (Bongaerts et al., 2015;Costantini et al., 2011;Vaz et al., 2016). However, mesophotic reefs are not immune to disturbance, and if they are to act as depth refuges, they need to be resilient and sustain populations that connect with those in other reef habitats (Abesamis et al., 2018;Bongaerts & Smith, 2019;Pinheiro et al., 2019). Understanding variability across reef habitats in protected isolated reefs subject to low levels of environmental disturbance may provide further insights into the overall capacity of reefs to survive rapid ongoing change under ecologically optimal conditions (Knowlton & Jackson, 2008;Skinner et al., 2020).
The Rowley Shoals is a cluster of three oceanic atolls located ~260 km from the Australian mainland that has no permanent human population, low visitation rates, and large no-take marine reserves with regular compliance activities (MPRA, 2015). The Rowley Shoals' distance from the Australian coast and Indonesia has likely contributed to minimal historical fishing pressure in comparison with reefs further north (e.g., Ashmore Reef and Scott Reef; Edgar et al., 2017;Russell & Vail, 1988;Serventy, 1952). As a result, they are often referred to as "pristine" and are regarded as one of the healthiest reef systems in the east Indian Ocean (Allen, 2000;Field et al., 2011). The impacts of warming oceans and extreme oceanographic events such as El Niño on coral health in the Rowley Shoals are low compared with other reefs (Hughes et al., 2018). The reefs have undergone multiple cycles of impact and recovery from localized exposure to cyclones and damaging wave action, with an overall increase of ~30% in mean coral cover over the past 22 years (Gilmour et al., 2019). Despite their pristine status, studies assessing the stability of fish communities through time are limited (Ruppert et al., 2013), and none have investigated spatial differences among reef habitats and depths.
Further, we know little of how abundances of species vulnerable to fishing compare among remote reefs in the region (but see Barley et al., 2017;Bennett et al., 2018;Edgar et al., 2017).
The aim of this study was to assess fish assemblages and regionally fished species across multiple reef habitats at the Rowley Shoals as a baseline for eastern Indian Ocean coral reef atolls. We also assess potential changes in fish assemblages and regionally fished species through time and compare these regionally fished species with other research campaigns that used the same methodology to sample remote oceanic coral reefs in the region with different geomorphology and fishing pressures. We use baited remote underwater video systems (BRUVS) to survey fishes, because they are not limited by depth, provide accurate estimates of length, and capture a wide diversity of reef fishes, particularly large predators often underrepresented in diver-based visual surveys Watson et al., 2010). The objectives of this study are to (a) assess how fish assemblages at the Rowley Shoals differ among reef habitats and depths through time; (b) determine whether the assemblage and abundance of regionally fished species changed through time; (c) compare the abundance of regionally fished species at the Rowley Shoals to other remote reefs; and (d) compare the size of predatory species at the Rowley Shoals to other remote reef systems.

| Study site
The Rowley Shoals are located approximately 260 km offshore from mainland Western Australia between 17°07′S, 119°36′E and 17°35′S, 118°56′E (Figure 1). They consist of three uninhabited oceanic atolls (Imperieuse, Clerke, and Mermaid reefs) which are between 30 and 40 km apart. Clerke and Imperieuse are partially protected from fishing (24% of the marine park is zoned no-take), and Mermaid is fully protected, totaling 752 km 2 of no-take zones across the three atolls (Table 1; MPRA & DEC, 2007). Although recreational and charter fishing is permitted in some areas, most (>80%) fish caught by charter operators are released and retained species are primarily pelagic (e.g., Gymnosarda unicolor, Caranx melampygus, Thunnus albacares, and Acanthocybium solandri; MPRA, 2015). All Epinephelidae (cod/groupers) and Labridae (wrasse) species are totally protected throughout the Rowley Shoals. Each atoll is similar in size, shape, and orientation, with outer reef flat and crest enclosing a lagoon (Collins, 2011).

| Comparison sites
The relative abundance of regionally fished species at the Rowley Shoals was compared with five other remote reefs in the eastern Indian Ocean, which had been sampled using standardized BRUVS methodology (Langlois et al., 2020) (Meekan et al., 2006;Russell & Vail, 1988;Serventy, 1952). North Scott Reef has an enclosed la-

| Sampling technique and design
We assessed variation in fish assemblages across reef habitats and depth with a comprehensive survey of the Rowley Shoals in 2018.
We placed 4-5 replicate BRUVS within the lagoon (5-12 m depth), on the fore-reef slope (5-14 m) and at mesophotic depths (40-75 m) at two sites at each of the three atolls ( We sampled during daylight, at least one hour outside of crepuscular periods to minimize potential variability in fish assemblages with time of day (Birt et al., 2012;Myers et al., 2016). Each deployment was randomized in time with respect to the treatment and left to record for 60 min. The exception is the Browse Island slope samples, which were only deployed during the morning and should therefore be interpreted with caution as this may have influenced comparisons with other locations in the afternoon (Birt et al., 2012).
We discarded two deployments due to limited field of view (facing substrate) resulting in a total of 88 deployments across the three atolls.
Historical surveys conducted at the Rowley Shoals that used the same BRUV methodology were used for temporal comparisons. In 2013, sampling was completed at the same two sites within the lagoon and fore-reef slope habitats, except at Mermaid atoll, where only one site was sampled within the lagoon. There were no deployments in mesophotic habitats in 2013. In 2004, BRUVS were only deployed in the mesophotic habitat, sampling two sites at each of Clerke and Imperieuse atolls. Replicates were haphazardly placed in all years and habitats, except for lagoon sites in 2013 and 2018 where they were deployed using the same GPS coordinates.
Therefore, statistical analyses in the lagoon followed a repeatedmeasures approach.
To compare the Rowley Shoals to other isolated atolls, data were sourced from a range of different sampling programs, and therefore, the numbers of replicates per site and reef habitats surveyed differed (Table 1). However, the slope was sampled at all locations, the mesophotic habitat was sampled at Browse Island, and the lagoon was sampled at Cocos (Keeling) Island. Reef habitats were in similar depths in each location. Differences in design were accounted for in the statistical analyses (see Section 2.5). Logistical constraints meant that both the Rowley Shoals and Scott reef could only be sampled on the sheltered (eastern) sides of the atolls. Previous research (Raedemaecker et al., 2010;Floeter et al., 2007;Wilson et al., 2003) and our own preliminary comparison between exposed and sheltered sites at Cocos Island (Appendix S2) have revealed that although fish communities on sheltered and wave-exposed reefs may differ, the regionally fished species examined here are likely at similar or higher abundances at exposed locations. Therefore, we expect observations on the sheltered sides of the Rowley Shoals and Scott Reef to yield conservative estimates of regionally fished species.

| Image analysis
We used the EventMeasure software (www.seagis.com.au) to identify and count fish. To avoid repeated counts of the same fish and for standardization among historical and interinstitutional datasets, we counted the maximum number of individuals in the field of view at one time (MaxN) to estimate relative abundance (Cappo et al., 2003

| Statistical analyses
We used multivariate permutational analysis of variance We did not compare fish assemblages in the mesophotic reef habitat through time due to differences in image quality impacting species detection and identification (see Section 2.4).
Multivariate fish species abundance data were fourth-root transformed to reduce the influence of highly abundant species after data visualization using shade plots. Similarities between species were based on the Bray-Curtis resemblance matrix as it does not treat the absences of species as similarities and emphasizes the composition of the assemblage rather than the relative abundance of individual species (Anderson et al., 2008). Significance levels were obtained using 9,999 permutations of the data for each term with type III sums of squares and permutation of residuals under a reduced model in the PRIMER v7 software with the PERMANOVA add-on (Anderson et al., 2008;Clarke & Gorley, 2006). Multivariate patterns were further explored by characterization of multivariate data using metric multidimensional scaling (MDS) with vectors representing fish species with greatest influence on observed differences overlaid (Pearson correlations > 0.55 with MDS axes).
Euclidean distance resemblance matrices were constructed with raw or square-root-transformed (for heterogeneous data) relative abundances of individual species using PRIMER (Anderson et al., 2008).  (Benjamini & Hochberg, 1995;Lee & Lee, 2018). Where the number of unique permutations was low, Monte Carlo (MC) p-values were obtained (Anderson et al., 2008).
We compared length distributions among reef habitats at the Rowley Shoals and between the Rowley Shoals and Cocos (Keeling) Islands using kernel density estimates (KDEs) when at least 20 individuals were observed in each level, using the "gg-plot2" package in R to estimate the probability density function of the length-frequency data (Wickham, 2016). Bandwidths were chosen using a "plug-in" style selection that did not make assumptions about the distributions of the data (Sheather & Jones, 1991).
Outliers greater than 1,600 mm (eight sharks) affected the selection of appropriate bandwidths and were therefore removed from analysis (Bond et al., 2018). Bandwidths were then estimated using the "dpik" function in the R package "KernSmooth" (Wand, 2015).
We compared the area between the two sets of KDEs using permutations of the data as random pairs. p-values were obtained following the approach outlined by Langlois et al., (2012) using the function "sm.density.compare" in the R package "sm" (Bowman & Azzalini, 2018).

| Rowley Shoals regionally fished species abundance
The relative abundance of selected regionally fished species was consistent between years in all of the reef habitats at the Rowley Shoals, but had marked differences in abundance among habitats (Appendix S3; Figure 3). Shoals were pooled in Figure 3 for ease of interpretation with all species having similar abun-    High abundance of C. amblyrhynchos was also found in the mesophotic habitat at the Rowley Shoals and Browse Island.

| Predator species length distributions among habitats at the Rowley Shoals
Larger predatory fishes were observed in the mesophotic zone with smaller individuals observed in slope and lagoon habitats ( Figure 5).
Eight large sharks (>1,600 mm) were excluded from this analysis, of which seven were present in the mesophotic habitat. The lagoon had higher density of the smallest (<175 mm) and medium-sized (250-450 mm) fish. The slope had higher estimates of small (175-250 mm) fish, and the larger (550-750 mm) fish were most abundant in mesophotic samples. Of the three predator taxa that were recorded frequently enough for independent distribution plots, L. bohar occurred at a wide range of lengths on the reef slope with a higher proportion of smaller individuals in this habitat than either the lagoon or mesophotic habitats; larger individuals were mostly found in the mesophotic habitat. Conversely, L. olivaceus × microdon were smaller (<400 mm) in the lagoon than in the slope and mesophotic habitats.
Plectropomus spp. also had higher density of medium and smaller (<500 mm) fish in the lagoon, and higher density of large (>500 mm) fish on the reef slope.

| Predator species length distributions at Rowley Shoals and Cocos (Keeling) Islands
Length distributions for selected predator species differed between

| D ISCUSS I ON
The abundance of large-bodied and iconic species, many targeted by fishers, has not changed at the Rowley Shoals across three surveys spanning 14 years. This includes the endangered and vulnerable C. undulatus and B. muricatum (Chan et al., 2012;Russell, 2004).
The abundance of these species, as well as most other regionally fished species analyzed, was higher at the Rowley Shoals than at five other remote reef systems in the northeast Indian Ocean. This stability and high abundance of fish populations was evident in each of the distinct lagoon, slope, and mesophotic fish assemblages. This suggests that the Rowley Shoals is a rare example of a reef system that meets all the "NEOLI" (no-take, enforced, old reserves, large reserves, and isolated) criteria for successful marine conservation (Edgar et al., 2014), which are all contributing to observed temporal stability.
Coral cover has remained consistently high at the Rowley Shoals with an increase of ~20% coral cover during our 14-year sampling period despite bleaching and cyclonic activity that has caused declines on reef systems at Christmas Island, Ashmore Reef, Scott Reef, and other locations along the WA coast (Gilmour et al., 2019). This is important, as coral communities and their associated structural complexity provide habitat for many coral reef fishes (Pratchett et al., 2008), including fish recruits for which live coral is an essential habitat during early postsettlement life-history stages (Jones et al., 2004;Wilson et al., 2016). While we found that fish assemblages at the Rowley Shoals remained stable, Ruppert et al., (2013) documented changes in herbivore, corallivore, and planktivore densities following a severe cyclone in 1996 which resulted in lower coral cover (<30%) than was observed during our survey period (>50% cover; Gilmour et al., 2019). Temporal changes in fish assemblages have been documented at other remote locations in the eastern Indian Ocean, such as Ashmore Reef (Speed et al., 2018) and Scott Reef (Halford & Caley, 2009). These community changes are likely due to a combination of reduced fishing pressure at Ashmore Reef and damage to habitats by heat stress and cyclone events at both locations (Gilmour et al., 2019). These changes can effect species with key functional roles, such as herbivores and piscivores (Garpe et al., 2006;Gilmour et al., 2013;Wilson et al., 2019) and persist for extended periods, contributing to long-lasting shifts in ecosystem function .
Lagoons provide a unique environment that supports a range of coral habitats essential for many juvenile and adult fish species. The Rowley Shoals have 92 km 2 of enclosed lagoon habitat, with a variety of microhabitat shelters to facilitate recruitment of diverse reef fishes. This includes highly valued species such as P. areolatus with its juvenile phase relying almost exclusively on coral rubble habitats (Tupper, 2007) as well as B. muricatum and C. undulatus which recruit into branching corals within wave-sheltered environments (Bellwood & Choat, 2011;Hamilton et al., 2017;Tupper, 2007).
These microhabitats are characteristic of the Rowley Shoals lagoons (Morrison, 2009).
We recorded high abundances of C. undulatus and B. muricatum at the Rowley Shoals; the Cocos (Keeling) Islands were the only other location with a notable number of these species. Both species are sensitive to fishing pressure, being large-bodied, slow-growing, and late-maturing (Bellwood & Choat, 2011;Fenner, 2014;Hamilton et al., 2019;Sadovy de Mitcheson et al., 2019). The ecological roles of these iconic species contribute to healthy ecosystem function.
Cheilinus undulatus shapes benthic communities via predation on mollusks, crustaceans, and echinoids, including the crown of thorns starfish (Cowan et al., 2017;Kayal et al., 2012;Kroon et al., 2020;Randall et al., 1978), while B. muricatum is the most important bioeroder and the largest coral predator (Bellwood & Choat, 2011), holding a keystone role in the maintenance of coral community structure and stability of coral reefs (Bellwood et al., 2012). Consistent with previous studies that used underwater visual census techniques in shallow water, our study using BRUVS identified high abundances of these large iconic labrids at the Rowley Shoals and relatively low abundances at other atolls and islands (Bellwood et al., 2012;Edgar et al., 2017).
Relative abundances of other regionally fished species were generally high at Rowley Shoals. Given fishing for these species is prohibited at the Rowley Shoals, the high abundance of epinephelids (genera Plectropomus, Variola, and Cephalopholis) indicates that fishing pressure on remote reefs may have an influence on the abundance of these taxa. However, differences in geography and island morphology among these reef systems may also contribute to the differences in abundance across locations (Bennett et al., 2018).
For example, the absence of Plectropomus spp. at both Browse and Christmas Islands is likely driven by a lack of lagoonal habitats in these locations (Hobbs et al., 2014). However, the presence of lagoon habitat may not always lead to high abundance of target species, particularly in the absence of enforced protection from fishing (e.g., Hamilton et al., 2019). For example, fishing in the Cocos (Keeling) lagoon may be contributing to the lower abundances of Plectropmous spp. and Lethrinus olivaceus × microdon. Understanding the relative importance of fishing and habitat on mesopredators is complex and may be further confounded when the abundance of apex predators is high, as was observed at Cocos (Keeling) Islands (Robbins & Renaud, 2016;Speed et al., 2019).
The reason for the high C. amblyrhynchos abundance at Cocos (Keeling) is unclear and an exception to the population status of this species at a global scale (Simpfendorfer & Dulvy, 2017). Previous studies have attributed high abundances of sharks at Cocos (Keeling) to a lack of historical fishing pressure (Robbins et al., 2006). However, in this study, sharks were significantly more abundant at Cocos (Keeling) than at other reefs with long-term protection in the region.
Other factors such as geographical position, oceanography, and island morphology may be contributing to this pattern. Moreover, historically high-intensity shark fishing in Indonesia and north-west Australia may have disrupted the connectivity of shark populations with lower-latitude reefs such as the Rowley Shoals, slowly decreasing the observed populations (Momigliano et al., 2017).
As at many reefs around the world, there are clear differences in fish assemblages and regionally fished species among habitats at the Rowley Shoals. Spatial differences in community structure relate to different environmental conditions and regimes of disturbance.
Accordingly, programs tasked with monitoring the effect of stressors on coral reefs typically focus on areas of high ecological value, where impacts are likely to be the greatest (Emslie et al., 2008;Frade et al., 2018;Simpson et al., 2015) while recognizing that this may not be representative of the entire fish assemblage from that reef.
This approach may be problematic if species move between habitats or migrate to deeper depths following disturbance. We show that BRUVS are a useful tool for understanding relative abundance and fish lengths across a range of habitats, including the deeper mesophotic zone which is generally not accessible to divers but is an important habitat for many species (Lindfield et al., 2016).
The use of BRUVS in monitoring programs may also broaden our understanding of fish migrations. Lagoons are often thought to be key habitats for juvenile fish, which then move onto the reef slope and into deeper water as they increase in size and age (Skinner et al., 2020 (Friedlander & DeMartini 2002;Heenan et al., 2019;Stevenson et al., 2007). However, the many complex environmental and physical factors that make a reef system will cause differences in the natural state of a reef in the absence of humans and further research will be required to disentangle the key drivers of the differences detected in the predatory fish lengths seen here .

| CON CLUS ION
The Rowley Shoals is a rare reef system that meets the criterion set for global conservation targets (Edgar et al., 2014). This study demonstrates temporal stability in fish assemblages (5 years) and regionally fished species (14 years) in lagoon, slope, and mesophotic habitats. The Rowley Shoals also has high abundances of regionally fished species compared with other isolated reefs in the region, including endangered and vulnerable C. undulatus and B. muricatum.
Lagoon habitats appear to be important for supporting these and other species, especially during the early stages of their lifecycle.
However, variability in abundances and length distributions across reef habitats suggest there are varied ontogenetic shifts and habitat preferences among species, some of which include mesophotic habitats. Differences among the locations seen here are likely due to multiple factors such as geomorphology, geographical location, and historical fishing pressure. Importantly, isolation from human populations does not necessarily translate to conditions that support stable fish assemblages and abundances of regionally fished species, further highlighting the rarity of reef systems like the Rowley Shoals and the importance of developing meaningful baselines to quantify the impacts of disturbances on coral reefs.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used in this study are available at GlobalArchive (https:// globa larch ive.org/geoda ta/data/campa ign/get/991).