The Azores is a Portuguese archipelago composed of nine islands situated on the Mid-Atlantic ridge (Figure 1) with an extensive EEZ of about 1 million km². As a volcanic archipelago of recent origin, the islands have narrow shelves and steep slopes, and the surrounding waters have an average depth of 3000 m with only 0.8% of the EEZ being less than 500 m deep. The highly irregular submarine topography contains vast undersea mountain ranges, with around 100 large and 400 small seamounts-like features (Morato et al., 2008, 2013), deep-water coral gardens and reefs (Sampaio et al., 2012; Braga-Henriques et al., 2013; de Matos et al., 2014; Tempera et al., 2015), sponge grounds (Tempera et al., 2012, 2013), and hydrothermal vents (Cardigos et al., 2005; Cuvelier et al., 2009). In winter a deep mixed layer is present at 150 m and average sea surface temperature (SST) is about 15–16°C. During summer, a seasonal thermocline develops at 40–100 m and the average SST is typically 22–24°C (Amorim et al., in review).

The region is characterized by very complex ocean circulation patterns. Large scale circulation is dominated by the eastward-flowing Gulf Stream, which forms a current system with many unstable eddies and meanders, the cold North Atlantic Current in the north, and the warm Azores Current in the south (Santos et al., 1995; Alves and Verdière, 1999; Johnson and Stevens, 2000; Bashmachnikov et al., 2009). Various water masses are present around the Azores. North Atlantic Central Water occurs above a permanent thermocline, located at depths shallower than 700 m, North Atlantic Deep Water is the dominant water mass below 2000 m depths, and at intermediate depths, northern sub-polar waters and Antarctic Intermediate Water predominate, but Mediterranean Outflow Water can also occur (Santos et al., 1995; Johnson and Stevens, 2000; Mann and Lazier, 2006).

For this study, we confined the study area to the boundary of the EEZ, which covers an area of 954,563 km². The area includes the deep-sea, open-ocean, some seamounts, parts of the Mid Atlantic Ridge and island slopes (Figure 1). The reference year of 1997 was chosen as most of the data used to construct the base model (diet and growth parameters) originated from that year.

Marine resources are central to the Azores' local economy, but bottom fishing grounds are limited and scattered on the island slopes and seamounts (da Silva and Pinho, 2007; Diogo et al., 2015). The Azores fleet is dominated by a small-scale artisanal fishing fleet (Carvalho et al., 2011) with only 10–20% of the fleet being a large-scale, semi-industrial fishing fleet. An overview of the main fisheries in the Azores, their gear types, target species, fishing vessels and regulations was taken from Gaspar (2011). A total of 11 Azorean fisheries were included in the model: the deep-water bottom longline and handline fisheries targeting mostly deep-water demersal fishes such as blackspot seabream (Pagellus bogaraveo), wreckfish (Polyprion americanus), alfonsinos (Beryx spp.) and the blackbelly rosefish (Helicolenus dactylopterus); the Azores pelagic longline, Portuguese mainland pelagic longline, and the foreign pelagic longline fisheries targeting swordfish (Xiphias gladius) and blue shark (Prionace glauca); the pole and line tuna fishery (including the live-bait); the small-size pelagic fisheries targeting mostly blue jack mackerel (Trachurus picturatus) and chub mackerel (Scomber colias); the drifting deep-water longline targeting black scabbardfish (Aphanopus carbo) which is a recent fishery in the Azores (Machete et al., 2011); the commercial coastal invertebrates; the recreational fishing; the experimental bottom trawling; and the squid (Loligo forbesii) fisheries.

Ecopath with Ecosim (EwE) is a food-web modeling facility that can be used to build trophic static mass-balanced snapshots (Ecopath) and to create temporal dynamics (Ecosim) of an ecosystem (Christensen and Pauly, 1992; Walters et al., 1997, 2000; Pauly et al., 2000; Christensen and Walters, 2004; Christensen et al., 2008). EwE has been widely adopted all over the world (Colléter et al., 2015) and has led to some ground-breaking science (Pauly and Christensen, 1994; Pauly et al., 1998; Watson and Pauly, 2001; Branch et al., 2010; Smith et al., 2011; Irigoien et al., 2014). However, it should be used with caution to avoid common mistakes and pitfalls (Ainsworth and Walters, 2015). EwE has been described in detail elsewhere (e.g., ecopath.org; Christensen and Pauly, 1992; Walters et al., 1997; Christensen and Walters, 2004), with the best practice in Ecopath described recently (Heymans et al., 2016).

Ecopath models parameterization is based in two master equations, one describing the production term and the other the energy balance for each functional group. The first master equation ensures a mass balance between groups and expresses production as a function of the catch, predation, net migration, biomass accumulation and other mortality (Equation 1). The second master equation is based on the principle of conservation of matter within each group (Equation 2; Christensen and Walters, 2004). Each group is parameterised with its biomass (B, t·km⁻²), production over biomass ratio or production rate (P/B, year⁻¹), consumption over biomass ratio or consumption rate (Q/B, year⁻¹), the prey-predator interaction in the form of a diet composition (DC) table, ecotrophic efficiency (EE_i), the biomass accumulation rate (BA_i, year⁻¹) and the net migration rate (E_i, year⁻¹).

The current version of the Azores model was built upon previous models developed for this region and associated seamounts (Guénette and Morato, 2001; Morato and Pitcher, 2002; Morato et al., 2009). The present model focused mostly on intermediate and deep-water species present in the Azores ecosystem and used, when possible, recent and local data for model parameterization. Species with biological and ecological similarities were grouped into functional groups or biomass pools. Fish species lists were compiled from previous models, and completed with more recent biodiversity studies of the Azores (Supplementary Data Sheet 1). The present model took into consideration 387 fish species representing about 66% of the known marine fish biodiversity (WoRMS Editorial Board, 2016). Non-fish functional groups were defined based on a previous Ecopath model of the Azores (Guénette and Morato, 2001) and an Ecopath model for a hypothetical seamount in the North Atlantic (Morato et al., 2009). With the exception of marine mammals (16 most common species, representing 66% of the known biodiversity; Mónica Silva, pers. comm.), seabirds (8 most common species, 73% of reported nesting species; Verónica Neves, pers. comm.), and sea turtles (3 most common species, 60% of the reported species; Marco Silva, pers. comm.), most of the non-fish groups were poorly represented in the model due to the limited amount of information available. In this model, energy related parameters are expressed in t·km⁻² of wet weight and the temporal unit is year⁻¹.

Fish species present in the Azores EEZ were compiled based on a checklist of marine fishes of the Azores (Santos et al., 1997), an updated list of commercial species caught in the Azores for the period 1950–2010 (Pham et al., 2013), a list of fish species caught on fisheries research cruises (Menezes, unpublished data), a list of deep-pelagic fishes compiled during mesopelagic trawling surveys (Sutton et al., 2008), and a list of coastal species sighted during a sub-aquatic visual census program (Afonso, 2002). Of the 387 fish species complied, only 223 (representing 38% of the known fish biodiversity) were included in the model because of data limitations. All of the selected species were allocated stepwise to 29 functional groups after compiling a dataset with diet composition, asymptotic length and average habitat depth for each species, gathered from local studies and completed with Fishbase data (Froese and Pauly, 2015). In addition, some fish were separated into single species functional groups because of their commercial interest and/or to allow specific management simulations. These are: H. dactylopterus, Conger conger, Pontinus kuhlii, Raja clavata, Phycis phycis, Pagrus pagrus, Beryx splendens, Beryx decadactylus, P. bogaraveo, Mora moro, L. caudatus. The model presented here consisted of 45 functional groups: one detritus group, two primary producer groups, eight invertebrate groups, 29 fish groups, three marine mammal groups, one sea-turtle and one seabird group (Supplementary Data Sheet 1).

Model parameters, P/B, Q/B, and production of consumption ratio (P/Q, unitless) were estimated from the literature, with preference to studies within our area or from similar areas, or using empirical equations (Pauly, 1980; Palomares and Pauly, 1998). Habitat area fraction, which is the habitat area to total model area ratio, for each group was calculated using habitat depth ranges compiled from local studies (e.g., Menezes et al., 2006) and Fishbase, and converted into surface areas using bathymetric grid of the Azores. Details on the calculations or sources of these model parameters and the habitat area fraction are presented in Supplementary Data Sheet 1.

The model pedigree describing the origin and quality of each parameter was calculated and used to analyse our hypothesis that there is sufficient data to construct an ecosystem model of the Azores. This was done by comparing with the estimated pedigree values with the reported ranges in Colléter et al. (2015). The model pedigree is also used to assign confidence intervals to the data inputs (Pauly et al., 2000).

A diet matrix was assembled using preferentially local literature on stomach content analyses, completed with other literature and adapted using empirical knowledge (Supplementary Table 1). To constrain the model and due to a lack of direct biomass assessments, biomass expressed in tonnes of wet weight per square kilometer of species' habitat, was estimated empirically for the two primary producer groups, the detritus group, and four top predator groups. For the remaining groups, biomass was left to be estimated by the model, assigning different EE values to the different groups (Supplementary Table 1). To balance the model, diet compositions were modified, and for some fish groups, ratios of P/B were left to be estimated by the model, using empirical P/Q values as input.

Total marine catch data was obtained from Pham et al. (2013). The dataset contains both official fishery statistics and estimated illegal, unreported, and unregulated catch (IUU) within the Azores EEZ. However, discards (fish returned to the sea), were not reported separately. Species catch data was assigned to the different fishing fleets and the functional groups. Catch data that could not be assigned to a specific functional group or fleet (e.g., unidentified marine species) were redistributed into the groups exploited by the various fleets. Catch data was constructed for the reference year 1997 (Figure 2; Supplementary Table 2) and then expressed in tonnes of wet weight per square kilometer of the model area. The tuna pole and line fishery is together with the deep-water bottom longline and handline fishery accountable for the majority of the catch of marine resources in the Azores during the studied period (Figure 3). Before the prohibition of bottom trawling in a large part of the EEZ, some trawling experiments were performed in 2001–2002 and were responsible for a significant amount of catch (Melo and Menezes, 2002). The total marine production in the Azores currently stands at over 20,000 tonnes, corresponding to about € 60 million in landed value (Pham et al., 2013).

After mass-balancing the model, a trophic network analysis was performed. For each functional group, a fractional trophic level (TL) was calculated based on the diet. An omnivory index, representing the trophic specialization of the predator (Christensen et al., 2008), was calculated for each consumer group. The “Keystoneness” index (KS), as defined by Libralato et al. (2006) was also calculated for each functional group. The KS allow the identification of the keystone species, i.e., relatively low biomass groups that have a structuring role in their food webs, in the given ecosystem. The mixed trophic impact (MTI) routine, developed by Ulanowicz and Puccia (1990), was applied to evaluate the impact of direct and indirect interactions on the static food web model. The routine was used to assess the theoretical impacts of increased biomass of a particular group on the biomass of the other groups, assuming that the trophic structure remains the same.

A selection of ecosystem indicators were calculated, allowing for a comparison of ecosystem properties with other models. Amongst others, the sum of all consumption, exports, respiratory flows, flows into detritus, and the ratio of total primary production/total respiration were calculated. The System Omnivory Index (OI), defined as the average OI of all consumers weighted by the logarithm of their consumption, was calculated. This index is a measure of the trophic specialization of the whole system. The Total System Throughput (TST) was calculated by totalling all biomass fluxes occurring in the system.

The various parameters for the balanced Ecopath model of the Azores ecosystem are presented in Table 1. Additional to those parameters, the Pedigree index was estimated to be 0.53 which although being similar to many other models (e.g., Corrales et al., 2015) showed that some input data should be improved. Due to the scarcity of biomass data, most of the ecotrophic efficiencies (EE's) had to be estimated based on expert knowledge. Nevertheless, EE of the top predator groups (group 40–43, Table 1) that were estimated by the balanced model were generally low, ranging from 0.13 (toothed whales) to 0.46 (baleen whales) and indicating that large fractions of the production of those groups is not being used in the modeled system. The EE's of phytoplankton and algae are 0.11 and 0.013, respectively; suggesting low utilization of primary production in the system.

The total biomass (excluding detritus) of the modeled ecosystem for the whole area was calculated as 24.7 t km⁻² (Table 1). Primary producers form 15.4% of the total biomass (3.8 t km⁻²) and fish biomass contributed to 14.8% (3.7 t km⁻²). The largest part of the total ecosystem biomass, 17.1 t km⁻² (69.3%), was composed by the invertebrate and zooplankton groups, while the non-fish groups occupying the higher trophic levels (seabirds, dolphins, baleen whales and toothed whales) contributed only to 0.5% of the total biomass (0.12 t km⁻²). The trophic spectra of the ecosystem is shown in Figure 4.

The Azores ecosystem model included five trophic levels with toothed whales and C. conger, presenting the top predators in the ecosystem with TL of 4.64 and 4.61, respectively. Other functional groups with a TL > 4 included the deepwater sharks, large-size pelagic fish, large-size bathydemersal fish, L. caudatus, large-size demersal fish, dolphins, pelagic sharks, M. moro, R. clavata, rays and other sharks, seabirds, tunas, H. dactylopterus, P. phycis, P. bogaraveo, and P. kuhlii. The remaining fish groups have a TL ranging from 2.99 (small-size pelagic fish) to 3.90 (bathypelagic fish). Invertebrate functional groups were estimated to have a TL between 2.05 (benthic filter feeders) and 2.77 (shrimps), with the exception of cephalopods who were estimated to have a TL of 3.72. Zooplankton functional groups have a TL of 2.00 (small-size) and 2.58 (large-size and gelatinous). The food web and flow diagram demonstrated the complex structure of the ecosystem (Figure 5).

The mean trophic level for the total marine catch of the Azores (Figure 6) was estimated to be 3.95, matching the trophic level of the bathypelagic and medium-size pelagic fish groups, but situated below the trophic levels of most of the top predators in the system. Drifting deepwater longline and the pelagic longline showed the highest values of 4.40 and 4.34 respectively, approximating the trophic levels of some top predator groups. Bottom logline and handline fishery showed a trophic level of 4.19, while the small-size pelagic fisheries showed the lowest trophic level of 2.99.

The Omnivory Index (Table 1) showed that most groups were feeding on few trophic levels. The OI ranged from 0.04 to 0.58, with shallow-water large-size fish, cephalopods, shallow-water medium-size fish, and rays and other sharks showing the highest index and sea turtles, benthic filter feeders, small-size bathydemersal fish and toothed whales showing the lowest values of OI.

The Mixed Trophic Impact (MTI) analysis (Figure 7) revealed the direct and indirect impact of an increase/decrease in biomass of an impacting group or fisheries catch on an impacted group or fishery. The MTI indices ranged from 1.8, representing a strong positive effect of cephalopods on toothed whales, to −2.0, revealing a strong negative effect of toothed whales on its main prey. The MTI analyses showed the influence of toothed whales, pelagic sharks, cephalopods and small-size pelagic fish in the ecosystem, having both strong positive and negative impacts in many components of the ecosystem. For example, toothed whales had a strong negative impact on their main prey (e.g., cephalopods) but also had a positive impact on other groups or species (e.g., large-size pelagic fish or Beryx spp.), through mixed trophic links such as removal of their predators (Figure 7). On the other hand, small-size pelagic fish had a strong positive impact on its predators (e.g., Seabirds or tuna) and a negative impact though complex trophic links (e.g., toothed whales). Most groups will have a positive impact of their fisheries while most fisheries showed a strong negative effect on its target species (Figure 7). However, some groups had a negative impact on some fisheries as for example, toothed whales and cephalopods were shown to have a negative impact on the pole and line and pelagic longline fisheries. Dolphins were found not to have a significant impact on any type of fisheries.

The Keystoneness index calculated according to Libralato et al. (2006) were highest for pelagic sharks (#38, KS = 0.66), toothed whales (#44, KS = 0.64) and cephalopods (#6, KS = 0.61; Figure 8). These groups also showed the highest relative total impact, highlighting their importance in the ecosystem structure. The groups of the small-size pelagic fish and bathypelagic fish (group #14 and #18) also showed high keystoneness (KS = 0.39 and 0.30, respectively) indicating an important role as prey in the food web.

Ecological summary characteristics of the system are represented in Table 2, along with general ecosystem statistics for other deep-sea models. The ratio of total primary production/total respiration (Pp/R) was 8.2 while the total primary production/total biomass ratio (Pp/B) was 67.7. The total transfer efficiency of the system was estimated to be 18.8%.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.