Until recently, research, monitoring, and management of coral reef fishes primarily relied on data collected using underwater visual censuses on open-circuit scuba between 0 and 30 m (herein denoted as “shallow water” in the context of this study). However, many fishes present in shallow water habitats are depth-generalists, able to reside in “mesophotic” depths of 30–150 m or more (Thresher and Colin, 1986; Ginsburg, 2007; Brokovich et al., 2008; Kahng et al., 2010; Slattery et al., 2011; Bridge et al., 2013; Bejarano et al., 2014). In addition, while not fully protected from environmental or biological disturbance events, mesophotic coral ecosystems (“MCEs”) may be partially shielded from some of the influences impacting shallow water coral reefs, and serve as population reservoirs for depth-generalists targeted by fishers in 0–30 m depths (Glynn, 1996; Riegl and Piller, 2003; Bak et al., 2005; Lesser et al., 2009; Bongaerts et al., 2010; Slattery et al., 2011; Kane et al., 2014; Lindfield et al., 2014, 2016; Tenggardjaja et al., 2014; Baker et al., 2016). Conversely, while shallow water coral reefs and associated habitats (e.g., pavement or rubble flats) may shelter depth-restricted specialist fishes incapable of inhabiting deeper depths, MCEs and other deep-water mesophotic benthic habitats (“MBHs”) can likewise host distinct communities and species of reef fishes not found in 0–30 m depths, with depth, habitat type, structural complexity, and biotic cover acting in concert with geographic extent and oceanographic drivers to structure assemblages and functional-level groupings (Thresher and Colin, 1986; Beukers and Jones, 1998; Brokovich et al., 2008; MacNeil et al., 2009; Harvey et al., 2013; Komyakova et al., 2013; Jankowski et al., 2015; Andradi-Brown et al., 2016; Heyns-Veale et al., 2016; Rosa et al., 2016). As a result, limitations of many marine science research programs include missing portions of reef fish populations that are utilizing deeper habitats, or omitting species of potential conservation or management importance that are restricted to mesophotic depths. However, despite increased mesophotic research over the past two decades and the potential importance of these systems, Pacific MCEs remain understudied and relatively unassessed in comparison with their shallower counterparts (Bridge et al., 2013; Kahng et al., 2014).

Historic mesophotic research in the Main Hawaiian Islands (“MHI”) has primarily focused on exploration (Brock and Chamberlain, 1968; Strasburg et al., 1968), surveys of specific taxa or target species (Grigg, 2004; Kahng and Grigg, 2005; Tenggardjaja et al., 2014), habitat and zone characterization (Locker et al., 2010; Rooney et al., 2010; Blyth-Skyrme et al., 2013; Costa et al., 2015; Veazey et al., 2016), spatially focused investigations, e.g., host reef fish assemblages inhabiting mesophotic black coral (Antipathes) patches and deep artificial reefs (Grigg, 1965; Moffitt et al., 1989; Boland and Parrish, 2005), or the effects of localized environmental variables and depth-delineated reef fish functional groupings (Kane and Tissot, 2016). While Fukunaga et al. (2016) characterized reef fish assemblages, functional groups, and endemism patterns holistically across shallow water to mesophotic gradients in the Northwestern Hawaiian Islands (NWHI), similar investigations in the MHI remains comparatively sparse.

To a large degree, the lack of MCE research and monitoring, in comparison with shallow water coral reef ecosystems, has been due to logistical, technical, and financial constraints associated with the use of technical mixed-gas or closed circuit rebreather (CCR) diving and the use advanced remote sampling technologies, e.g., submersibles (Pyle, 1996, 2000; Kahng et al., 2010). The advent of baited remote underwater stereo-video (stereo-BRUVs, herein denoted as “BRUVS”) represents an alternative, cost-effective approach that has been increasingly used to assess MCE reef fish populations (Pearson and Stevens, 2015; Lindfield et al., 2016). Here, we analyzed BRUVS data collected from 107 sites around the MHI, with the objective of characterizing changes in reef fish community structure from shallow water to mesophotic depths. We focus on changes to overall reef fish assemblages and functional-level partitions along depth gradients and their relationships to a range of habitat variables (Boland et al., 2011; Kane and Tissot, 2016). We also explore the potential for mesophotic depth-refugia of reef fish target species (those subjected to commercial or recreational fishing extraction) and whether relative abundance of endemic species or the proportion of the fish community they make up changes with depth (Kane et al., 2014; Fukunaga et al., 2016; Kosaki et al., 2016).

A total of 5,583 fish belonging to 36 families were recorded over 107 BRUVS sites (Figure 1). Herbivores, planktivores, and mobile invertivores were the largest components of fish assemblages (Figure 2 and Figure S1), ranging between 86–93% of total abundance and 74–82% of species richness at hard-bottom sites, and 77–83% abundance and 59–79% of species richness at unconsolidated sediment sites, depending on depth. Remaining groups constituted between 0.5–4% of total abundance per depth-habitat combination, with the exception of piscivores (8%, 53–100m unconsolidated sediment sites, Figure S5B) and generalist macropiscivores (9–15% at unconsolidated sediment sites in upper and lower mesophotic zones, Figure 8D). We therefore focused analysis on the three prominent functional groups, along with generalist macropiscivores.

This study provides the first in situ BRUVS assessment of reef fish communities across shallow water to mesophotic zones in the main Hawaiian Islands. Depth, habitat complexity, macroalgal cover, and unconsolidated sediment acted as influential reef fish assemblage drivers. While a variety of other potential environmental co-contributors, ranging from temperature, hydrodynamics, and sedimentation could also affect the distribution of organisms in mesophotic depths, these remain unaddressed in the scope of this work (Locker et al., 2010; Kahng et al., 2014).

Community shifts similar to patterns observed in other tropical (e.g., Red Sea, Marshall Islands, Puerto Rico, Honduras) and sub-tropical ecosystems, (e.g., South Africa, NWHI), included declines in herbivore abundance with depth, even in mesophotic habitats hosting high levels of macroalgal cover (Thresher and Colin, 1986; Feitoza et al., 2005; Brokovich et al., 2010; Bejarano et al., 2014; Kane et al., 2014; Andradi-Brown et al., 2016; Fukunaga et al., 2016). While planktivore relative abundance and richness measures were relatively similar across depth strata (Figure 6B), their proportional abundances was highest in the lower mesophotic zone (Figure S1), aligning with depth-based planktivore density and/or biomass peaks recorded in other mesophotic studies (Thresher and Colin, 1986; Feitoza et al., 2005; Fukunaga et al., 2016).

The decline of herbivores outside of shallow waters, and increased numbers of mobile invertivores in the upper mesophotic zone, indicate possible shifts in benthic primary productivity sources and compels additional nutrient cycling research between depth and habitat strata (Hilting et al., 2013; Fukunaga et al., 2016). In addition, while isotopic evidence suggests predators remain heavily reliant on resources between 0 and 30 m acting as nutrient conduits to mesophotic depths (Meyer et al., 2001; Wetherbee et al., 2004; Papastamatiou et al., 2006, 2015; Hilting et al., 2013), marked increases in schooling mobile invertivores (e.g., goatfishes) and planktivores (e.g., C. jordani, Decapterus macarellus) may serve as deep-water prey-bases for predators in the MHI (Smith and Parrish, 2002). As with schooling behaviors observed on Halimeda meadows (see below), potential prey species were observed seeking shelter in algal canopies or retreating to hard-bottom interstitial spaces during transits by generalist macropiscivores—particularly jacks and A. virescens—within the BRUVS frame of view (J. Asher, pers. obs.).

Of particular interest were mesophotic sand flats and lower-complexity hard-bottom habitats which hosted extensive calcareous Halimeda meadows, which generally harbored more limited reef fish communities in comparison with more structurally complex habitats. The abundance of mobile invertivores (70% of total community), rather than herbivores, was noted in these areas as Halimeda are relatively unpalatable to most herbivorous reef fishes (Lewis, 1985; Spalding, 2012). We observed generalist macropiscivores foraging in these habitats Figures 12B–D), with individuals or small-mixed groups of C. melampygus, C. orthogrammus, Caranx ignobilis, and/or A. virescens seen transiting though Halimeda meadows, typically with prey species fleeing or seeking shelter in algal canopies (J. Asher, pers. obs). Juvenile tiger sharks (Galecerdo cuvier) were also observed transiting along Halimeda sp. meadows, indicating possible habitat use as part of their generalist approach to feeding on a wide variety of potential prey items (Werry et al., 2014). Finally, juvenile bottom-fish P. filamentosis (9–30 cm) were also seen schooling in mesophotic Halimeda meadows in the Maui-Nui region in as little as 54 m, suggesting that those habitats may be foraging or refuge areas used by bottom-fish, prior to ontogenic migration into deeper habitats upon maturity.

Despite the apparent linkages maintained by depth-generalists, particularly those inhabiting shallow water and upper mesophotic strata, (Tenggardjaja et al., 2014; Papastamatiou et al., 2015), community linkages and species movements between strata are still largely uncharacterized in the MHI. This remains an important focus for future MHI research, particularly for those species subject to high fishing pressures around human population centers. The majority (70%) of target species were encountered in mesophotic depths, including all but one of the species where there is strong evidence of fishery depletion based on shallow water surveys (Figure 10), albeit with the majority of species having lower overall mesophotic abundance levels than in 0–30 m. Generalist macropiscivores, which remain one of the more susceptible groups to fishing pressures, had ~50% greater abundance on hard-bottom substrates in the upper mesophotic zone compared to shallow water. Higher mesophotic abundances of some groups, and changes in predator communities is a potential indicator of depth refugia (Thresher and Colin, 1986; Bejarano et al., 2014; Andradi-Brown et al., 2016; Lindfield et al., 2016), particularly for the fishery-depleted target species that were more abundant in mesophotic depths (e.g., A. virescens, Caranx ignobilis, Figures 8D, 10), or for redatory species with larger mesophotic body sizes (e.g., C. melampygus, Figure 9C). Given that assessments of MHI reef fish stock exploitation rates and annual catch assignments remain largely constrained to fishery-independent, open-circuit diver depths (i.e., ≤30 m) or fisheries-dependent, commercial catch/recreational survey data obtained from indeterminate depths (Nadon et al., 2015), the use of BRUVS serves as a promising, shallower compliment to the deep-water camera system (BotCam) utilized for Hawaiian bottom fish stock assessments (Merritt et al., 2011). Finally, additional research parsing depth refugia vs. ontogeny effects in structuring reef fish communities would be beneficial, as larger-sized mesophotic planktivores (Figure 9B) could be attributed to ontogenic migrations (Andradi-Brown et al., 2016) and may be less prone to mesophotic predation in lower-complexity habitats that do not provide adequate shelter to more vulnerable size classes.

Mesophotic reef fish community breaks have been proposed at ~60 m (Slattery et al., 2011; Fukunaga et al., 2016), with community, functional group, and assessed environmental structural outputs generally supporting this premise. However, community compositions were largely distinct between shallow water and upper mesophotic zones, indicating the potential for depth-based refugia may be limited to depth-generalists and not depth-zone specialists, e.g., specific mobile invertivores and generalist macropiscivores that are equally, if not more abundant in 30–53 than in <30 m, with the possibility of refugia further declining when transitioning to more comparably depauperate lower mesophotic communities. While mesophotic reef fish communities may provide meaningful refugia for some species, shallow-water specialists are clearly unlikely to be able to benefit in that way (Fukunaga et al., 2016). Conversely, given that many MHI depth generalists decline when transitioning from shallow water to mesophotic systems seen here and in other studies (Pyle et al., 2016), mesophotic reef fish communities affected by hypothetical biological impacts (e.g., lionfish invasions in the Atlantic) or anthropogenic perturbations (e.g., dredging) may end up reliant on shallow water systems for repopulation. Outside of a small number of species investigations (e.g., Chromis verater), the movements of fish larvae between shallow water and mesophotic strata remains largely unknown and remains an important focus for future research (Tenggardjaja et al., 2014).

MCEs in the NWHI appear to be reservoirs of extremely high levels of endemic biodiversity (Kane et al., 2014; Kosaki et al., 2016). In our study, BRUVS showed comparable MHI shallow water and upper mesophotic endemism levels as those documented by underwater visual census surveys in <30 m in the MHI (Randall, 1998; DeMartini and Friedlander, 2004; Pyle et al., 2016). However, declines in proportionate abundance (excluding two schooling species) and richness (Figures S3B–E) in the lower mesophotic zone conflict with patterns of those seen during technical dive surveys in the NWHI and submersible/technical dive surveys in the MHI. This may be attributed, in part, to these studies targeting specific habitats, e.g., hard-bottom, structurally complex slopes and ledges (Kane et al., 2014; Kosaki et al., 2016) or the Leptoseris sp. beds in the Maui-Nui region known to host large, localized endemic populations of fishes. In contrast our MHI mesophotic BRUVS surveys tended to sample mostly lower complexity habitats, e.g., low-lying aggregate Montipora reefs, rubble flats, and sand flats, which appear to be the most common habitats in those depths at our study locations.

Results from BRUVS surveys in the Main Hawaiian Islands should be interpreted with several caveats. Largely as a byproduct of the incomplete bathymetric data, the majority of lower-mesophotic deployments occurred on 100% unconsolidated sediment (27 sites), with only 10 sites sampling hard-bottom habitats (aggregate reef, aggregate patch reefs, consolidated rubble flats). None of our deeper sites were of the, in some places extensive, 90 m+ Leptoseris coral communities that have been shown to host rich reef fish communities in the Maui-Nui region (Costa et al., 2015). As a result, several functional groups (e.g., corallivores) were largely absent in our surveys outside of 53 m, although we know they can be abundant in some habitats.

As with all underwater visual censuses, BRUVS rare subject to possible sampling biases. These include the potential inflation of density estimates due to fish being drawn from outside visible sampling areas, unknown areas of attraction as a byproduct of variable bait plume dispersion, alteration of fish behaviors, competitive exclusion, and/or preferential sampling of predator and scavenger populations with corresponding reductions to other functional groups (Harvey et al., 2007; Colton and Swearer, 2010; Dorman et al., 2012). However, comparisons between baited and unbaited camera stations have shown that while carnivore and scavenger abundances tend to increase in the presence of bait, no commensurate changes are typically detected in herbivore or omnivore abundances (Watson et al., 2005; Harvey et al., 2007). The lack of bait-induced declines among non-carnivorous functional groups could be explained by possible “sheep effects,” whereby species not directly attracted to bait plumes are attracted to the feeding activities of others around BRUVS, or conspecific social attraction behaviors (Watson et al., 2005, 2010; Harvey et al., 2007; Dorman et al., 2012). Finally, underwater sampling visibility was, in general, much higher than the required BRUVS sampling minimum (7 m), even in mesophotic depths to 100 m. While the authors detected no depth-associated, functional group or species-level behavior alterations as a result of reduced light attenuation in deeper strata, coral reef fishes are known to exhibit behavioral shifts in response to varying light levels, which merits additional consideration for future mesophotic research (Rickel and Genin, 2005).

Mesophotic reefs remain infrequently explored throughout the Indo-Pacific region and likely still host numerous undiscovered fish species (Pyle, 2001). While shallow water and upper mesophotic reef fish communities are highly connected, these zones have their own distinct functional group assemblages, becoming even more dissimilar in lower depths. In light of anthropogenic and climate-based pressures in shallower waters, coupled with substantial data-gaps for the many reef fish species present in both shallow water and mesophotic habitats, there is a strong need to continue research into depth zone connectivity, along with species and life-stage distributions, in order to develop appropriate and comprehensive coral reef resource management strategies.

JA, IW, and EH conceived and designed the experiments. JA performed field experiments, analyzed the data, wrote the main manuscript, and prepared all figures unless otherwise specified. JA and EH contributed materials and analysis tools. All authors reviewed the manuscript.

Funding for ship time, support staff, and field materials/consumables was provided by the National Oceanic and Atmospheric Administration (NOAA), National Marine Fisheries Service, Pacific Islands Fisheries Science Center (PIFSC) Director's Office, the Ecosystem Sciences Division, Science Operations Division, and the Fisheries Research and Monitoring Division. Funding for camera equipment was provided by the University of Western Australia and Curtin University. Videographic processing and analysis were funded by an IPRS/APA Scholarship, the NOAA Coral Reef Conservation Program, the NOAA PIFSC Ecosystem Sciences Division, Science Operations Division, and the Pacific Young Scientists Opportunity Program. All funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.