2.1. Ecopath

The model was built in Ecopath with Ecosim (EwE) version 6.7. Ecopath is a mass balance model across functional groups (FGs) that represents a snapshot of the ecosystem in a given year. FGs are represented by species or groups of species that share similar predator-prey interactions, temporal biomass trends, ecosystem functioning, and habitat preferences.

FGs were selected based on the composition of the Black Sea commercial fishery [20] and ecosystem dynamics, as observed by Akoglu [16]. In addition, to assess and compare ecosystem structures across MSFD regions, the model was constructed following the same rationale described by Piroddi et al. [24], aiming to represent the diversity of the overall ecosystem and support best policy decisions.

Thus, in contrast to previously built Black Sea models [16, 23] top-predator mammal species (bottlenose dolphin, common dolphin and harbour porpoise), seabird species (seagulls & cormorants, terns and pelicans), benthic invertebrates (‘mobile’ and ‘sessile’ benthos groups) as well as seagrass and seaweed primary producers were included as FGs.

Input data (biomass, production/biomass, consumption/biomass, diet, and catch/discards) were extracted and/or calculated from in situ literature, stock assessments, or, when lacking, inferred from the Mediterranean model [24].

Catches (1995–2021) from the General Fisheries Commission for the Mediterranean (GFCM) (https://www.fao.org/gfcm/data/capture-production/ar/) were extracted to obtain a list of species that mainly represented the commercial species in the ecosystem. The species were subsequently organized within the defined functional groups (FGs, Table 1). Catches data were separated into landings and discards, by functional groups across the fishing fleets (Beach-Seiner, Trawler, Dredge, Gillnet and Trammel net, Longline, Traps and pods, Purse Seine and Midwater Trawl) following the catch and discard proportions used by Piroddi et al. [24] for the Eastern Mediterranean region, and when available, specific local fishery data by fleets were included (details in the S1 Table in S1 Appendix).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Model input parameters of the Black Sea balanced model for 1995.

https://doi.org/10.1371/journal.pone.0312170.t001

The biomasses of Atlantic bonito and bluefish were extracted from the literature and scaled down to consider the migratory behaviours of these stocks as they move into the Marmara and the Mediterranean Seas for wintering [4]. This approach was also applied to the Atlantic mackerel group. To account for changes in biomass, either temporally, outside the period considered, (i.e., species overfished before the model selected year) or spatially, outside the model domains (i.e., migration), the model allows to include a biomass accumulation term, negative or positive to represent the “Non-Steady-State Condition” of those species [31]. We applied a positive biomass accumulations for those species with migratory behaviours (Atlantic bonito, Bluefish, Mackerels), whilst we set a negative biomass accumulations for rapa whelk (Rapana venosa, Valenciennes, 1846) and picked dogfish (Squalus acanthias, Linnaeus, 1758) to better represent the temporal decline observed by their stock assessments (S1 Table in S1 Appendix).

The biomasses for all the other fished groups in 1995 was inferred using catch per unit of effort data (catch from the GFCM and reconstructed effort (for details, see section below)). Catches per Unit of Effort (CPUE) measured only the component of the population that is vulnerable to the gear (gear selectivity, size and age of fish, horizontal and vertical distribution of fish); to account for the total population, we corrected it applying a coefficient factor estimated by the model. For benthic invertebrates, data from literature were used (S1 Table in S1 Appendix).

Ecopath is based on the assumption that the production of one FG is equal to the sum of all predation, non-predatory losses, exports, biomass accumulations, and catches, as expressed by the following equation: (1) where B_i is the initial biomass (t*km^-2), (P/B)_i is the production rate per unit biomass of group i (year^-1), (Q/B)_i is the consumption rate per unit biomass of group i (year^-1), DC_ji is the relative proportion of prey i in the diet of predator j, NM_i is the net migration of prey i, BA_i is the biomass accumulation of prey i, Y_i is the yield (landings + discards, t*km^-2year^-1) of prey i, and EE_i is the ecotrophic efficiency of group i, that is, the proportion of production used in the system.

In an Ecopath model, the energy input and output of all the FGs must be balanced under certain ecological and thermodynamic rules. To balance the Black Sea model, we applied a manual mass-balance procedure following Mackinson and Daskalov [48], which modifies the appropriate input parameters employing six criteria organised by trophic levels (values of ecotrophic efficiency, production/consumption rate, respiration/biomass rate, slope of the biomass, slope of the production/biomass rate and slope of the consumption/biomass rate) (S2 Fig in S1 Appendix). We balanced the model following the best practice guidelines [36, 49, 50]: mass balance was obtained primarily by adjusting the diet matrix and the coefficient factor used to estimate biomasses from the CPUE which are the data source with higher uncertainty. Here we presented selected factors, among all parameters, defining the ecological and thermodynamic rules for balancing Ecopath models.

2.2. Ecosim

Ecosim uses Ecopath’s outputs as starting information to simulate the dynamics of each FG over time. In Ecosim, a series of differential equations describe the changes in biomass for each FG (i) over time, as follows: (6) Where dB_i/dt is the biomass change of group i in the time step dt, (P/Q)_i is its production/consumption ratio, i.e., gross growth efficiency, Q_ji is the consumption of group j (predator) on prey group(s) i, Q_ij is the consumption for predation by all predators j on group i (prey), I_i is the immigration rate, B_i is the biomass at time t, M_i and F_i are the natural and fishing mortality rates of group i, respectively, and e_i is the density-dependent emigration rate. Fishing mortality or fishing effort is used to drive the model.

Ecosim predicts biomasses and catches and compares them to observed time-series data of biomasses and catches using a log-likelihood sum of squares for calibration [58]. For each predator, consumption (Q_ij) is calculated based on the “foraging arena” theory, which divides the biomass of the prey into a vulnerable and a non-vulnerable fraction and the transfer rate or vulnerability between the two fractions determines the trophic flow between the predator and the prey [59]. The vulnerability concept incorporates density-dependent processes and expresses how far a group is from its carrying capacity. Default values of vulnerability (v = 2) represent a mixed trophic flow, a low value (v < 2) indicates a ‘bottom-up’, donor-driven, flow control and a situation closer to carrying capacity, while a high value (v > 2) indicates a ‘top-down’, predator-driven, flow control and a situation further away from carrying capacity. For each predator-prey interaction, the consumption rates are calculated as: (7) where a_ij is the rate of effective search for prey i by predator j; T_i represents prey’s relative feeding time; T_j is the predator’s relative feeding time; M_ij is the mediation forcing effects due to the presence of a third indirectly interacting species with the prey-predator pair in the food web; v_ij is the vulnerability parameter, which is akin to the half-saturation constant in Holling-type functional responses [60]; D_j represents the effects of handling time as a limit to consumption rate; and f(Envfunction,t) is the environmental response function that restricts the size of the foraging arena (or consumption) to account for external environmental drivers changing over time [34, 61]. Species-specific response functions to environmental factors were defined by minimum, 10^th and 90^th percentiles and maximum values of their optimal niches extracted from AquaMaps (https://www.aquamaps.org/) within the Black Sea domain (S3 Table in S1 Appendix). For functional groups defined by multiple species, the parameters were calculated as weighted means by species biomass composition.

For the Black Sea model, we used relative temporal changes in fishing efforts by fleet and net primary productivity for small and large phytoplankton, derived from the Blue2MF hydrodynamic biogeochemical model (GETM-BSEM [47]) to drive the model and predict temporal biomass and catch changes (1995–2021).

"Fishing effort” (kW*days⁻¹) was calculated as the product of the number of fishing vessels, Kilowatt (kW) per vessel, and the number of days spent fishing. For EU countries, we used data from the official European Commission databases (STECF https://stecf.ec.europa.eu/index_en), while for other countries, we considered the available literature [2, 62–66]. To account for improvements in technology (e.g., mobile phones, GPS, sonar, radio) that were not captured by kW as a measure of effort, a conservative technological “creep factor” of approximately 0.8% per year was used, as observed by Damalas et al. [67] and the European Court of Auditors [68].

Stock assessment data were used to validate biomasses and catches for turbot, whiting, red mullet, picked dogfish, horse mackerel, sprat, anchovy, and rapa whelk, while the International Commission for the Conservation of Atlantic Tunas (ICCAT) database (https://www.iccat.int/en/accesingdb.HTML) was used for large pelagic fish. The model was fitted [69] versus time series of observations of relative biomasses for stock-assessed species and relative catches for all fished functional groups (reported by the GFCM database). The fitting procedure is a routine that minimizes the sum of squares (SS) ratio between predicted and observed data as a metric for assessing the model performance. The SS per functional group can be used to evaluate and compare the quality of the group’s fitting to its respective reference time series data (for equal weight and number of observation points across the time-series). The fishing effort, primary production (PP) of large and small phytoplankton groups over time were used as the main forcing time series to drive the model.

The fitting procedure followed the approach described by Mackinson [58]. This method uses the Akaike Information criterion (AIC) [70, 71]: (8) where n is the number of observations, minSS is the minimum sum of squares resulting from the comparison of predicted with observed datasets, and k is the number of parameters to test statistical hypotheses related to changes in predator-prey dynamics, changes in productivity, impact of fishing, and possible combinations of the above-mentioned factors. The AIC is a statistical measure for model selection and penalizes for fitting too many parameters. AIC is used to choose the “best” model (the one yielding the lowest AIC) considering a good fit and the least number of estimated parameters. In this study, we used the second-order Akaike Information Criterion (AIC_c) calculated as follows: (9) to account for small sample sizes (n = observations) in the dataset.

Thirty-nine independent time series were used during the calibration process using different weights assigned to the time series to reflect the uncertainties of the data. We identified the best-fitted model using “by-predator” setting; additional vulnerabilities were found using “by-predators-preys” setting with a hybrid fitting approach [72, 73]. A thousand Monte Carlo runs were performed with the “best” fitted model to assess the uncertainties (5^th and 95^th percentiles) of the predictions using 10% variability in the input data (B, P/B, Q/B, EE).

Statistical significance of the changes in the temporal trends of simulated biomasses were analysed with time series and trend analyses. First, time series of simulated biomasses for the functional groups with observed data were decomposed into their seasonal, trend and irregular components using moving averages. The time series were then detrended by subtracting the time series of seasonal component from the time series of biomasses using the “stats” package in R [74]. Finally, nonlinear trend analysis was done using the “zyp” package in R [75] and the data were pre-whitened prior to trend analysis to eliminate autocorrelation as model simulated values are known to be hampered from autocorrelation following the Zhang’s method [76].

2.3. Ecospace

Ecospace is the spatiotemporal dynamic module of EwE and, as for Ecopath and Ecosim modules, it is coupled with the hydrodynamic biogeochemical model GETM-BSEM outputs, namely depth, annual bottom oxygen concentration, temperature, salinity and monthly primary/secondary productivity (small and large phytoplankton and zooplankton) which drive the spatial distributions.

Since Ecospace does not explicitly account for dynamics in the water column [77], we extracted data on salinity, temperature, primary productivity and planktonic biomass for distinct depth ranges: surface (0–10 m), euphotic zone (0–50 m), entire water column (0–2000 m), and (4) bottom (maximum depth). All species are distributed in space according to their dispersal rates (S2 Table in S1 Appendix) in a two-dimensional spatial grid (~3.7 km x 3.7 km) in relation to their depth-optimal niches (S3 Table in S1 Appendix) from the coastline and to a maximum depth of 2212 m (Fig 1A). The bottom oxygen concentration was used to drive the benthic species distributions to avoid their distributions from anoxic bottom conditions (Fig 1B) (< 10 mmol O₂/m²). Salinity integrated over 50m depth was only applied to the bivalve group, as values below 10‰ have a negative influence on the Mediterranean mussel, Mytilus galloprovincialis [78], the dominant species within this functional group. For the model hindcast, temperature and salinity (Fig 1C and 1D) were not used, with the above exception. S3 Table in S1 Appendix shows the response functions to environmental variables; in bold, the functions applied in the hindcast.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Black Sea environmental and LTL biological outputs used as inputs and drivers in the HTL model.

(A) Depth (m); (B) Bottom oxygen (averaged over 1995–2021); (C) Temperature averaged over 50m (averaged over 1995–2021); (D) Salinity averaged over 50m (averaged over 1995–2021); (E) Small phytoplankton integrated over 50m (averaged over 1995–2021); (F) Large phytoplankton integrated over 50m (averaged over 1995–2021); (G) Small zooplankton integrated over 50m (averaged over 1995–2021); (H) Large zooplankton integrated over 50m (averaged over 1995–2021); (I) Primary production integrated over 50m.

https://doi.org/10.1371/journal.pone.0312170.g001

Spatiotemporal changes in planktonic groups (small and large phytoplankton and zooplankton) were used in Ecospace as absolute biomass (Fig 1E–1H), whereas the Net Primary Production of 1995 was used in the Ecospace base maps (Fig 1I).

Ecospace simulations were run with 5 years spin-up, a burn-in period with no forcing or disturbances that allows species to be distributed in space in relation to their environmental niches before applying the temporal changes in planktonic biomasses and fishing effort [77].

Within the Ecospace module, the Habitat Foraging Capacity (HFC) allows to assess the suitability areas of the modelled species/functional groups based on the cumulative effects of the applied environmental conditions [79]. The HFC maps were extracted and qualitatively validated against AquaMaps suitability maps, downloaded and standardised for the Black Sea domain [80] (https://www.aquamaps.org/).

3.1. Ecopath

We modelled the entire Black Sea excluding the Azov Sea, covering approximately 423,000 km² with a maximum depth of 2,212 meters.

The final Blue2MF Black Sea HTL model, representing a snapshot of the ecosystem in 1995, has 47 functional groups (FGs), consisting of three dolphin species, three seabird groups, 24 fish groups, eight invertebrates groups, five planktonic groups, seagrass and seaweed, discards and detritus, and eight fishing fleet types (Fig 2, flow diagram). The model was balanced following the PREBAL ecological and thermodynamic rules (S2 Fig in S1 Appendix). The full list of species/FGs and the basic estimates of the balanced model are given in Table 1, while details on input data calculations and references per FGs are presented in S1 Table in S1 Appendix. Other inputs (landings, discards, effort and diet in S2 Appendix).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Energy flow and biomass diagram of the Black Sea ecosystem (1995).

Functional groups (grey) and fleets (red) are represented by nodes. The relative sizes of the nodes for functional groups are proportional to their biomasses in the ecosystem, whereas the relative sizes of the nodes for fishing fleets are proportional to their catch volumes (fishing fleets: Beach-Seiner, Trawler, Dredge, Gillnet and Trammel net, Longline, Traps and pods, Purse Seine, Midwater Trawl). The functional groups are ordered by their trophic levels (y-axis) and their preferred domain (bottom x-axis Pelagic vs Demersal vs Benthic). Predator–prey relationships are expressed with a gradient yellow (predator) to blue (prey) line. The graph was created by the authors using MATLAB vR2014b.

https://doi.org/10.1371/journal.pone.0312170.g002

3.1.1. Ecosystem indicators.

Functional groups are ranked by their TLs ranging from 1 (primary producers) to 4.29 (marine mammals); the highest TLs were found for ‘Common dolphin’ followed by ‘Bottlenose dolphin’, and ‘harbour porpoise’ (TL ≥ 4) (Table 1). ‘Seagulls & cormorants’ and ‘terns’, despite being considered top predators, showed relatively low TLs (3.38 and 3.36 respectively) due to the presence of discards and zooplankton in their diet. In EwE, discards are assigned to a detritus group with TL = 1 and seabirds feed considerably on discards, lowering the TL of those groups [81]. Several groups belonging to the pelagic and demersal domains had a TL greater than 3, while most of the functional groups with a TL smaller than 2 belonged to the benthic domain (Table 1 and Fig 2).

Other ecosystem indicators are listed in Table 2. Trophic level of the community (TL>1) was 2.5, which was slightly higher than that in the Mediterranean Sea in 2000 but lower than that in the Black Sea ecosystem in 1960. The TL of the catches in 1995, which indicates the exploitation status of fish populations [54], was much lower (2.97) than that in the Black Sea in the 1960s (3.52) and the 1980s (3.08), highlighting the continuous degradation of this ecosystem [16, 20].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary statistics for the Black Sea and Mediterranean Sea.

https://doi.org/10.1371/journal.pone.0312170.t002

The relative proportions of system throughput (TST) were in agreement with previous findings [16, 20], and were composed of consumption flows (38%) and respiration flows (23%), followed by flow to detritus (22%) and exports (17%). Comparing the absolute values of the total system throughput is not straightforward because this index is sensitive to the model structure [82]. Previous Black Sea models (Table 2) reported a TST of ~10,210 t km⁻² yr⁻¹ for 1995 and ~2007 t km⁻² yr⁻¹ for 1960; however, these models showed simpler structures than the model developed in this study, with only 10 and 22 functional groups, respectively. The TST value calculated in this study (Table 2) concurred with the value for the Mediterranean Sea model [81] that had the same model structure.

The ratio of total primary productivity to total respiration (TPP/TR), which is expected to be close to 1 in developed ecosystems, is an indication of ecosystem maturity [51]. Although the Black Sea was in a more mature state than the Mediterranean Sea (Table 2), our results showed that the Black Sea was losing this balance as TPP/TR increased since the 1960s. The transfer efficiency (TE) has also been decreasing over time in the Black Sea ecosystem, indicating inefficient functioning of the food web in terms of energy transfers to higher trophic levels. This could be a reason for the lack of recovery of higher-trophic-level predatory fish species in the ecosystem in addition to their overexploitation by fisheries.

The system omnivory index (SOI) indicates the range of feeding interactions in the food web; therefore, this index is also sensitive to the model structure, which is confirmed by our results showing a similar SOI with the Mediterranean Sea model [81].

Results of system ascendency and overhead, which represent the organizational degree in an ecosystem and its resilience against stress [55, 83], indicated higher values for both indices for the Black Sea than for the Mediterranean Sea (Table 2).

Heymans et al. [82] highlighted that a direct comparison of absolute indicator values generated by EwE across different ecosystems should take ecosystem traits such as type, size, depth, and location as well as model’s topology into consideration. In this study, we compared these indicators with a Mediterranean model built with the same model structure [24] and with other Black Sea EwE models built with different structures for different periods. The overall analysis of these statistical indices revealed that TST and SOI, as well as indices based on the trophic level greater than 1 (either TL of the community and catches) and transfer efficiency, were comparable with similar model structures, while TPP/TR, ascendency, and overhead were more ecosystem-dependent. However, our comparisons indicated a more deteriorated ecosystem condition in the Black Sea with respect to the results provided by earlier modelling studies.

MTI analysis (Fig 3) revealed that the majority of predators exerted a direct negative impact on their prey due to diet preferences. Functional groups negatively impacted themselves due to cannibalism or inter-species competition. Planktonic groups had predominantly positive effects on all pelagic and demersal groups in the system (e.g., through a bottom-up effect). However, these planktonic groups had negative impacts on benthic groups, as they were preyed upon by the aforementioned groups (e.g., through a top-down effect).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Mixed trophic impact relationships between functional groups and fisheries.

Positive values (from white to blue) indicate positive impacts, and negative values (from white to red) indicate negative impacts. “Green check mark” symbol indicates MTI value greater than 0.2, “yellow exclamation mark” symbol indicates MTI value between 0.2 and -0.2, “red cross mark” symbol indicates MTI value smaller than -0.2.

https://doi.org/10.1371/journal.pone.0312170.g003

The results also identified strong direct and indirect negative impacts of gillnets & trammels on dolphins (directly as by-caught by this fleet, and indirectly for targeting the same “prey” species) and on turbot and demersal sharks. An indirect positive effect was found for prey species of the above groups. Trawlers showed a positive indirect impact on seagulls that feed on discards, and a negative direct impact on the main trawling target demersal species. Longlines had a negative impact on seagulls & cormorants as by-caught by this fleets (Appendix 2 -Discards), and an indirect positive impact on terns due to their competition with seagulls. Purse-seiners had direct negative impacts on bluefish and anchovy as well as on common dolphins by direct by-catches, but also indirectly, as this fleet targets dolphin’s prey. It is interesting to note that, fishing fleets are substantially acting as top-predators exhibiting negative effects on their target species as “preys” and competing with other top-predators such as mammals and seabirds. Declining fish stocks determined an increase of interactions between top-predators and fishers, in particular, interactions between gillnets/trammels and dolphins are well documented in the Black Sea as well as in the Mediterranean Sea [84–86] having negative consequences for both cetaceans and fishers [87, 88]. As for the interaction between seabirds and fisheries, despite the lack of scientific evidence in the Black Sea region [89], several studies have documented this phenomenon worldwide. For example, negative interactions between longline fisheries and seabirds is well documented in the Atlantic and Pacific Oceans [90, 91] whilst the positive effect of trawler’s discards on gulls is also well reported in the North Sea [92, 93].

The species with the highest keystone index value (Fig 4) in the system was large zooplankton (40); however, by definition, keystone species are the species with a relatively low biomass, but their biomass changes would have a disproportionately large effect on the ecosystem structure [94]. According to this definition, seagulls& cormorants (4), followed by sprat (15), mugilidae (21), mobile benthos (37), horse mackerel (12) and bottlenose dolphin (1) were the functional groups with the highest keystones index values. In the Black Sea model, representing the early 1960s, demersal sharks and dolphins were the keystone species [16]. This contrasting finding indicated that a change of control on the food web occurred since the 1960s, from a top-down to a bottom-up control.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Relative total impact (ε_i) versus keystoneness (KS_i).

The size of the circles is proportional to the species/group biomass.

https://doi.org/10.1371/journal.pone.0312170.g004

3.2. Ecosim

The fitting routine improved the model performance by 24% compared to the baseline fishing scenario (Table 3). Fig 5 and S3 Fig in S1 Appendix displayed the best fitted model predictions (lines) plotted against observations (points) when available for the 1995–2021 period, showing the fitting for stock assessed species (Fig 5) horse mackerel, anchovy, sprat, red mullet, whiting, turbot, picked dogfish and Rapana venosa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Modelled biomass predictions of the best-fitted model.

Predictions (lines) versus observed (points) estimated by statistical stock assessments for the 1995–2021 period.

https://doi.org/10.1371/journal.pone.0312170.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Comparison across selected stepwise fitting interactions.

https://doi.org/10.1371/journal.pone.0312170.t003

The fitted model showed an overall decreasing trends: all the commercially exploited species except red mullet and turbot declined since 1995 (Fig 5). Good agreements of the model fits with observed fluctuations were found for anchovy adults (SS = 0.89), sprat (SS = 1.08) and Rapana venosa (SS = 0.74), prediction was also good for Atlantic bonito (SS = 1.18) whilst a higher sum of squares was found for horse mackerel (SS = 4.13) due to the more scattered pattern of the observations. The model missed on representing the initial trends of the time series for whiting (SS = 3.65) and the steep observed decline of picked dogfish (SS = 6.04) whilst for turbot (SS = 2.10) the prediction did not replicate the end of the observed time-series. Red mullet (SS = 3.50) did not show clear trends neither for the model prediction and the observations.

According to the trend analysis, all the simulated biomasses except juvenile anchovy had significant decreasing trends (Fig 5 and S4 Table in S1 Appendix). The overall declines could be attributed to poorly-managed fisheries at national level and the lack of a basin-wide management policy. The Black Sea fish stocks are in fact shared across countries due to the high level of connectivity and the migratory behaviours of the primarily exploited species [95]. Stock assessment for picked dogfish (observations in Fig 5) showed a very steep declining trend since the 90’s, due to over-exploitation and the use inappropriate fishing gear [96]: the observed and modelled trend agrees with the classification of “vulnerable” (high risk of extinction in the wild) given by the IUCN Red List Threatened Species for Squalus acanthias in the Black Sea region [97].

Red mullet was the only stocked assessed species that did not show a clear predicted declining trend, with highly scattered observed calibration data (Fig 5). Similarities between congeneric species Mullus barbatus and Mullus surmuletus (red mullet and striped mullet respectively) often lead to ‘voluntary’ inaccuracies in species reporting by fishermen [98, 99]. Red mullet are often reported as striped mullet that have a smaller minimum sampling size to avoid penalties and to land more catch. This practice raises concerns about the accuracy of catch statistics for these two species.

The model was unable to reproduce the recent increase in turbot biomass (2017–2021) predicted by the stock assessment. This increase could be captured by forcing the observed turbot catches for those years (S4 Fig in S1 Appendix). This model uses the observed total catches per functional group for all countries fishing in the Black Sea (Ukraine, Romania, Bulgaria, Russia, and Turkey). Overfishing has caused a strong decline in turbot since 1980; for this reason, since 2016, a fishing management plan has been put in place by the European Commission for Romania and Bulgaria in the Black Sea [100]. Turkey, the main fishery country for turbot in the Black Sea, did not support these management restrictions. Accordingly to Hulak et al (2021), the decrease in turbot landings reported in the Turkish catch statistics, after 2002, is a consequence of the effort of riparian countries (Ukraine, Romania, and Bulgaria) to stop illegal Turkish turbot fisheries in their EEZ. We presume that abovementioned restrictions/controls could be the consequence of the increase in biomass predicted by stock assessment models.

Ongoing efforts have been made to manage and conserve key fish stocks by European Commission who aiming to achieve sustainable fishery for the Mediterranean and Black Seas in line with the 2030 Strategy of the GFCM. Recent management measures proposals includes total allowable catches (TAC) and quotas for turbot and catch limits also for sprat [101].

3.3. Ecospace

Ecospace output represented the species distribution of the functional groups driven by prey distributions, and avoidance of predators (including fisheries) within their preferred habitat preference (e.g., depth, oxygen) (S3 Table in S1 Appendix). It should be noted that, the output distributions of the modelled functional groups should be considered with caution, due to the lack of spatial data for many Black Sea species. Yet, qualitative validation can be cross-checked with available information. For instance, the predicted spatial biomass distributions of marine mammals (S5 Fig in S1 Appendix, upper panel) in shallower regions for bottlenose dolphins and harbour porpoises and offshore areas for common dolphins are in line with aerial surveys and species distribution models [102, 103]. Seabirds showed distributions over waters shallower than 500 m (S5 Fig in S1 Appendix, lower panel), mainly in the North and West of the basin where their preys are distributed following a similar pattern as observed by Piroddi et al. [26] for the neighbouring Mediterranean Sea.

The distributions of commercial migratory pelagic species, such as Atlantic bonito, bluefish and the mackerels groups were only driven by their prey distribution as migration patterns of these species were not considered in this study (Fig 6 and S6 Fig in S1 Appendix).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Biomass spatial distributions of the stock-assessed species.

Data are presented as t*km^-2 averaged over all modelled years 1995–2021.

https://doi.org/10.1371/journal.pone.0312170.g006

Among the stock-assessed species (Fig 6), sprat and anchovy were the species that mainly drove the distribution of many pelagic predators in the model.

The anchovies exhibit a strong seasonal pattern in which, in winter, they migrate between the North-Western shelf and the south of the basin. Notably, over winter, the highest anchovy biomass and a significant portion of anchovy catches are associated with the southeast region [95, 104]. Annual averaged maps presented here may not represent these spatial seasonal patterns successfully; however, overall, our predicted distributions agreed well with those produced by stock assessment models [95, 104].

Benthic species had a confined distribution (Fig 6 and S6 Fig in S1 Appendix) over the continental shelf, due to the anoxic conditions that occur in deeper waters where the annual bottom oxygen concentration is lower than 10 mmol O₂/m² (Fig 1C), whilst many demersal groups have confined distributions for trophic relationships on benthos. The spatial distribution of picked dogfish and turbot (Fig 6) well agreed with previously reported data [96, 105].

A qualitative validation of the Ecospace habitat foraging capacities (HFCs) of stock assessed demersal species whiting, turbot, red mullet and picked dogfish showed an overall agreement with the maps of habitat suitability model extracted from AquaMaps [80] (Fig 7). Differences such as lower habitat capacity values towards the centre of the basin could be due to the different model resolutions between Ecospace and AquaMaps and an overestimation of the species sensitivity at lower bottom oxygen concentration.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Qualitative spatial validation of stocked assessed species.

Habitat capacity computed by Ecospace model (left panels) versus AquaMaps suitability maps (right panels).

https://doi.org/10.1371/journal.pone.0312170.g007

Overall, our results confirmed the highest biodiversity confined in the North-Western shelf, particularly in the shallow neritic waters [22], highlighting even further the necessity to prioritize spatial management actions to preserve the remaining biodiversity.