Comprehensive spatial distribution of tropical fish assemblages from multifrequency acoustics and video fulfils the island mass effect framework

Tropical marine ecosystems are highly biodiverse and provide resources for small-scale fisheries and tourism. However, precise information on fish spatial distribution is lacking, which limits our ability to reconcile exploitation and conservation. We combined acoustics to video observations to provide a comprehensive description of fish distribution in a typical tropical environment, the Fernando de Noronha Archipelago (FNA) off Northeast Brazil. We identified and classified all acoustic echoes into ten fish assemblage and two triggerfish species. This opened up the possibility to relate the different spatial patterns to a series of environmental factors and the level of protection. We provide the first biomass estimation of the black triggerfish Melichthys niger, a key tropical player. By comparing the effects of euphotic and mesophotic reefs we show that more than the depth, the most important feature is the topography with the shelf-break as the most important hotspot. We also complete the portrait of the island mass effect revealing a clear spatial dissymmetry regarding fish distribution. Indeed, while primary productivity is higher downstream, fish concentrate upstream. The comprehensive fish distribution provided by our approach is directly usable to implement scientific-grounded Marine Spatial Planning.

Video data. To identify species and bottom habitat characteristics, four different optical systems were deployed: (i) a towed video camera; (ii) a video camera fixed on the transducer support close to the surface; (iii) a video camera deployed vertically; and (iv) a remotely operated vehicle (ROV) (see Supplementary Fig. S1 online). The towed video camera and the video camera fixed on the transducer were used to capture videos along transects. Both provided a view of the water column and allowed for substratum identification in shallow water. The towed video was set on a downrigger to deepen the camera, which looked downwards and dragged with a 15 m long rope at a vessel speed of 1.5 m s −1 . The videos captured from the camera fixed on the transducer support were especially useful in very shallow waters. Vertical videos and ROV were deployed during stations whose location was driven by observations of important quantity of fish on the echogram with the purpose of species identification (Fig. 1). Vertical videos were made using a camera fitted on a fishing line. The ROV was operated from the vessel with live stream video.
Towed, fixed and vertical videos were captured with a GoPro TM Hero3+ operating in HD at 1080 p and 30 frames per second during FAROFA 1 and 2 and 60 frames per second during FAROFA 3. ROV videos were performed using a Blue Robotics BlueROV2 system operating in HD at 1080 p and 30 frames per second. To synchronize acoustic and video observations form the towed video, a delay of 9.7 s was subtracted to the video time to adjust with the echosounder time. Each video was annotated using the Solomon Coder software 85 to identify and enumerate observed species and sediment characteristics classed over nine types (Table 1 and see  Supplementary Table S1 online). For the species censuses, we used the maximum number of individuals (MaxN) of a given species present in a single video frame 86 . In video stations, the MaxN was directly used on each frame to avoid double counting of individuals. In video transects, the MaxN over 3 s of record was used. To estimate the abundance of each taxon observed by video, we used the sum of the maximum values of the MaxN of each video (TMaxN). www.nature.com/scientificreports/ Acoustic data. Acoustic data were collected continuously throughout the survey with two SIMRAD EK80 echosounders connected to two 7° split beam transducers centred on the frequencies of 70 and 200 kHz and operated simultaneously in narrow band (continuous wave) transmission. Transducers were attached with a stainless-steel pole to the port side of a 10-m-long sport fishing boat. The ping rate was set to 'maximum' for a maximum acquisition range of 100 m (over the continental shelf) and to 1 ping s −1 off the continental shelf, where the maximum acquisition range was set to 400 m. Vessel speed was ~ 2.5 m s −1 during acquisition of acoustic data and pulse duration was set at 1.024 s. Acoustic data were converted to HAC files using Hermes software 87,88 . Processing was completed using the Matecho tool 89 and Movies3D software 88 . Details on acquisition and calibration parameters as well as on acoustic pre-processing steps from data acquisition i.e., data conversion, bottom detection, filtering and manual cleaning are available in Salvetat et al. 90 .
Acoustic data processing. To discriminate fish echoes from other organisms, multifrequency approaches, generally rely on the property that swimbladder-bearing fish have, well beyond the resonance of their swimbladder, high and homogenous backscattering response to frequency 91 . To discriminate between scatters attributed to fish (fish-like) and those originated by other organisms (no-fish), e.g. gelatinous and crustaceans, we developed an approach based on thresholds on (i) volume backscattering strength Sv (Sv, in dB re 1 m −1 ; see 92 for acoustic definitions), (ii) the bi-frequency sum of Sv, and (iii) the variance of Sv. See Supplementary methods and Supplementary Figs. S2 and S3 online for a detailed description of the methodology. To study the horizontal distribution of fish-like and no-fish echoes, we used the Nautical Area Scattering Coefficient (NASC or s A , in m 2 nm −2 ) 92 , an index of acoustic biomass, for each ping integrated over the water column. Since fish-like and no-fish data were highly correlated at the two frequencies, only the 70 kHz echograms were used for further analyses.
Combining surveys. The three FAROFA surveys were conducted at different seasons and years. To determine whether the surveys could be combined to provide a more comprehensive picture of fish distribution, we verified if, locally, s A values were sufficiently similar between surveys with regards to the natural variability observed within surveys (see Supplementary Fig. S4 online). Punctual comparisons were not possible given that observations of the different surveys were not located at the same geographical points. We thus selected the pixels containing observations from different surveys at a pixel sizes of 100 m (Fig. 2). The inter-survey comparison was based on the difference between the mean log10(s A + 1) of different surveys in the selected pixels. Meanwhile, www.nature.com/scientificreports/ the intra-survey variability was computed by selecting records at least one hour apart within the same grid cell. The inter-survey differences of fish-like log10(s A + 1) were centred on 0 ( Fig. 2a) whatever the pair of surveys considered or the pixel size, and were comparable to intra-survey ones. Based on these results indicating a strong spatial stability in the horizontal fish-like log10(s A + 1) distribution around FNA over years and seasons, we combined the data from the three surveys to provide a comprehensive spatial coverage.

Spatial distribution.
To interpolate the horizontal distribution of fish-like and no-fish acoustic biomass outside acoustics transects around FNA, we applied a tailored geostatistical approach. Spatial interpolation was adapted to face the fact that the domain area was elliptical with radial transects (Fig. 3a,b). The geographic reference system was thus irrelevant to describe the orientation between observations. For instance, North-South did not mean the same thing in different parts of the survey area. The relevant orientations to consider were rather parallel or perpendicular to the coast, which required projecting the data in a system conformal to these two main orientations.
To unfold the sampling area, we covered the domain by a series of trapezes that were then aligned and resized one by one so that the distance perpendicular to the coast ranged from 0 (coast line) to 1 (offshore border of the trapezes), and the distance parallel to the coast ranged from 0 (beginning of the first trapeze chosen conventionally) to the sum of the length of the bases of the trapezes ( Supplementary Fig. S5 online). This projection was bijective so that we could move back and forth between the geographical space and the unfolded space. In particular, the s A and the kriging grid cells whose coordinates were defined in the geographical space were projected in the unfold space to compute their variogram and their kriging values. To avoid border effects at the edge of the unfolded system, the starting and ending trapezes were duplicated prior to kriging. So, the interpolation of the left side of the first trapeze was made taking into account the data of the last trapeze also. Interpolations were performed over regular cells of 55 m longitude × 44 m latitude.
Fish assemblages. Despite the simultaneous acquisition of acoustic and video, except for two triggerfish species, it was not possible to allocate each fish-like echo to a given species. However, consistent fish assemblages with characteristic echotypes were observed on echograms (Table 2). To attribute each fish-like acoustic scatter to an assemblage, we labelled all fish-like echoes. Label assignment, hereafter called "labelling", was based on video observation and the presence of characteristic structures in echograms. Video footages made it possible to identify the species observed simultaneously by the cameras and the echosounder. This experience was then used to label the echograms not monitored by videos. For the three cruises, 70 kHz fish-like echograms were labelled manually by the same operator using the software Matecho 89 , which allows drawing polygons to encompass scatters corresponding to a given assemblage. All fish-like echoes inside a polygon were allocated to a given assemblage. In total ten assemblages were defined ( Table 2). In addition, two species, the black triggerfish Melichthys niger and the ocean triggerfish Canthidermis sufflamen could be identified on echograms due to their characteristic shoal shape. The black triggerfish forms large loose shoals occupying the whole water column distributed over the shelf from 6 to 40 m depth exhibiting different body orientations. Ocean triggerfish form smaller looser shoal generally found on deeper depth (~ 17 to 70 m) close to the shelf-break. The fish-like s A of each label corresponding to the 10 assemblages and the two triggerfish species was echointegrated over the water column by 25 m-long elementary sampling distance unit (ESDU).   www.nature.com/scientificreports/ to one). This became a real issue given the very large number of s A data available. To solve this problem, the five spatially mutually correlated indicators variables were transformed into five factors called Min-max Autocorrelation Factors-MAF 101 . These factors are linear combinations of the input georeferenced variables, and are uncorrelated at null and at short distances. Assuming that MAFs were also uncorrelated for larger geographical distances, allowed using them independently from the others. In this context, we performed the global estimation of each of the five MAFs over the study polygon by global kriging 99 and then recombined them to get the kriging estimates of the mean overall s A together with its estimation variance 102 . Using all black triggerfish labels, we delineated its area of main presence, concentrated in the east side of FNA. In this area, based on the estimation of the mean overall acoustic value described above, we estimated the biomass by Eq. (1): where the Surface corresponds to the total surface of the delineated area (in m 2 ), W is the triggerfish mean weight estimated at 485 g for 27.8 cm long black triggerfish and TS is the target strength of the black triggerfish at 70 kHz (TS = − 39.3 dB for 27.8 cm long black triggerfish, the mean size during the surveys 90 ).
Environmental drivers. We investigated the relationships between the fish-like and no-fish acoustic biomasses to a series of categorical environmental variables: • Wind/current exposure: FNA was categorised in two sides, leeward and windward, according to the exposition to main winds and currents. Indeed, FNA is under the trade wind regime and washed by the central branch of the South Equatorial Current (cSEC) that both flow from east to west 103,104 .  (Table 1).
We used both univariate and multivariate statistics to relate the distribution of the acoustic biomasses log10(s A + 1) of fish-like and no-fish data as well as the acoustic biomasses of the ten assemblages plus the two triggerfish species to the environmental factors. To seek for significant differences in acoustic biomass according to each environmental factor, we used a non-parametric Kruskal Wallis test followed by pairwise Wilcoxon tests since the distribution of the data did not follow a normal distribution.
Classification And Regression Trees (CART 105 ) were used to explore the relationships between the fish-like acoustic biomass and the fish assemblages (plus the two triggerfish species), and environmental variables. In the first case, we used the rpart package 106 while we used the diet package 107 in the second case. The diet package, originally designed to study diet composition, allows a non-parametrical exploratory and predictive approach for identifying complex relationships between environmental variables and assemblages composition. Classification tree using the diet package uses a bootstrap technique similar to Breiman 108 and Kuhnert et al. 109 . The diet package also allows visualizing the bagged predictions by mapping the predictions 110 . Both trees were pruned to the smallest cross-validated relative error 105 .
In addition to the categorical environmental factors, we added two continuous explicative variables, the nofish acoustic biomass and the bottom local slope (absolute value of the difference between the first and the last depth in a given ESDU relative to the length of the ESDU expressed in %). All statistical analyses were performed with R 111 .
All statistical analyses were performed twice, including and not the sediment types. Indeed, the sediment types were extracted from video observation and were thus not available for all ESDU but mostly restricted to the shallow areas where videos observations are available and the sediment observable.
Fish-like and no fish acoustic biomass. The geostatistical interpolation of fish-like s A reveals a heterogeneous distribution with the presence of several hotspots, mostly on the windward (east) side of FNA, in the vicinity of the shelf-break (Fig. 3c). The lowest fish-like acoustic biomass was observed on the north-western side, in particular at the mid-shelf. The pattern was different for the no-fish acoustic biomass that was concentrated off the shelf-break with a main aggregation on the leeside (Fig. 3d).
The regression tree relating the fish-like acoustic biomass to the environmental parameters without considering the sediment (Fig. 4a) reveals that the main driving factor is the wind exposure with much higher acoustic biomass per ESDU windward (log10(s A + 1) = 1.2 m 2 nm −2 ) than leeward (0.58 m 2 nm −2 ). A second split occurs in the windward side with higher acoustic biomass (1.3 vs. 0.35 m 2 nm −2 ) when the slope is greater than 0.08% meaning that fish-like acoustic biomass is very low in flat areas. When considering the reduced database (restricted to the neritic zone) containing sediment (Fig. 4b), the first explicative variable is the sediment. The most complex sediment (SaStCoRhAl) encompasses a much higher biomass (2.3 m 2 nm −2 ) than the others (0.93 m 2 nm −2 ). For sediments other than SaStCoRhAl, the next splits are wind exposure, sediment types and no-fish biomass with the highest fish-like biomass (1.9 m 2 nm −2 ) distributed windward where the no-fish acoustic biomass ranges between 2.3 and 1.6 m 2 nm −2 and over the more complex sediments (SaCoRhAl, SaLrAl, LrAl, SaUn).
Univariate analyses provide some additional information (see Supplementary Fig. S6 online). Fish-like and no-fish acoustic biomasses significantly varied according to wind and current exposure, protection levels, sediment types, and bottom depth strata. Indeed fish-like acoustic biomass was significantly lower on the leeward even if some ESDU encompassed very high acoustic biomass (log10(s A + 1) up to 6.8 m 2 nm −2 ) in presence of small pelagic schools distributed in the upper mesophotic zone (40-60 m) (see "Fish assemblages"). On the opposite, the no-fish acoustic biomass was slightly (but significantly) lower windward (2.1 m 2 nm −2 ) than leeward (2.2 m 2 nm −2 ).
Video observations allowed a good identification of the species present for most of the groups. However, the composition of four groups (bottom weak fish detection, sand fish, loose school, shelf-break school) could not be fully validated by the videos.
Small pelagic school presented the highest total acoustic biomass, followed by the black triggerfish (Fig. 5). The percentage of ESDU with presence of a given assemblage also mostly followed the trends in acoustic biomass with some notable exceptions. Small pelagic fish school that encompassed the highest acoustic biomass was only observed in 1% of ESDU (Fig. 5), indicating that they were concentrated within some large schools. On the opposite, M. niger was the most frequently observed assemblage (in 9.9% of ESDU) followed by bottom weak fish detection (9.4%) that ranked eighth in terms of total acoustic biomass.
Fish assemblages presented specific spatial patterns of distribution (Fig. 6). Four assemblages (bottom small fish school, bottom weak fish detection, individual demersal and mix reef fish) presented the most comprehensive distributions over the shelf, all around FNA. The other assemblages associated to the shelf were loose school and sand fish, mostly distributed close to the coast and M. niger, mainly distributed on the windward side of FNA. Small pelagic schools were distributed both over the shelf and at the shelf-break. The other groups were mostly associated to the shelf-break, with shelf-break schools and shelf-break large fish distributed all over FNA while small pelagic and predators and C. sufflamen were mostly distributed on the windward side.
The regression tree relating the fish-like acoustic biomass to the environmental parameters without considering the sediment (Fig. 7a,c) reveals that the main driving factor is depth strata, discriminating between areas shallower (upper and lower euphotic, upper mesophotic) and deeper than 60 m depth (mid and lower mesophotic). Bottom depths shallower than 60 m correspond to the shelf where neritic assemblages dominate: bottom small fish school, mix reef fish, bottom weak fish detection, M. niger. Pelagic assemblages logically dominate in deeper areas: small pelagics and predators, shelf-break large fish and shelf-break school. Over the shelf, the next splits of the trees are depth strata, wind exposure, MPA protection and s A no-fish. Mix reef fish constituted 79% of the assemblages on the upper euphotic strata, it was also the dominant group (27%) in the lower euphotic and upper mesophotic strata on the leeward side outside the MPA. M. niger dominates (37%) on lower euphotic and upper mesophotic strata on the windward side with low no-fish acoustic biomass. On the pelagic side, shelf-break large fish dominated (29%) on the mid-mesophotic zone 60-80 m and small pelagic and predators (47%) on the lower mesophotic zone 80-100 m. When considering the reduced database containing sediment information (Fig. 7b,d), the first explicative variable was the sediment. On SaStCoRhAl sediment, M. niger dominates (73%). For sediments other than SaStCoRhAl, the next splits of the trees are sediment, no-fish acoustic biomass, wind www.nature.com/scientificreports/ exposure, MPA protection and depth strata. On the sediments LrAl, SaAl, SaRhAl, SaLrAl, SaStAl 52% of the assemblages belong to mix reef fish. Univariate analyses showed that the percentage of space occupation was substantially higher on the windward side for half of the assemblages (C. sufflamen, bottom weak fish detection, shelf-break large fish, shelf-break school, small pelagics and predators, M. niger) ( Supplementary Fig. S7 online). The highest acoustic biomass of all groups corresponded to small pelagics school in the leeward side, followed by M. niger in the windward side and mix reef fish in the leeward side.
All assemblages ( Supplementary Fig. S7 online) varied substantially according to the bottom depth strata. The acoustic biomass percent of presence were higher in the upper and lower euphotic strata for mix fish and M. niger, respectively. The acoustic biomass and percent of presence of demersal assemblages, sand fish and mix reef fish, decreased with depth. The opposite occurred for pelagic groups shelf-break large fish, shelf-break school, small pelagics and predator, C. sufflamen that peaked at mid and lower mesophotic depths, except for small pelagic school that presented highest acoustic biomass in the upper mesophotic zone (40-60 m). Bottom weak fish detection and bottom small fish school, individual demersal fish were distributed and presented higher acoustic biomass either on the lower euphotic or upper mesophotic.
Sediment identification was only possible on shallow water and represented a small portion of the data. In this dataset, some assemblages presented clear association with one sediment (Supplementary Fig. S7 online). In particular mix reef fish were strongly related with SaLrAl, M. niger with SaStCoRhAl and bottom small fish school with Sa. Mix reef fish and M. niger, although showing a higher occurrence on a particular substrate, were the two only groups that appeared on all sediment types.
The no-take MPA effect was clear on fish distribution as six assemblages (bottom weak fish detection, shelfbreak large fish, bottom small fish school, shelf-break school, small pelagics and predators, M. niger) were more present and had higher acoustic biomass inside the MPA (Fig. 8). C. sufflamen was virtually absent outside the no-take MPA. The five remaining assemblages (sand fish, loose school, individual demersal fish, mix reef fish and small pelagic school), were all more present and presented higher acoustic biomass outside than inside the no-take MPA.  Table 1  www.nature.com/scientificreports/ Black triggerfish biomass. The black triggerfish, the second group in terms of fish acoustic biomass, was mostly concentrated in the East side of FNA. In this area (Fig. 9), its actual biomass was estimated to 700 tonnes (19 g m −2 ) with an estimated CV of 40%. Its distribution inside the area was heterogeneous and organised in patches.   www.nature.com/scientificreports/

Discussion
By combining acoustic and video observations, we provided for the first time, a comprehensive vision of tropical fish distribution from the near-shore line to off the shelf-break with a description of (i) fish diversity, (ii) horizontal fish-like and no-fish distribution, and (iii) a focus on the black triggerfish. The data gathered also provided important evidence to (iv) revisit the Island Mass Effect (IME), and (v) give some insights for Marine Spatial Planning (MSP).

Fish diversity.
In about 50 h of video footage, we recorded 47 fish species of 29 families (Table 3). This is only a small fraction of the ichthyofauna of FNA that, with a total of 250 species and 77 families, harbours the greatest richness of marine fish among the oceanic islands of the South Atlantic 96,114 . Indeed our objective was not to perform an exhaustive description of the fish diversity but to capture the fish composing the bulk of the biomass. Still, compared with other visual census techniques we observed a similar number of families (27-28 96,115,116 ) but much fewer species (50)(51)(52)(53)(54)(55)(56)(57)(58)(59)(60)(61)(62)(63)(64)(65)(66) 96,115,116 . Only, Soto 95 depicted much more families (68) and species (167) but his inventory included pelagic species combining visual census, fisheries surveys and literature records. The most diverse families (Serranidae, Labridae and Pomacentridae) are underrepresented in our study since we focused on pelagic and demersal species that can be assessed with acoustics and did not put efforts on filming benthonic and cryptobenthic communities. In our study, two species, the tropical sardine Harengula sp. and the black triggerfish M. niger accounted for more than 90% of the fish recorded. This confirm the fact that the biodiversity in FNA is represented by few very abundant species 117 . M. niger was the second most abundant and the most often observed species. Such results differ from other studies in FNA. Indeed, most studies 96,115,118 report Thalassoma noronhanum, Haemulon chrysargyreum and Stegastes rocasensis as dominant species. The difference between our and other studies is likely due to our extensive depth coverage compared with other that mostly focused on shallow (< 20 m) waters. Only Schmid et al. 116 observed a dominance of M. niger but their study was performed using baited remote underwater video stations and black triggerfish are voracious bold species 119 . Even if far form exhaustive, our video records likely provide a robust picture of the main pelagic and demersal species present in FNA and a robust information to complement and identify acoustic observations. Fish echotypes are known to provide a heuristic description of fish species, assemblages or communities 120,121 . Combining video observations with the scrutinising of acoustic echograms allowed for the identification of consistent fish assemblages (Table 2). We acknowledge that, in some cases, some fish echotraces may not have been correctly assigned to the proper fish assemblage. However, since these assignations were sustained by ~ 50 h of video footage we are confident that potential mis-assignation should not have significantly impacted our results.
Comprehensive tropical fish distribution. The algorithm we applied on acoustic data allowed discriminating between fish-like and no-fish echograms. The access to simultaneous video observation and the care taken to validate the algorithm with these images makes us confident that the fish-like echograms do indeed correspond to fish.
The strong stability in acoustic fish biomass distribution between FAROFA surveys (Fig. 2), through years and seasons that allowed merging the data from the three surveys, suggests a bottom-up structuration of fish assemblage. Most fish observed by video and acoustics are demersal and pelagic. Demersal fish are classically associated with typical habitat in terms of sediment, benthic composition, structural complexity or depth 24,[122][123][124] (see also "Revisiting the Island mass effect (IME)"), which may explain site-fidelity. In the same way, except for Harengula sp., the pelagic fish species were concentrated at the shelf-break, a known hotspot for pelagic fish 75 . Seasonal variations (e.g. rainy vs. dry season) do not imply significant environmental changes. Indeed, the seasonal variation gradient sea surface temperature is minimal (varying from 26.5 to 28 °C) 34 and does not seem to significantly impact the distribution of the fish as observed by acoustic. A similar result with no change regarding the season was found in fish predator diet 125 .
The comprehensive spatial coverage we achieved allows providing the first map of fish acoustic biomass around FNA (Fig. 3). Such a picture cannot be completed with classic methods based on visual census or fishing operations (in systems where fishing activity is allowed). More generally, this is the first comprehensive highresolution map of fish distribution of a tropical system from the near-shore to the shelf-break. To our knowledge at least one example of map of tropical fish biomass was produced from acoustic data covering a fraction of the U.S. Virgin Islands 126 . However, this study did not use simultaneous video observation limiting the skills of species identification. By combining acoustics and video, we provide the spatial distribution of all acoustically detected fish (Fig. 3) but also of a variety of fish assemblages (Fig. 6). Our results reveal a strong heterogeneity in the distribution of fish acoustic biomass with the presence of hotspots. This reinforces the fact that punctual observations may miss hotspots in comparison with our extensive continuous sampling. The comprehensive maps we provide have several benefits since they can help defining areas for further sampling strategies, and are key elements for management in particular for implementing MSP 6,38,77 . Black triggerfish: a key tropical player. In phase with video observations, the black triggerfish was the second most important species in terms of acoustic biomass and the one occupying most space. In its main zone of distribution, we estimated the black triggerfish biomass to 700 tonnes (i.e. 19 g m −2 ). M. niger is one of the very few reef fish with a circumtropical distribution 127 . It is known to form large shoals of more than one hundred individuals and has been reported to exhibit high densities around remote oceanic islands such as Ascension Island 128 127 . This is probably due to its long pelagic phase that enables its settlement in remote location and its high plasticity www.nature.com/scientificreports/ in resource use. Indeed, M. niger broad omnivory gives him the potential to forage opportunistically on a variety of prey, including pelagic algae remains or dolphin vomit and faeces 97 . In addition, isolated oceanic Islands such as FNA, are impoverished and the functional richness is low 96,134,135 . M. niger has the ability to take advantage of an empty niche as demonstrated by Mendes et al. 119 at St Peter and St Paul Archipelago where it endorses the functional role of opportunistic grazing herbivore. FNA lacks of large herbivore that are represented by few small Scaridae of the gender Sparisoma 136 and roving herbivore represented by few Kyphosidae and Acanthuridae. In our study we observed one species of parrotfish Sparisoma amplum (3 individuals in 2 videos), one species of sea chub Kyphosus sectatrix (183 individuals in 4 videos) and one surgeon fish Acanthurus chirurgus (11 individuals in 1 videos). Those species are found with higher abundance in southern location of the Brazilian coast such as Abrolhos, Bahia for parrot fish and surgeon fish, or even higher latitude such as Arvoredo Island, Santa Catarina and Arraial do Cabo, Rio de Janeiro for sea chubs 134 . The black triggerfish is mostly distributed on the northeast side of the Island. This windward pattern of distribution is facilitated by its high swimming abilities 96 . The windward side of FNA also concentrates the more complex sediment, containing reef algal-vermetid barriers along rocky shorelines. M. niger directly benefits from this sediment complexity as it lays its eggs in the sand, feeds on epilithic algal matrix (EAM) and sleeps in rocky reef 137 . Adult, M. niger is strictly reef-associated as it sleeps in the same hole every night, which attests to a high level of site-fidelity 138 . If the bulk of the M. niger occurred in the selected area for biomass estimation, we also observed M. niger on the leeward side close to the shore where large rocks occur. In this case, M. niger were juvenile associated with other reef fish (e.g. surgeons fish, sergeants). Although abundant, easily catchable, edible and appetent, the black triggerfish is not a commercial species and is only consumed occasionally.
Revisiting the Island mass effect (IME). The IME describes well the higher primary and secondary productivity on the leeside due to turbulent mixing and advection created by eddies on the wake of islands [139][140][141] . However, the IME on tertiary productivity remains quite unknown or is described as isotrope around the island 27 . Our data allows us to better describe the IME and propose a new conceptual figure (Fig. 10) with an asymmetrical response of the fish distribution.
Horizontal fish distribution patterns were very different regarding the position of the archipelago face to the main wind/current flows (windward vs leeward). Overall, fish were much more abundant on the windward side. This is true for most assemblages, especially for pelagic fish and triggerfishes. According to the IME, an overall increase in island-related production is expected. This effect is expected to be more marked on the leeward side due to hydrodynamic retention processes 27,29,34,142,143 . The no-fish acoustic biomass, mostly composed of zooplankton (including gelatinous) was indeed higher on the leeward, in particular off the shelf where dense layers of gelatinous were observed. Lessa et al. 74 144,145 .
If the IME is well described in terms of turbulent processes and further primary productivity 30 , there is a lack of fine scale information on the consequences in terms of fish distribution. We reveal that most pelagic fish are concentrated windward facing the main flow, where the productivity is expected to be lower. This study does not allow concluding about the mechanisms explaining this feature, but we can propose at least three hypotheses. First, the current flow reaching the island topography likely concentrates the flux of particles 143 including zooplankton, favouring the feeding behaviour of medium size planktivorous pelagic fish (e.g. C. sufflamen) or intermediate predators that feed on small fish, shrimps or invertebrates (e.g. Caranx crysos) and are targeted by top predators (e.g. large Scombridae and Carangidae, Istiophoridae and sharks). The aggregation of planktivorous fish on the windward side of island has been observed by Hamner et al. 146 that described a "wall of mouth" composed by planktivorous fish picking up zooplankton before it reaches the reef 146,147 . Second, the water is much more turbid leeward than windward where visual predators concentrate. Third, a behavioural pattern consisting in facing the current (reotrope) to keep associated to the island and avoid advection may also play a role.
Interestingly, most demersal assemblages were also more abundant on the windward side. Structural habitat complexity is known to be an important factor for fish richness and abundance 24,148,149 and higher acoustic biomass was associated with the more complex sediment SaStCoRhAl, a mosaic of different substrates. This may be a consequence of turbulent processes. Indeed, on the windward side, the strong oceanic flow flush soft sediments, favouring the development of complex habitats that concentrate fish 148 . On the other hand, soft sediments deposit leeward, forming sandy habitats that are less populated 148 . Wind and current exposure influence the underwater landscape with the windward side, characterized by extensive reef barriers along rocky shorelines, and the leeward side, mainly composed of descending slopes along a rocky shoreline with large scattered boulders and small reefs scattered on sandy habitats 150 . This difference in sediment between the windward leeward due to the wind and wave exposure is a common feature observed elsewhere. e.g., in the Madeira Archipelago 139 . IME effect is thus not just an enhancement of primary productivity though physical process in the leeward side of islands but it also drives the fish distribution by shaping the habitats. Wind and current cause erosion and a transformation of the shoreline and sediment distribution which also impacts the distribution of fish. This results in an anisotropic distribution of fish with schools of small pelagic and associated predators flourishing at the shelf-break on the windward side taking advantage of the vertical mixing, the current and the clear water.
Depth strata. Fish acoustic biomass and assemblage composition varied according to the depth strata. The highest overall fish acoustic biomass was observed on the mid mesophotic zone ( www.nature.com/scientificreports/ the shelf-break. By providing a cross-shore along-depth acoustic biomass estimations of demersal and pelagic fish, we confirm the importance of the bathymetry 151 and quantify and rank the fish distribution among depth strata. Specifically, the shelf-break was a hot spot for five assemblages (Fig. 6). Indeed, at the shelf-break, we   www.nature.com/scientificreports/ observed the steady presence of shelf-break schools as well as small pelagics and predators. It is in this area, in particular at the windward, that SSF targeting both demersal and pelagic species concentrate 152 . With our results, it is difficult to determine whether the depth and associated light conditions (euphotic vs mesophotic) or the structural characteristic of the vertical zones are the main drivers of diversity and abundance. Indeed, in our case the mid mesophotic zone matches the shelf-break. This zone has specific characteristics. It presents a steep slope associated to high structural complexity 18 . In addition, by being at the interface of neritic and oceanic domains it concentrates organisms from both domains 153 . Mesophotic reefs associated to the shelf-break are thus hot spots for marine life. However, our result do not confirm that mesophotic reefs per se (independently of the shelfbreak) concentrate more life than euphotic ones. Despite their importance, shelf edge reefs are seldom included in marine protected area network, in particular in Brazil 23 . Shelf-breaks should thus be considered as important area for biodiversity conservation 24 .
Insights for marine spatial planning. MSP is a natural extension of practices including integrated coastal management and multi-use MPA management 36 . MPAs are regarded by many marine scientists as a major management tool needed to tackle fisheries collapse and regular loss of marine biodiversity [154][155][156] . Data and analysis should be central to decision-making. The data we provide here have the advantage of being comprehensive. We show that fish acoustic biomass was significantly higher inside than outside the no-take zone ( Supplementary Fig. S6 online) indicating a potential effect of the MPA. Such positive effect was witnessed by Ilarri et al. 115 on shallow-reef fish communities. Here, by providing a comprehensive covering of the entire FNA we provide much more robust estimates. If most assemblages had higher acoustic biomass inside than outside the no-take MPA, it is difficult to fully unravel between protection effects and habitat characteristics. Indeed the no-take zone encompasses the entire windward shelf that is characterised by higher bottom habitat complexity while the no-take MPA area encompasses the area close to the port that is susceptible to enhance the productivity through the eutrophication and the presence of rocks and structures. In all cases, with the objective to protect fish biomass and assemblages, the no-take zone seems overall well designed. It covers most of the shelf, protecting reef fish that exhibit high densities in near shore shallow waters areas that classically suffer great anthropic pressure linked to multiple use activities including artisanal and recreational fishing. However, the no-take zone stops at the 50 m isobaths and thus leaves the shelf-break unprotected. As an important zone for biodiversity, foraging and spawning, protecting the shelf-break could favour species reliant on shelf-break mesophotic reefs. However, since most SSF operate at the shelf-break it is important to let a sufficient portion of the shelf-break open to fisheries, in particular for those targeting pelagic species that temporally use FNA as a shelter during their long journey.

Conclusion
By combining multifrequency acoustic data and video, we provide the first comprehensive description of demersal and pelagic fish distribution of a tropical ecosystem. We also provide the first biomass estimation of the black triggerfish Melichthys niger, a key tropical player. More generally, we pictured the distribution of a variety of fish assemblages and related their spatial patterns to biotic and abiotic environmental features. Comparing the effects of euphotic and mesophotic reefs we show that more than the depth, the most important feature is the topography with the shelf-break as the most important hotspot. Beyond, this approach allowed us to revisit the IME. We completed the IME portrait and revealed that it is an asymmetric process regarding fish distribution. Indeed, while primary productivity is mostly enhanced in the leeward, higher trophic levels concentrate on the windward side. We also tested for the impact of the no-take MPA of Fernando de Noronha Archipelago on the distribution and acoustic biomass of demersal and pelagic fishes. This MPA protects the most complex habitats that shelter the highest fish diversity and biomass. Still, an important hot spot, the shelf-break, currently unprotected could be partly included given that sufficient space is left for fishing activities, in particular for pelagics. Maintaining pelagic small-scale fisheries in FNA is indeed socially and economically important. More generally, describing fish distribution and associated environmental features is the first step toward understanding how fish communities are spatially structured and is a necessary step to conduct MSP and operate relevant protection policies. www.nature.com/scientificreports/