Scenarios of fish waste deposition at the sub-lagoon scale

Analysis framework

We developed an analysis framework to aid zoning and site selection for aquaculture based on predicting environmental impacts of fish farming in a broad range of explorative scenarios (i.e. situations or developments that are considered possible) (Börjeson et al., 2006). Its five steps (Figure 1) involved the use of three models (Figure 2):

Open multimedia modal

Open multimedia modal

1. The main farm types in Mayotte were defined based on surveys, and fluxes of solid waste (faeces and uneaten feed) from each farm type were estimated using the FINS farm-scale model. This step was performed in a previous study (Chary et al., 2019).

2. Zones where aquaculture could be developed within the study area were determined as a function of physical (depth) and logistical (distance from the coast) constraints of the farming systems.

3. Contrasting site profiles, including different hydrodynamic regimes (barotropic currents and depth) were selected within aquaculture zones using a clustering method. Simulations of hourly barotropic currents were obtained for each selected site from the three-dimensional hydrodynamic Regional Ocean Modeling System (ROMS) (Chevalier et al., 2017).

4. Dispersion of fish farming waste was simulated for all relevant farm and site combinations using NewDEPOMOD (Black et al., 2016).

5. The factors that most influenced predicted extents of deposition on the seabed (i.e. deposition footprints) were examined, and the results interpreted in light of thresholds of deposition rates (based on the literature) to develop recommendations for zoning and site selection.

Application to the Mayotte case study and farm types

Case study

Mayotte is part of the Comoro Islands, located in the northern end of the Mozambique Channel, 300 km northwest of Madagascar and 450 km east of Mozambique. Mayotte (374 km²) is composed of two main islands—Petite-Terre and Grande-Terre—enclosed in a 1500 km² lagoon created by a double-reef system. The second and external reef is discontinuous and forms three main complexes: the NE reef, the reefs south of Petite-Terre, and the South reef. A marine nature park covers all Mayotte’s EEZ, including the entire lagoon. The park pursues several objectives, including protection of the park and sustainable development of marine activities, including aquaculture.

We studied the northeastern part of the lagoon (NE Lagoon, 25 km long, 2–10 km wide, Figure 3), where the first and largest (production capacity of 400 t year⁻¹) fish farm in Mayotte was established in 1999 and where a plan for a large-scale farm (production capacity of 1500 t year⁻¹) is currently being investigated. The NE Lagoon has two inner passages connecting adjacent lagoons (Chevalier et al., 2017). Hydrodynamics within the NE Lagoon are driven mainly by the tide, with typical mesotidal variations ranging from 0.8–2.7 m and up to 4 m during extreme tides (Dinhut et al., 2008^; Chevalier et al., 2017). The lagoon has a relatively weak swell due to Mayotte’s sheltered location in the Mozambique Channel and to the presence of the double-reef system (De La Torre et al., 2008). The dominating SE–SSW winds (50% of all winds) are relatively weak, reaching 4.3 m s⁻¹, but tropical storms or cyclones can bring high winds from January to April (Jeanson et al., 2013). The lagoon’s water temperature is relatively stable and high (26–29°C), which can increase the growth potential of aquaculture species.

Open multimedia modal

Farm types

The fish farming sector in Mayotte produces mainly red drum, and the range of possible farming conditions in Mayotte was described in a previous study by three main farm types according to farm scale, production objectives, and technical and socio-economic criteria (Chary et al., 2019). Possible fish farming systems in the area were categorized from a survey of existing and planned farms. These categories were used to build three hypothetical farm scenarios: Small, Medium, and Large farms. The farm scenarios were distinguished by specific production systems (farm scale and associated technological level), production objectives, and rearing practices. The Small farm produced 23 t year⁻¹ and had simple aquaculture technology (rectangular floating plastic cages). Because the Small farm required sheltered areas and easily accessible sea-cage facilities (to facilitate farming operations and security), it was restricted to nearshore coastal areas. The Medium farm’s system was an upscaled version of that for the Small farm, with similar design and technology but more cages, which can produce 299 t year⁻¹. The use of safer and larger boats, however, and of surveillance cameras allowed the Medium farm to sit further from the coast. The Large farm’s system (2079 t year⁻¹) had more advanced technology and large service vessels to store feed and house employees. Thus, its cages were assumed to be permanently accessible for farming operations and security. See Table 1 for a summary of farm characteristics (i.e. dimension, number of cages, feed inputs, stocking density); more details can be found in Chary et al. (2019).

Open multimedia modal

Aquaculture scenarios

Explorative aquaculture scenarios were built to examine the potential benthic waste deposition footprint of the three farming systems in the study area. First, preliminary zoning based on farm siting constraints (site depth and distance from coast) identified zones where at least one of these three farming systems could be established. Then, sites with contrasting hydrodynamics were selected in each of these zones. Finally, relevant aquaculture scenarios were built by defining relevant combinations of each farming system and site.

Models: structure, inputs, and key forcing variables

For each aquaculture scenario, a finfish farm-scale model (FINS; Chary et al. 2019) was used to simulate waste fluxes, and a particle-dispersion model (NewDEPOMOD, version 20181109-125931-1541605374) was used to simulate waste dispersion and the deposition footprint. The structure of the models and their associated input data and key forcing variables are described below (Figure 2).

Waste dispersion simulated with NewDEPOMOD

The general structure of NewDEPOMOD has the same four modules as those initially developed for DEPOMOD (Cromey et al., 2002): (i) grid generation, (ii) particle tracking, (iii) resuspension, and (iv) benthic fauna response (benthic impacts). In this study, only the grid-generation and particle-tracking modules were used to obtain primary (initial) deposition footprints. Deposition and impact metrics were then calculated based on literature thresholds.

Grid-generation module

In the grid-generation module, domains covering an area of 4 km² with a horizontal resolution of 10 m × 10 m and a homogenous flat seabed were generated, and cage layouts were set in the centre of the grid to build the 54 scenarios. The total area and the grid-cell resolution defined (Table 2) were a compromise among decreasing particle export out of the domains, providing sufficient spatial detail for the smallest cages (6 m × 6 m) and reducing computing time. We made the simplifying assumption of a homogenous flat seabed throughout the domain to ease comparison of deposition footprints predicted under different currents and depths. Simulated particles were released from random starting locations within each cage.

Open multimedia modal

Particle-tracking module

Current intensity layer and time steps. In the particle-tracking module, we provided hourly horizontal current fields for three depth layers, as recommended by Hills et al. (2005): near-surface (3 m below the surface), near-bottom (3 m above the seabed), and mid-depth (midway between the surface and seabed). Tidal currents over a 15-day period (spring-to-neap tidal cycle) extracted from ROMS for the 36 grid cells, which corresponded to the 36 sites selected, were used. Currents were assumed to be the same throughout the domain, as required by NewDEPOMOD.

Feed pellet and faeces settling velocities. Feed pellet and faeces settling velocities are important species-specific parameters for modelling dispersion of aquaculture waste (Reid et al., 2009^; Bannister et al., 2016^; Broch et al., 2017). Since data on red drum waste-particle settling velocity were not available in the literature, an experiment was performed to predict waste dispersion accurately (see the Supplementary material for a full description). Briefly, we measured faeces-settling velocity rates of four commercial size categories (mean weight, small: 648 g, medium: 1152 g, large: 1913 g, very large: 3155 g) of cultured red drum and a range of red drum commercial feed (Nutrima^® diet, i.e. NUTRImarine 1.2, 2.2, 3.2, 4.5, 6.0, and 9.0 mm pellets) commonly used on farms in Mayotte. The method was based on a previous experiment (Magill et al., 2006^; Cromey et al., 2009^; Perez et al., 2014), with the design improved to obtain fresh undamaged faecal material by connecting the fish tank directly to a settling column (Figure 4). Particle settling was recorded using an HD camera, and the videos were analysed with particle-tracking software to determine individual settling velocities of many fecal (N = 2848) and feed (N = 186) particles. Settling data were analysed statistically to determine theoretical distributions of feed- and faeces-settling velocities. In NewDEPOMOD, the settling velocity of simulated particles remains constant during the growing cycle; only the statistical distribution of velocities could be set. When modelling the entire growing cycle is the goal, as in the present study, mean values can be used because all fish size categories are usually grown simultaneously on a farm. Consequently, faeces-settling velocities for the four fish categories were pooled, and means and standard deviations (SD) were used in the dispersion model. Feed-settling velocities also needed to be pooled. In NewDEPOMOD, we used the mean and SD of settling velocities of the three largest pellets (4.5, 6.0, and 9.0 mm), which represented more than 90% of the annual mass of farm feed input under routine farming operations (Chary et al., 2019).

Open multimedia modal

Resuspension module

Although the resuspension module in NewDEPOMOD has been updated since DEPOMOD and should simulate bed processes and resuspension mechanisms more accurately (Black et al., 2016), no studies of its application to or validation with other sites have been published yet. Using the resuspension module can change model outputs considerably (Chamberlain and Stucchi, 2007), and several studies that used DEPOMOD tended to turn off resuspension to optimize model performances (Chamberlain and Stucchi, 2007^; Cromey et al., 2012^; Keeley et al., 2013^; Chang et al., 2014). Consequently, we deactivated it as well. Deposition predictions were thus considered as “primary deposition footprints” (Keeley et al., 2013), i.e. deposition patterns and rates without any sediment reworking or mixing (e.g. resuspension, diagenesis).

Deposition metrics and impact threshold

Particulate waste deposition (i.e. kg solids m⁻²year⁻¹ in 100 m² grid cells) was simulated for the 54 scenarios with NewDEPOMOD. Deposition predictions were exported to MATLAB version R2015b (The Mathworks Inc., 2015) to calculate metrics that characterized the deposition footprint: (i) the area under farm influence (AUI_F, m²) at a given deposition rate (F) or range of rates (see below) and (ii) the maximum distance from cages at which F can be detected (DC_F, m).

Most studies defined environmental thresholds associated with fish farming in temperate environments and observed ecological effects across a broad range of deposition rates spanning two orders of magnitude (i.e. 0.1–10 kg solids m⁻²year⁻¹) [reviewed in Keeley et al. (2013)]. Several studies defined the threshold of detectable impacts in the range of 0.01–2 kg solids m⁻²year⁻¹ (Hargrave, 1994^; Cromey et al., 2002^, 2012^; Chamberlain and Stucchi, 2007^; Keeley et al., 2013). Since the relationship between solid fluxes and the biotic index available in the benthic module of NewDEPOMOD was developed and validated for temperate environments, it was not suitable to predict benthic impact in our study. Nonetheless, to illustrate how relationships between hydrodynamic regimes and deposition metrics developed in this study could be used to predict impact and help select sites, we used deposition rate thresholds for fish farms in tropical environments published in the literature, although information was scarce. Indeed, to our knowledge, only two studies have reported impact thresholds in tropical environments. Riera et al. (2017) observed significant decreases in sediment ecological status at deposition rates greater than 12 kg solids m⁻²year⁻¹ for fish farms in Macaronesia, but they did not examine rates greater than 16 kg solids m⁻²year⁻¹. In the Philippines, predicted deposition rates of 0.4–5.5, 5.5–27, and >27 kg solids m⁻²year⁻¹ were categorized as moderate, high, and severe impacts, respectively (Aquapark, 2015^; White and Lopez, 2017). Based on these values, three categories of impact were set as a function of F (kg solids m⁻²year⁻¹) in the present study (Figure 5):

Open multimedia modal

• F [0.5; 12]: detectable impact (D). We set the lower threshold to 0.5 instead of 0 because effects below it would be subtle and difficult to distinguish from the background of natural variability in nutrient enrichment, potentially resulting in less useful conclusions. For this range of deposition rate, AUI_D and DC_D thus represented the seabed area with detectable impacts and the farthest distance at which a detectable impact was predicted in the domain, respectively. To ease comparison with other studies, this range is equivalent to 0.2–4.1 g C m⁻² d⁻¹, assuming 45.5% C content in feed, 31.6% C content in faeces, and a 20:80 ratio of feed and faeces to total solids (Chary et al., 2019).

• F [12; 30]: moderate (M) impact, equivalent to 4.1–10.3 g C m⁻² d⁻¹

• F [30; max]: severe (S) impact, equivalent to more than 10.3 g C m⁻² d⁻¹

We thus calculated the AUI and DC for detectable (AUI_D and DC_D), moderate (AUI_M and DC_M), and severe (AUI_S and DC_S) impacts (Figure 5). The total AUI (AUI_F>0.5 = AUI_D + AUI_M + AUI_S) and peak deposition in the grid of each simulation (F_max) were also calculated.

Regression models were used to establish and test relationships between site hydrodynamic regimes [mean barotropic current intensity (i.e. mean of I_LT and I_HT) and depth as dependent variables] and the deposition metrics (F_max, AUI_F>0.5 and DC_D as explanatory variables). These equations were used to identify critical current intensities beyond which detectable, moderate, and severe benthic impacts were not predicted to occur, including uncertainty. Two ranges of uncertainty in each critical current intensity threshold were calculated as (i) the regression models’ 95% confidence intervals and (ii) the estimate that NewDEPOMOD predicts F_max with an accuracy of ±40%, based on a literature review of DEPOMOD’s accuracy in previous studies (Relationships between site hydrodynamics and deposition section). The higher critical current intensity (i.e. the conservative value) of the two ranges was then selected.

The maximum deposition footprint [i.e. largest AUI (AUI_max) and farthest DC (DC_max) for each deposition rate] was compiled to summarize the worst possible impact for each farm type. A regular deposition rate vector (F_n) was built that ranged from the minimum detectable flow (0.5 kg m⁻²year⁻¹, as mentioned) to the highest peak deposition (i.e. the maximum of the 54 values of F_max). Corresponding vectors of the two metrics AUI_n and DC_n were provided for the 54 scenarios, and the largest values of AUI_n and DC_n for each type of farm were selected to build the upper boundary of the envelope curve.

Suitable areas for fish farming in the NE Lagoon

In the coastal zone, 13.31 km² were suitable for Small farms, while 1.38 km² were suitable for Medium and Large farms. The off-coast zone covered 45.94 km² that were suitable for Medium and Large farms. The deep zone covered 23.63 km² that were suitable for Large farms. The total area suitable for each type of farm in the NE Lagoon was obtained from these values.

Contrasting site hydrodynamics in the NE Lagoon

The 36 sites selected in the NE Lagoon had contrasting hydrodynamic regimes (Figure 6 and Supplementary Table S1). To visualize this variability, we represented a tidal cycle averaged over the 15-day period from spring tide to neap tide of hourly barotropic current intensity using rose distributions of current for several of the sites. Most barotropic currents were oriented parallel to the coastline: northwest-southeast at most sites and north-south at sites between Grande-Terre and Petite-Terre (e.g. sites 10 and 11) (Figure 7). Rose distributions with the most elliptical shapes (e.g. sites 4, 6, and 13) were associated mostly with low currents (<5 cm s⁻¹) and representative of non-established barotropic current (Figure 7). In contrast, sites with higher intensity currents had more established direction and thus the least elliptical shapes (Figure 7).

Open multimedia modal

Open multimedia modal

Feed- and faeces-settling results

Settling velocities

Faeces-settling velocities of all red drum size categories had a mean (± 1 SD) of 0.64 ± 0.39 cm s⁻¹ (Table 3) and differed significantly between all four fish size categories tested (Kruskal–Wallis test, p < 0.001). Faecal particles from small fish settled significantly faster than those of other size categories, but other pairwise comparisons showed no clear distinctions between size categories (Table 3). Feed-settling velocity ranged from 3.67 to 15.68 cm s⁻¹ and generally increased with pellet diameter (Table 4). Differences in settling velocities between all pellet categories were large and significant (Mann–Whitney U test, p < 0.001 for all pairwise comparisons).

Open multimedia modal

Open multimedia modal

Predicted waste deposition footprint

Deposition footprint results

Deposition footprints tended to have an elliptical shape and extend in the direction of local tide currents. Among the 54 scenarios, the model predicted the highest deposition rates directly beneath cages, which decreased as distance from the cages increased (Table 5; details in Supplementary Table S2). A scenario with a Large farm (Oco-L₂₄, off-coast zone, Large farm, site 24) had the farthest DC_D, with a detectable impact up to 850 m. In 38 of 54 scenarios, particles released from cages were exported from the domain (i.e. >1000 m); thus, waste can be deposited even farther from the cages. Up to 21% of the particle mass released was exported from the domain; however, since deposition rates on domain edges were always below the detectable threshold, the scenario simulations effectively captured the area impacted by farm waste.

Open multimedia modal

The type of farm strongly influenced the predicted deposition rate and area of seabed impacted. Farm types differed in production scale, but also in the number of cages, cage type, and cage layout (Figure 8). The highest deposition rate (44 kg m⁻²year⁻¹) was generated by a Medium farm (Co-M₁, coastal zone, site 1; Figure 8) and Medium farms always had higher F_max than Large farms at a given site. For AUI, Large farms had the highest AUI_F>0.5 (0.1–0.5 km²), covering up to 58 times the total cage area. Impacts of Small farms remained detectable, with a maximum F_max of 10.7 kg m⁻²year⁻¹ at site 2 (Co-S₂) and an AUI_D that ranged from 1500 to 4700 m². Impacts of Medium farms were detectable (33% of the scenarios), moderate (54%) or severe (13%). AUI_M and AUI_S ranged from 200 to 2600 m² and 400 to 1500 m², respectively. No severe impact was predicted for Large farms, and moderate impact was predicted for 19% of the scenarios. AUI_M ranged from 3200 to 29 500 m². In the deep zone, impacts of Large farms remained detectable, with a maximum F_max of 11.9 kg m⁻²year⁻¹ at site 28 (Dl-L₂₈) and an AUI_D that ranged from 3200 to 29 500 m². When examining the largest AUI and farthest DC as a function of deposition rate among farms (Figure 9), the AUI was larger for Large farms than for Medium farms for F up to ca. 26 kg m⁻²year⁻¹; above this threshold, the AUI for Large farms decreased sharply.

Open multimedia modal

Open multimedia modal

Relationships between site hydrodynamics and deposition

Results of simple linear regression between current or depth and three deposition metrics (F_max, AUI_F>0.5, and DC_D) are presented in the Supplementary Table S3. Mean barotropic current intensity was strongly and negatively correlated with F_max and strongly and positively correlated with DC_D, at least for Medium and Large farms. These relationships indicated greater dilution of waste as current intensity increased. For instance, in Medium farm scenarios, F_max was 82% lower at the site with the highest current intensity (site 24, 19.2 cm s⁻¹) than at the site with the lowest current intensity (site 1, 0.6 cm s⁻¹). In contrast, correlations between depth and F_max, AUI_F>0.5, or DC_D were weak (R² < 0.2) and never significant (p > 0.1) for Small, Medium, and Large farms, indicating little influence of depth on deposition metrics. The influence of depth was notable mainly under low current intensities. For instance, in scenarios Co-S₁ and Co-S₂, in which current had a similar low intensity (<1 cm s⁻¹)_, Small farms had F_max of 5.1 and 10.7 kg solids m⁻²year⁻¹ at a depth of 33 and 9 m, respectively.

Regressions between current and the deposition metrics yielded the following equations for mean barotropic current intensity (cm s⁻¹) as a function of F_max (kg m⁻² y⁻¹): ln (y) = −0.290x + 2.466 (R² = 0.578, p < 0.05), −0.090x + 3.095 (R² = 0.906, p < 0.01), and −0.124x + 2.987 (R² = 0.805, p < 0.01) for Small, Medium, and Large farms, respectively (Figure 10). Critical current intensities beyond which detectable, moderate, and severe benthic impacts were not predicted to occur varied among farm types (Table 6). According to these current intensity thresholds and considering model (regression and NewDEPOMOD) uncertainties, moderate impact can be prevented in Medium and Large farms if mean current intensity exceeds 10.2 cm s⁻¹ (9.9% of the area suitable for Medium farms in the NE Lagoon) and 6.8 cm s⁻¹ (18.3% of the area suitable for Large farms), respectively. In contrast, to prevent detectable impact, mean currents of ca. 25–30 cm s⁻¹ would be required for all three farm types, but no sites in the NE Lagoon have this current intensity. These critical limits of current intensity could be adapted for other deposition thresholds using the relationships between current intensity and maximum deposition rate provided in this study.

Open multimedia modal

Open multimedia modal

Modelling limits and advantages

The present study used three different models to compare waste deposition footprints of different scenarios and to identify those that would minimize benthic impacts of aquaculture at the sub-lagoon scale. A hydrodynamic model was used to study the hydrodynamics in the study zone and extract current and depth data for 36 contrasting sites, a farm-scale model was used to estimate waste emissions from three farm types, and a waste dispersion model was used to simulate deposition footprints in 54 aquaculture scenarios. Limits and advantages related to the parametrization of each model, as well as their consequences on model outputs and potential application of the study’s framework elsewhere are discussed below.

First, the hydrodynamic model (ROMS) used in this study considered the tide as the only forcing variable. Although tide is the main forcing variable in this system (De La Torre et al., 2008^; Chevalier et al., 2015), it can vary spatially and temporally (seasonally or inter-annually) due to wind, swell, and intra-lagoon waves. These forcing variables may modify current intensity and/or direction locally at some sites. For instance, wind likely has a greater influence at shallow sites, due to greater impact of friction on the seabed, whereas swell increases water fluxes along the reef, pushing water out of reef openings and passages (Chevalier et al., 2015). Studying current variations at a finer scale, by including wind, swell, or wave forcing and their effects on waste deposition, could be a future course of research. Applying this study’s framework to another region would require examining the specific characteristics of that region’s hydrodynamics and probably the effects of these forcing variables.

In this study, NewDEPOMOD was applied for the first time to red drum farming, which required adapting only species-specific parameters (Cromey et al., 2009), such as particle-settling velocities, feed digestibility, and feed and faeces composition. In this study, we used the FINS farm-scale model, which use red-drum specific digestibility data and red drum feed composition data, to provide realistic and detailed time series predictions of waste fluxes under routine farming conditions. The advantage of using a farm-scale model such as FINS to do so, rather than using NewDEPOMOD alone, is that waste emissions can vary at the cage scale as a function of daily stocked biomass and feeding practices instead of remaining constant over time. Providing detailed rearing input data to DEPOMOD has been shown to increase its accuracy (Cromey et al., 2012^; Chang et al., 2014). We thus believe that combining FINS and NewDEPOMOD may increase the accuracy of NewDEPOMOD predictions. In this study, new faeces and feed-particle settling data for the commercial feeds used by farms (and parametrized in FINS) were acquired. They may also help increase the accuracy of the predicted deposition rates and decrease parameter uncertainties in NewDEPOMOD.

We examined effects of differing hydrodynamics on deposition footprints but did not consider other important highly site-specific processes, such as particle resuspension and effects of wild fish, which can also influence results greatly. Resuspension can pick up previously settled particles and redistribute them over larger areas, thus decreasing organic-matter accumulation and its ecological effects near cages (Keeley et al., 2013^; Broch et al., 2017). A seabed’s sensitivity to resuspension depends greatly on substrate type and texture (Law et al., 2016^; Carvajalino-Fernández et al., 2020a), as well as vertical structure, disturbance history and, for sediment substrates, resident biota (Sanford, 2008). Dispersion models with spatially explicit sediment resuspension parameters have been found to be more accurate than those with constant default resuspension parameters or those that ignore resuspension (Carvajalino-Fernández et al., 2020b). Substrate characteristics can vary greatly at the sub-lagoon scale and were not examined in this study; therefore, we did not intend to set specific parameters for each site. In previous studies, wild fish have shown a potential to reduce the amount of organic matter deposited on the seabed by 14% in some farms (Ballester-Moltó et al., 2017) by scavenging uneaten feed and, to a lesser extent, fish faeces (Fernandez-Jover et al., 2008). The magnitude of particulate removal by wild fish can vary, however, according to the biomass and species composition of their assemblages (Dempster et al., 2005), which also depends on site distance from the coast and farm scale (Dempster et al., 2002). Thus, because the wild fish effect also likely varies at the sub-lagoon scale, it was not simulated. The absence of resuspension and wild fish effects probably resulted in overprediction of deposition rates, particularly near the cages, where most uneaten feed settles. Model outputs should thus be considered as primary footprints and conservative worst-case deposition rates.

We used NewDEPOMOD in a new environmental context without validating it, since this study analysed exploratory scenarios with hypothetical farms. Model validation is an integral part of the modelling process that improves understanding of a model’s error, overall accuracy, and spatial applicability. The accuracy of DEPOMOD and some of its derived models (CODMOD, MACAROMOD) has been assessed for a variety of hydrodynamic conditions, including both depositional and highly dispersive sites. Differences between observations and DEPOMOD predictions of waste deposition for salmon farms in Scotland with low-to-medium mean current intensity (3.6–6.9 cm s⁻¹, depending on the layer) ranged from 13 to 20%, depending on the site (Cromey et al., 2002). In a seabass and seabream farm at a site with low mean current intensity (1.3–2.2 cm s⁻¹) in the eastern Mediterranean, MERAMOD (Cromey et al., 2012) had a mean accuracy of 44% (with optimized parameters), with more accurate predictions in high-flux zones (±35%) than in low-flux zones (± 111%). More recently, MACAROMOD was validated for four farms in the northeastern Atlantic that produced seabass, seabream, and meagre. Its accuracy ranged from 12 to 34% [calculated from Figure 5 in Riera et al. (2017)] at three sites with medium-to-high mean current intensity (8–25 cm s⁻¹) but was lower (±94%) at the fourth site (6 cm s⁻¹). Other studies of salmon farms in Canada (Chamberlain and Stucchi, 2007) and New Zealand (Keeley et al., 2013) reported similar patterns between observed and predicted footprints for optimized models, but model accuracy was not estimated or reported. This demonstrated that DEPOMOD’s transport equations are generally transferable to different hydrodynamic environments, including the conditions tested in this study, as long as good input data are provided and models are optimized. Most studies optimized model accuracy by deactivating DEPOMOD’s resuspension module (Chamberlain and Stucchi, 2007^; Cromey et al., 2012^; Keeley et al., 2013^; Chang et al., 2014), except for one (Cromey et al., 2009), which opens a debate on the quality of the models and parameters used to represent resuspension. The main changes made to NewDEPOMOD compared to DEPOMOD concern the modelling of resuspension and seabed processes (Black et al., 2016^; SAMS, 2020). Since resuspension was not simulated in this study, we expect NewDEPOMOD’s accuracy in our scenarios to be similar to that of the original model. In future research, it would be necessary to examine NewDEPOMOD’s accuracy for real and operating farms in Mayotte Lagoon.

Variability in impact across hydrodynamic regimes and farm characteristics

Systematic exploration of multiple aquaculture scenarios had the main advantage of allowing quantitative relationships between site hydrodynamics and deposition footprint metrics to be estimated. The physical properties of depth and current intensity are often presented as equally desirable attributes promoting dispersion (Hargrave, 2002^; Belle and Nash, 2008). In this study, depth did not influence waste dispersion significantly, while a significant decrease in solid accumulation near the cages and a broader deposition footprint were observed as current intensity increased. The lack of a significant influence of depth may be explained in part by the narrower range of depths in the scenarios (factor of 6 between minimum and maximum values) than of current intensities (factor of 30). A previous large study that evaluated environmental impacts of finfish and shellfish aquaculture across Europe (Borja et al., 2009) found that mean current intensity had a much larger influence than depth in reducing biotic index values. In our study, the significant relationships between current intensity and three deposition metrics (F_max, AUI_F>0.5, and DC_D) allowed the use of current intensity to predict waste-particle deposition. The relationship between mean barotropic current intensity and maximum deposition rate (F_max) decreased logarithmically, which indicated that a small increase in current intensity can provide a large dilution effect.

The amount of fish waste produced and released into the environment is related to the total fish biomass in stock; however, our results showed no correlation between maximum deposition rates and farm scale, and a strong influence of the farm layout. When simulated under identical hydrodynamic regimes, scenarios with Medium farms always had higher maximum deposition rates F_max (over small benthic areas) than Large farms. Since the Large farm releases much more waste per day than the Medium farm, this difference is probably due to cage density and layout (Figure 8). Indeed, Small and Medium farms have interdependent cages to reduce installation costs and the number of mooring lines (Cardia and Lovatelli, 2015). This interdependence reduced cage spacing (ca. 3 m vs. 20 m on Large farms), thus concentrating waste emission on a smaller area, which logically increased maximum waste deposition on the seabed locally. Aquaculture management guidelines generally recommend a “minimum farm distance” to avoid cumulative effects of multiple farms (Cardia et al., 2017). Based on this principle, “minimum cage spacing” could be added to reduce the risk of high waste accumulation on the seabed. These results demonstrate that farm scale is not the main factor that influences deposition rate; instead, management of the entire farm (e.g. cage position, cage spacing) must be considered to predict impacts accurately, as also highlighted in a recent paper (Burić et al., 2020).

Interpreting the deposition footprint

The impact metrics used may influence the interpretation of impact. The need to manage waste in intensive fish farming has led to a debate that opposes the approaches of concentrating waste over a small area vs. diluting and dispersing it to avoid excessive accumulation near the cages (Keeley et al., 2019). Depending on the approach, different metrics can be used to characterize the waste dispersion footprint and the degree of impact. Some may consider maximum deposition rate (regardless of the area impacted or distance of impact) as the most important metric, while others may consider the maximum distance (and thus area) of deposition (regardless the intensity of the flux) as the most important. Ideally, both metrics should be considered, since they can influence interpretation of potential impact on the seabed. For maximum deposition rate, Medium farms would have the highest impacts, since they always had the highest deposition rates, while for maximum distance, Large farms would have larger impacts than Medium farms, since their total AUI was always larger (4.4–18.4 times). However, reasoning in terms of “maximum” has no sense if the flux considered is not likely to cause any impact. This is why it is crucial to identify relationships between deposition rates and ecological responses as well as ecological thresholds beyond which significant change in ecosystem state and quality can occur (Groffman et al., 2006). These relationships are highly specific to a given ecosystem (Keeley et al., 2013) and have been developed and implemented mainly in temperate environments, while tropical environments have received little attention. For instance, a biotic index such as the multivariate AZTI Marine Biotic Index (M-AMBI), widely used to characterize the ecological status of benthos in temperate environments, has not yet been adapted or validated for tropical ecosystems or species. This is also true for tools that assess carrying capacity in general (Aubert et al., 2020). Therefore, we suggest below a possible interpretation and application of this study based on literature thresholds.

Based on ecological and regulatory thresholds from the literature, most predicted impacts of red drum fish farming in Mayotte’s NE Lagoon were located near farms. In ca. 70% of the scenarios, detectable impacts were predicted up to 500 m from the farm. In the most dispersive sites, they were predicted up to 850 m from the farm (Figure 9b and Supplementary Table S2), and traces of waste did disperse farther. In the literature, traces of fish waste have also been recorded 500–1000 m from a farm (Broch et al., 2017^; White et al., 2017^; Woodcock et al., 2018^; Keeley et al., 2019) and even farther (Bannister et al., 2016). Thus, a safe distance of ca. 1000 m between sites seems necessary to avoid cumulative effects of multiple farms and to protect zones of ecological interest from fish farm waste. In the range of hydrodynamic regimes tested, moderate and severe impacts were found up to 20 and 10 m from the cages, respectively. This result is in line with results of a meta-analysis of 70 aquaculture studies from around the world that estimated that the distance at which environmental impact is extremely unlikely (the “farm ecotone”) is 90 m (Froehlich et al., 2017). In the AZE framework, specific EQS (and thus impacts) are generally allowed up to 25–50 m around cages in EU countries (Katavić et al., 2005^; GFCM, 2012) and up to 200 m in other regions (e.g. United States, Australia, or New Zealand) (Wang et al., 2020). Based on these regulatory distances, moderate and severe impacts, if they occur, would remain in the AZE for all three farm types. If acceptable impact distances (AZE limits) and/or deposition rate thresholds used to define moderate or severe impacts change, these conclusions could change. For this reason, we plotted the maximum distance and maximum area impacted as a function of deposition rate (Figure 9). One can use these plots to adapt the conclusions to other ecological or regulatory thresholds. For instance, for a maximum EQS of 20 kg solids m⁻²year⁻¹ allowed within 50 m of the cages, some of the Medium and Large farm scenarios would have unacceptable impacts. Compared to a series of maps, this type of plot provides an easy-to-understand summary of possible benthic impacts of multiple farming scenarios in a given zone.

The areas in the NE Lagoon where hydrodynamic regimes could reduce benthic impacts below detectable, moderate, and severe impacts were examined for the three farming scenarios. The model’s results suggest that a low mean current intensity (3.2 cm s⁻¹) could prevent severe impacts of Medium farms. Based on this hypothesis, most of the area (83.5%) suitable for this farm type in the NE Lagoon would have sufficiently high current intensity to prevent such impact. Since severe impacts were predicted only for Medium farms, they should be unlikely to occur in the NE Lagoon. Based on the current limits, we can also hypothesize that moderate impacts on Medium and Large farms can be avoided only in limited areas: 9.9 and 18.3% of the area suitable for Medium and Large farms, respectively. Expecting zero impact is unrealistic, in any case, since very intense currents (25–30 cm s⁻¹) are necessary to prevent detectable impacts from the three farm types, and the maximum mean current intensity in the NE Lagoon was ca. 20 cm s⁻¹. Furthermore, such a high current intensity may reduce fish growth potential (Plaut, 2001), increase feed conversion ratio (Ferreira et al., 2012), and deform nets enough to endanger fish (GESAMP, 2001^; Cardia et al., 2017). Similar ranges of current intensity were proposed for salmon culture in Scotland as a function of biomass stock: 0–5 cm s⁻¹ for 0–499 t of biomass, 5–10 cm s⁻¹ for 499–999 t and >10 cm s⁻¹ for >1000 t (Cardia et al., 2017). Another classification of site sensitivity to fish farming based on current intensity was proposed in Norway: <3, 4–6, 7–10, and 11–25 cm s⁻¹ for very sensitive, moderately sensitive, slightly sensitive, and not sensitive, respectively (Velvin, 1999). This type of environmental recommendation should be considered at the same time as other biophysical requirements for farmed species and systems to ensure technically feasible and environmentally sustainable development of aquaculture.

Conclusion and perspectives

The present study proposed a generic five-step analysis framework to facilitate zoning and site selection based on predicting the deposition footprint in a broad range of explorative scenarios and applied to a case study of red drum farming in Mayotte. Legislation for fish farm licensing in countries such as France, and thus Mayotte, requires that environmental impact assessment be performed as a function of the amount of farm production expected. Our results indicate that this approach may be overly simplistic and that current intensity and cage layout are other important factors to consider. The combined use of hydrodynamic models, farm-scale models, and particle-dispersion models can consider many farm and site-specific factors that can influence benthic impact. Using such tools in the licensing process of new farms in France, as done in Canada and Scotland (Hills et al., 2005), would address potential impacts of aquaculture in a meaningful way. In the meantime, by using them for exploratory analysis at a large scale, we provided useful information to facilitate aquaculture zoning, site selection, and management. Worst-case impacts (i.e. maximum distance and area for a given deposition rate) were provided for multiple farm types. Practical recommendations for cage and farm spacing to reduce the concentration of farm waste on the seabed were also provided. Finally, critical current intensities were provided to provide information about the dispersive potential of sites in the NE Lagoon of Mayotte. These critical current intensities and the hydrodynamic-based site clustering method used in this study could be used to build spatial layers of environmentally suitable zones. Incorporating them in spatial analysis (e.g. GIS) and overlaying them with other physical, chemical, biological, and social attributes would provide a comprehensive multi-criteria assessment and create maps that could identify potential sites and help achieve successful marine plans more easily (Stelzenmüller et al., 2017). In the future, other near-field and far-field effects of fish farming in Mayotte Lagoon should also be examined. Given the abundance of coral reefs in the lagoon and their potential sensitivity to water nutrient enrichment (D’Angelo and Wiedenmann, 2014), we recommend studying in priority the effects of dissolved nutrient emissions from fish farming on coral reef ecosystems.

Acknowledgements

The authors gratefully acknowledge Cristele Chevalier for providing access to predictions of the ROMS model. We thank J.C. Falguière and S. Devillers for helping to perform the settling experiment. We thank Dr M.S. Corson for careful revision of the English. We thank R. Weeks and T. Adams for their support in using NewDEPOMOD. Finally, we thank the two anonymous reviewers for their valuable suggestions and comments.