The future floating wind farm studied in this paper will be implanted in 2025 in the Western Mediterranean Sea, about 18 kilometers off the coast of Gruissan (Aude, France) and about 30 kilometers from the edge of the continental shelf ( Figure 1 ). The depth in this area varies between 60 to 70 meters. The substrate in this area is essentially muddy (Hamdi et al., 2010). According to IFREMER SIH data about landings, the main commercial species in the study area are Engraulis encrasicolus, Octopus spp., Scomber scolias, Scomber scrombus, Diplodus spp.,.

The EolMed pilot project, developed by Qair SA, started in 2016, involves the installation of three wind turbines producing a total power of 30 MW. The three wind turbines should be built by the end of 2024. The top of the tower of the wind turbine is located at about 100 m above the surface. The blade is about 80 m long. The wind turbine has a rotor diameter of 160 meters. The originality of this project is based on its floating foundation. This structure has a square shape with a central damping pool, which dampens the movements of the swell. The dimensions will be 49 m long, 49 m wide and will reach an underwater depth of 13 m. The turbine will be anchored by a maximum of 8 anchor lines. After this pilot project, a commercial wind farm park, composed of 25 wind turbines ( Figure 1 ), is planned. The commercial wind farm (i.e. with 25 wind turbines and floating systems based on the design of the actual EolMed pilot project) was used as the focus of our this study. The spatial scope of this park, 100 km², will allow to create food web models of equivalent size to those found in the literature on wind turbines (Raoux et al., 2019, 2017; Wang et al., 2019).

Ecopath and Ecosim parameters and equations are presented in the Appendix A .

An initial Ecopath model was calibrated for the year 2019 corresponding to the year initial timing for the implantation of the wind farm, though due to several delays, this timeframe has now been shifted to 2025.

Species clustering in each of the functional groups included in the Ecopath model was based on their habitat, food preference, size and commercial importance. The final food-web model contained 27 functional groups (1 marine mammal, 1 seabird, 13 teleost fishes, 2 cartilaginous fishes, 5 invertebrates, 2 zooplankton, 1 phytoplankton, 1 alga and 1 detritus) ( Figure 2 ).

Input parameters were obtained from available data from published articles or from sampling campaigns conducted by IFREMER (the French Institute for the Exploitation of the Sea) and PELAGIS (French observatory of marine mammals and birds). Pre-processing was carried out in order to select the species presented in the vicinity of the implantation area, to transform the unit of measurement to the unit used in EwE (biomass/km²) and to group species into functional groups to obtain biomass for each functional groups. Final data are summarized in Table 1 (more detailed information are gathered in Appendix B ). Biomass values were obtained from the MEDITS campaign (Jadaud and Certain, 1994), the PELMED campaign (Bourdeix and Hattab, 1985), and the PELAGIS campaign (Laran et al., 2011). Production/biomass and consumption/biomass ratios were taken from the literature or obtained from empirical equations using asymptotic length, asymptotic weight and growth data (Nilsson and Nilsson, 1976; Banse and Mosher, 1980; Innes et al., 1987; Opitz, 1996; Pauly, 1980; Palomares and Pauly, 1998). Diet composition ( Table C1 ) was obtained by compiling published information primarily in the Mediterranean Sea. Assimilation rates were obtained from a published model (Coll et al., 2007; Bǎnaru et al., 2013). For almost every Ecopath model, many species have a part of their life cycle outside of the model area (Christensen et al., 2008). Two options are proposed to manage this issue by Christensen et al. (2008): 1) ‘diet import’ approach which consists of setting the diet import proportion to the proportion of time spent outside of the model area; 2) ‘model expansion’ approach which adds functional groups in order to represent the outside food web structure. The first option was chosen then a proportion of the diet composition ( Table C1 ) of these groups as imports to the ecosystem was added. Principal fishing activities present in the area were included in the Ecopath model: bottom trawling, pelagic trawling, seine, longline and other types of fishing. Fishing data were obtained from the IFREMER SIH. An index of confidence to the data (pedigree) was associated to each data (Christensen et al., 2008). Values of pedigree index are provided in Table C2 .

Balancing an Ecopath model requires compliance with certain constraints. Physiological and thermodynamic rates for each functional group must be respected (Heymans et al., 2016): (1) EE values must be inferior to 1, (2) P/Q for consumers must be between 0.05 and 0.3 and (3) R/B must be between 1 and 10 for fish groups (Christensen et al., 2008). All models were balanced using a step-by-step expert calibration procedure (Heymans et al., 2016). This means that in order to obtain a mass-balanced model, inputs (B, P/B, Q, EE and diets) were modified to satisfy the physiological and thermodynamic constraints. Procedures ( Method D1 ) and results ( Figures D2, D3 ) of the balancing step are provided in Appendix D .

A preliminary step that is required before an Ecosim model can be run to explore prospective scenarios is the calibration of the model through the estimation of a vulnerability matrix that allows for variations in biomasses to be reproduced, as observed in the ecosystem (Heymans et al., 2016). Thus, in order to fit the model to historical trends, biomass and capture sampling data from 2008 and 2019 were used from the databases previously mentioned in section 2.3. An Ecopath model, HIST ( Figure 3A ), was created for the first year of the historical data (2008). Biomass, P/B, Q/B and EE were calculated according to the data available in 2008. If data were missing, data from our BOWF model were used. Input parameters are presented in Table 2 . Data used to fit to biomass historical trends are presented in Appendix E . The strategy used to fit to time series thanks to the estimation of vulnerabilities was a “predator” search strategy meaning that a vulnerability value was set for each predator and not for each “predator-prey” relationship.

The fishing technique used in the MEDITS campaign does not allow for a reliable estimate of the biomass of some pelagic species. Then, admitted that biomass data for European Seabass and Medium Pelagic Fish were admitted as not reliable, then the fitting was conducted on the fishing data. No biomass data were available for Bivalves & gastropods, Crustaceans and Large Pelagic Fish; thus, fishing data were used for the fitting procedure. In order to use fishing data as reference, relative fishing effort data were used as input in the fitting procedure ( Table 3 ). The fishing effort time series were calculated with the following formula:

where P n is the power value (kWh) of the vessel for the n^th year, and Technological coefficient is a factor of evolution of the power of the vessel. These factors were retrieved from (Pauly and Palomares, 2010) and presented in the Table 4 . Initial power for 2008 were obtained from the EVOMED report (2011) (Sartor et al., 2011). The power for the year between 2009 and 2018 were obtained from the formula above using the n year power to calculate the n+1 year.

Two indicators were used to study the quality of the fit of the time series: the Akaike Information Criterion (AIC) and the Sum of Squares (SS) of deviations between model and data (Mackinson, 2013). The AIC is an indicator that penalizes fitting on too many parameters. The lower the AIC indicator is, the “better” the fitting. SS, sum of square difference, is an indicator which measures the accuracy of the fitting. A low SS indicated that the model fits well with respect to reference data.

The fitting had an AIC of 2.096 and an SS of 159.6. When comparing with other models in the Mediterranean Sea on the basis of fitting with only trophic interaction, the AIC and SS obtained seems to acceptable (Papantoniou et al., 2023; Piroddi et al., 2017; Tsagarakis et al., 2022; Vilas et al., 2021). The Monte Carlo routine was applied to assess the sensitivity of the biomass output to the input parameters (B, P/B, Q/B and EE). Input parameters were modified according to a normal distribution centered on the Ecopath input value and with a deviation parameter. In our case, pedigree index was applied to the coefficient of variation ( Table F2 ). 500 iterations were ran, which allowed for plotting in the 5^th and 95^th percentile for the fitted results. These results ( Figure F3 ) and the matrix of vulnerability ( Table F1 ) are available in the Appendix F .

Ecosystem indicators were used to compare scenarios by using the mean of the last four years based as in Agnetta et al. (2022); Ricci et al. (2023). The “Network Analysis” routine of the EwE suite was implemented in order to compute ENA indicators (Ulanowicz (1986). Indicators were separated into three themes: structure, maturity and/or resilience and diversity and are provided in Table 8 .

The pedigree index of the BOWF model was 0.525. Ecotrophic efficiencies (EE) ranged from zero to 0.979 ( Table 9 ). An EE of 0 was obtained for top-predator groups, Dolphins and Seabirds whereas the highest EE was obtained for Small Demersal which had a medium trophic level (3.15). Trophic level ranged from 1 to 4.69 ( Table 9 ). The groups with TL=1 were the primary producers and corresponded to Phytoplankton, Algae and Detritus. Dolphins occupied the highest TL. Other top-predator groups in the model had a trophic level superior to 4, such as Seabirds (TL=4.022), European hake (TL=4.049), Sharks (TL=4.132), Large Pelagic (TL=4.233) and Large Demersal (TL=4.279). Omnivory index ranged from 0 to 1.378. Null and low omnivory index (OI) were obtained for low trophic level groups such as Bivalves & gastropods, Benthos, European anchovy (OI=0), Zooplankton, European pilchard and Other Small Pelagic (OI=0.111) while a high omnivory index concerned high trophic level groups such as Large Pelagic (OI=1.071), Rays (OI=1.114), European Hake (OI=1.187), Sharks (OI=1.312) and Large Demersal (OI=1.378). Omnivory indices for all functional groups are gathered in Table H1 .

The living functional group with the highest biomass was Phytoplankton, it represented 20% of the total living biomass. The consumer group with the highest biomass was Benthos, it represented 19% of the total living biomass. For the fishes, Other Small Pelagic was the most abundant group followed by European pilchard and European anchovy, all of which are pelagic fishes.

Highest keystone index (K1, Libralato et al., 2006) was obtained for Crustaceans (K1=-0.0878). The second and third highest keystone indices were calculated for Benthos (K1=-0.22) and Zooplankton (K1=-0.237) ( Figure 4 ; Table H1 ).

Total System Throughput (TST, t.km^-2.year^-1) ( Table 8 ) increased in all four scenarios compared to BOWF. Highest TST was obtained for the OPTIM scenario, then a slightly lower TST was obtained for REEF. TST for TRANSFERT was lower than the OPTIM and REEF values of TST. Finally, TST decreased slightly between TRANSFERT and SCRAP ( Figure 5 ). Finn Cycling Index (FCI, %) ( Table 8 ) decreased for the 4 scenarios compared to the BOWF model. FCI values for the REEF and OPTIM scenarios were identical. FCI was identical for the SCRAP and the TRANSFERT scenario, which in turn was greater than the REEF and OPTIM scenarios ( Figure 5 ). Regarding Mean Trophic Level catch (MTLc) ( Table 8 ), the OPTIM scenario is the only scenario of the four which presents values above the BOWF model. Among the three other scenarios, it was the REEF scenario which had the higher MTLc., TLc was slighty higher for the TRANSFERT scenario compared to the SCRAP scenario ( Figure 4 ).

With the introduction of the REEF effect, the keystone index (KI) ( Table 8 ) changed ( Figure 4 ). Highest values of keystone index for the BOWF scenario were obtained for low trophic level organisms such as Crustaceans, Benthos, Zooplankton and Phytoplankton. For the REEF, SCRAP and TRANSFERT scenarios, the highest values of keystone index were obtained for a mix of fish groups such as Small Demersal and Medium Pelagic and low trophic level organisms such as Crustaceans, Benthos and Zooplankton. Highest keystone index values varied slightly for the OPTIM scenario compared to the other scenarios. Sea Bream became an important group while Small demersal, Medium Pelagic Crustaceans, Benthos and Zooplankton remained important structural groups ( Figure 4 ). The differences between BOWF scenario and the other scenarios may be linked with the structure of model (new groups added).

Ascendency (A, t.km^-2.year^-1) ( Table 8 ) increased for the whole scenario compared to the BOWF scenario, with the highest value of OPTIM only slightly higher than the value for REEF ( Figure 6 ). Regarding the scraping scenario, TRANSFERT scenario presented slightly higher values than the SCRAP scenario ( Figure 6 ). Ascendency for the SCRAP and TRANSFERT scenario were largely inferior to REEF and OPTIM scenarios. Redundancy (R, t.km^-2.year^-1) ( Table 8 ), proportion of flow to detritus (DET, %) ( Table 8 ) and entropy (H, t.km^-2.year^-1) ( Table 8 ) increased for the whole scenario compared to the BOWF scenario. Highest values for the three indicators were obtained for the REEF and OPTIM scenarios ( Figure 6 ). R, DET and H for SCRAP and TRANSFERT were largely inferior to the values for REEF and OPTIM scenarios. These indicators were slightly higher for the TRANSFERT scenario compared to the SCRAP scenario ( Figure 6 ).

Shannon diversity index (SI) ( Table 8 ) increased for the all scenarios compared to the BOWF scenario. SI values were much higher for the REEF and OPTIM scenarios compared to both SCRAP and TRANSFERT scenarios. SI increased slightly between the REEF and OPTIM scenarios while SI remained constant between the SCRAP and TRANSFERT scenario ( Figure 7 ). Kempton Q ( Table 7 ) increased slightly for the REEF and SCRAP scenarios compared to the BOWF scenario while it seemed constant for the TRANSFERT scenario and decreased largely for the OPTIM scenario. Kempton Q was much higher for the REEF scenario than the SCRAP scenario. Value of Kempton Q decreased slightly between the SCRAP and TRANSFERT scenarios. Kempton Q was largely inferior for the OPTIM scenario compared to all other scenarios ( Figure 7 ).

Regarding the biomass trends obtained at the end year of the simulation, several species seemed to benefit from the reef effect and the derived scenarios. Indeed, Small Demersal biomass was multiplied by between 1.03 and 1.189 for the REEF, 1.018 and 1.181 for the OPTIM, 1.055 and 1.117 for the SCRAP and 1.049 and 1.104 for the TRANSFERT scenarios. Sea Bream biomass also increased by a factor between 1.153 and 1.867 for the REEF, 1.177 and 1.904 for the OPTIM, 1.158 and 1.305 for the SCRAP, 1.079 and 1.262 for the TRANSFERT scenarios. Some groups seemed to benefit more from the partial closure of the fishery, with Large Demersal biomass multiplied by between 1.018 and 1.110 for the REEF, 1.073 and 1.171 for the OPTIM scenarios. Medium Pelagic biomass increased by a factor between 1.020 and 1.065 for the REEF, 1.252 and 1.294 for the OPTIM scenarios. Large Pelagic biomass was multiplied by between 1.034 and 1.076 for the REEF scenario while for the OPTIM scenario biomass was multiplied by between 1.33 and 1.384. In the TRANSFERT scenario, biomass for groups such as Flatfishes and Sea Bream increased significantly, after the hard substrate was scraped, by an approximate factor of 1.2 and 2.1, respectively, but this increase did not extend over time and biomasses returned to their initial values. Overall, biomass trends decreased for some groups such as Crustaceans, Jellyfish and Cuttlefish & Squids. For Cuttlefish & squids, biomass decreased by a factor between 0.981 and 0.997 for the REEF, 0.945 and 0.962 for the OPTIM scenarios, while they remained constant for the SCRAP and the TRANSFERT scenarios. Biomass declines appeared more severe in the OPTIM scenario ( Figure 8 ). Biomasses for the functional groups missing from Figure 8 are available in Figure I1 and values of biomass for each year are available in Table I2 .

Pedigree is the only indicator that addresses the quality of the input data that is included within an Ecopath model (Morissette, 2007). The Ecopath model BOWF had a high pedigree index, compared to the mean pedigree (0.44) value that was obtained by Morissette (2007) and revealed that input data in BOWF was of acceptable quality. However, the pedigree indices obtained in the BOWF model were lower than most of those obtained for the Ecopath model that was developed in the North-Western Mediterranean Sea (Coll et al., 2006; Bǎnaru et al., 2013; Corrales et al., 2015) which ranged from 0.62 to 0.67, and were slightly higher than the one obtained by Vilas et al. (2021) (0.50). In the present model, biomass data were mainly local and were obtained from detailed samplings. Like for the majority of EwE models, P/B and Q/B were obtained mainly from an empirical equation and trophic levels were lower than those included in other Ecopath models. Landings were local but not as precise as they could have been; even if we did collect landing values in six large areas in the Gulf of Lion. In parallel, diet remains the main source of incertitude in the present model. Even if most of the diet data came from sampling in the Mediterranean Sea, diets change and adapt to available resources both seasonally (Morte et al., 2002) and geographically (Rumolo et al., 2016). Catch and diet were certainly the most uncertain factors in our model and explain the differences in pedigree obtained here, compared to other models.

The initial Ecopath model, BOWF, was composed of 27 functional groups. According to the keystone index, the majority of the most ‘structuring’ group were within the low trophic level categories such as Crustaceans, Benthos, Zooplankton, characterized by a high keystone index and relative total impact. Bǎnaru et al. (2013) and Corrales et al. (2015) found that dolphins, seabirds and high trophic level fish were keystone species in the ecosystem. Both added that low trophic organisms played an important role in the ecosystem (Bǎnaru et al., 2013; Corrales et al., 2015). In the BOWF model we developed, diets of high trophic level organisms were essentially based on imports because of the limited size of our model compared to their feeding area; this may partly explain the differences in the key groups.

Artificial hard substrate introduction in the Mediterranean Sea are mainly colonized by mussels, which are pioneers in the colonization of this substrate (Fabi et al., 2002; Airoldi and Bulleri, 2011; Spagnolo et al., 2014). Similarly, we expect that the introduction of wind farm turbines will provide new habitat, for crustaceans and benthos organisms and an additional source of food that will be integrated in the trophic web. In our model, we thus assumed that this new source of food would lead to a change in a proportion of the diet for some demersal and pelagic groups. Results of the EwE model (REEF) showed that Sea Bream and Small Demersal functional groups could largely benefit from the introduction of the wind farm while Medium Pelagic and Large Pelagic fish functional groups only benefited slightly. European Hake as a separate functional group benefited from the reef effect, but in the fitting procedure, biomass data simulated by Ecosim were largely superior to the reference biomasses. As such, the increase in biomass of European Hake should be taken with caution. The Rays functional group benefited from the reef effect while the Rays functional group did not feed on the hard substrate. The high vulnerability obtained from the fitting procedure for Rays ( Appendix F ) means that the Rays as a functional group was far from its carrying capacity (Christensen et al., 2008). Rays biomass was probably driven by the “bottom-up” effect, and thus when Rays prey biomass increased, Rays biomass skyrocketed. Adding a reserve effect (OPTIM) from the wind farm led to a higher biomass for most fish groups including those which benefited from the reef effect and those with a high trophic level. As an example, Medium Pelagic and Large Pelagic functional groups benefited only slightly from the reef effect while they benefited largely from the reserve in the wind farm area. This result was expected, as one of the major effects of MPAs is the protection of high trophic level populations (Valls et al., 2012). Some low trophic groups were affected (e.g. Jellyfish) from the combined reef and reserve effect, mainly due to the increase of the biomass of their predators.

According to the scenario with the combined reef and reserve effect (OPTIM), the implantation of an offshore wind farm will have an impact on food web structure. Here, the quantity of flows in the ecosystem increased, characterized by the total system throughput. The quantity of flows with the introduction of the hard substrate was supposed to slightly increase in the case of monopile windfarm (Raoux et al., 2019; Wang et al., 2019; Nogues et al., 2021). Our approach suggests that the colonization of the floating hard substrate had an equivalent impact. Finn Cycling Index decreased with the reef effect and with the association with the reserve effect, suggesting that the structure of the food web changed due to the implantation of the offshore wind farm (Finn, 1976). Compared to the previous model developed for monopiles, the trend for the Finn Cycling Index in the present study was the opposite. Nogues et al. (2021) and Wang et al. (2019) obtained a slight increase in the FCI index. Energy recycling capabilities seemed to decrease with the implantation of floating wind farms while it increased with monopile wind farms (Raoux et al., 2017). A hypothesis to explain this pattern is that monopile wind farms provide benefits mainly to bottom fish groups that feed more on detritus compared to pelagic groups. The effects of the introduction of a wind farm on the mean trophic level of the catch were largely studied by Raoux et al. (2019). Like in their models, our results indicated that the increase in low trophic level biomass overwhelmed the increase in high trophic level biomass.

According to Ulanowicz (1986), Ascendency (A, t.km^-2.year^-1) is expected to increase as an ecosystem matures (Ortiz and Wolff, 2002; Ulanowicz, 1986; Raoux et al., 2019). Two other indicators were proposed in this study to characterize the maturity of the system: the proportion of flows to detritus (%) and entropy (t.km^-2.year^-1). When the system matures, a shift from herbivory to detritivory is expected (Christensen, 1995; Geers et al., 2016; Odum, 1969). For a mature ecosystem, species diversity is expected to be high but with EwE, several species are gathered in the functional group making measurement difficult. Christensen (1995) proposed to measure the flow diversity instead of species diversity by quantifying the statistical entropy (H, t.km^-2.year^-1) for all groups in the ecosystem. The results of the present work on these indicators pointed out that under a reef effect (REEF and OPTIM) the ecosystem seemed to be more mature than a model without a reef effect (BOWF). Even if a high value of Ascendency is correlated with a more mature system, it also means that since the system is more active in specific pathways, it may lose flexibility and thus lose resilience (Raoux et al., 2019). Redundancy (R, t.km^-2.year^-1) and Ascendency for both REEF and OPTIM increased compared to the scenario without a wind farm, meaning that the system did not lose its flexibility. Indeed, Redundancy measures the number of parallel trophic pathways that link the different functional groups (Ulanowicz, 1986). When there is an increase in Redundancy, this means that the flows are distributed in several alternative pathways to link one specific group to another (Heymans et al., 2007), and results in an increase in the resilience of the ecosystem (Heymans, 2003). Based on the results for Redundancy, it appears that the reef effect (REEF and OPTIM) led to a higher resilience compared to the model without the reef effect (BOWF). Raoux et al. (2019) found similar results with respect to the maturity and resilience after the introduction of the reef effect. Further, it appears that the introduction of both floating and monopile wind turbines led to a system that was more mature and flexible system in response to potential disturbances.

According to Shannon’s diversity index, diversity increased with the introduction of the reef effect. However, Kempton’s Q index showed that the diversity increased only for the reef effect and not for the combination of the reef and reserve effects. Different trends were obtained with the Shannon’s diversity index vs the Kempton’s Q index. Shannon’s diversity index studied the overall diversity of functional groups and equitability, and took into account all of the trophic levels in the system, while Kempton’s Q index only studied the diversity of the upper trophic levels (Kempton and Wedderburn, 1978). Overall diversity increased with the introduction of the reef effect (REEF and OPTIM) but the diversity for the upper trophic levels only increased with the reef effect and decreased with the reserve effect. Marine Protected Areas are expected to benefit all trophic levels (Soler et al., 2015). In the model with the reserve effect, there were dissimilarities between functional groups. Further, as several functional groups (Large Pelagic, Medium Pelagic, Rays) benefited largely from the reserve effect, an overall decrease in diversity might result for high trophic levels.

Functional groups which largely benefited from the reef effect, such as Sea Bream and Small demersal, experienced a reduction in their proliferation with the scraping scenario. When compared only to the reef effect, a reduction in biomass was common to the majority of groups. Adding the transfer of biomass to the seafloor benefited Flatfishes, Sea Bream, Octopuses and Small demersal functional groups for which biomass was largely increased. However, this increase was followed by an increase in fishing pressure, inherent to Ecopath, which nullified the benefit for these functional groups.

With regular scraping of the biofouling (SCRAP and TRANSFERT), the Total System Throughput (TST, t.km^-2.year^-1) decreased compared to the scenarios with reef effect only. However, the decrease in this indicator was predictable since biomass was extracted and thus the production from the extracted biomass was removed from the ecosystem. The trophic level of catch (TLc) is inevitably reduced because of the scraping on the hard substrate sessile organism which are low trophic level groups. The regular scraping of biofouling and its deposition to the sea bottom led to slightly higher value for TST and TLc than only the regular harvesting of the biofouling. The difference in TST is due to the periodic high increase in biomass ( Figure 8 ) of some groups such as Small Demersal, Large Demersal, Sea Bream, Flatfishes. This increase was not obtained in the SCRAP scenario and then it explained the difference. The same explanation can be use to explain the higher TLc in the TRANSFERT scenario compared to the SCRAP because with the periodic high increase in biomass of some medium and high trophic level, it balanced the impact of the harvest of the hard susbtrate group on the TLc indicator. The FCI values highlighted important structural differences (Finn, 1976) between the reef effect and the scenario in which reef organisms were scraped. Differences in structure between the scraping and sampling (SCRAP) and the scraping and settling (TRANSFERT) were negligible according to FCI.

In 1985, Odum proposed an evolution of the indicators that he developed previously (Odum, 1969) to determine trends in a stressed ecosystem. Odum, (1985) considers that a stressed ecosystem has a low efficiency in its conversion of energy to organic structure, which leads to a decrease in species diversity and a decrease in the redundancy of parallel processes. Proportion of flow to detritus, redundancy and entropy decreased from both scraping scenarios compared to the reef effect scenario. It appears that the regular scraping of the hard substrate led to a less mature system. This result is consistent with the assumption that under regular perturbations, an ecosystem is less mature (Nilsson and Grelsson, 1995). Also, the decrease in the Ascendency (t.km^-2.year^-1) from the reef scenario to the scraping scenario appears in line with these results according to the indicators studied in the present paper, where scraping and allowing the organisms settle on the sea bottom led to a more mature system than scraping and removing the organisms for the ecosystem.

Diversity of an ecosystem is expected to decrease with intense fishing effort (Bianchi et al., 2000). With the scraping of the hard substrate, the benthic organisms on the substrate were subjected to a very high fishing effort, for which we would expect the overall diversity to decrease. Kempton’s Q index and Shannon diversity index confirmed this assumption.

The model developed to study the potential effects of a floating offshore wind farm in the Gulf of Lion addressed the colonization of the hard substrate by groups that were present in minority in the food web (i.e buoys, shipwrecks). We proposed an original approach to simulate for the colonization of large floating devices. This approach relied mainly on the parameters, biomass accumulation, and vulnerability (see 2.7.1) and enabled us to follow the speed of the colonization observed on a floater in the Gulf of Lion. In order to better fit with reality, more field data for biofouling succession and biomass are needed. To better integrate this new food source in the trophic web, it will be essential to observe species present around the OWF and carry out stomach content analysis to ensure which functional groups feed on the biofouling organism. In the present study, network analysis indicators were used to quantify the effects of the wind farm on the ecosystem. These indicators permitted us to study the effects of the implantation of the wind farm in a global way (Raoux et al., 2019). However, the links between these indicators and the system maturity, or resilience, are complex (Raoux et al., 2019). The present study and other models for offshore windfarms (Raoux et al., 2019; Nogues et al., 2021, 2017) highlight a greater maturity and/or resilience of the ecosystem on the basis of the network analysis indicators, while Wang et al. (2019) who developed both a pre-implantation model and a post-implantation model with sampling for input parameters for both, found more mixed results for the maturity of the system.

In addition, the present model covered an area of 1,000km² in the Gulf of Lion. The use of a restricted area was required to study the impact of the wind farm turbines, and similar surface areas were used by Raoux et al. (2017), (2019), Wang et al. (2019). The main constraint of an EwE model on a small area was the population dynamics. One way of managing this limit is to introduce a proportion of imports into the diet of groups that were occasionally present in the area (Christensen et al., 2008), but this proportion was difficult to determine precisely and we opted for the percentage of the feeding area covered by the EwE model. Another way to manage this issue is to create an Ecospace model on a larger area, i.e Gulf of Lion, then simulate the migration of highly mobile groups.

The French Ministry of Ecological Transition announced the tendering of two 250 MW wind farms in the Gulf of Lion. In this context, the development of an Ecospace model representing the entire Gulf of Lion and simulating the effects of the implantation of these two wind farms is required. However, our model does not cover all of the potential impacts of wind turbines that can be simulated with EwE. An update EwE model should work on including the effects of noise (Debusschere et al., 2016; Serpetti et al., 2021), electric fields (Gill et al., 2014), pollution from float components (Leleyter et al., 2016), the bioengineering effect of the biofouling organism (Sadchatheeswaran et al., 2020) which can provide refuge for some species and reduce their vulnerability to predaction.