Discussion
Since the ecological role of taxa heavily depends on their position in the trophic network, we were interested in identifying whether the sensitivity and exposure of taxa to fishing pressure (i.e. vulnerability), completed by taxa’s centrality, could have consequences on the robustness of the trophic network.
The secondary extinction analysis conducted here highlighted that the robustness of the Celtic Sea to fishing is not closely related to the sensitivity of its taxa to this pressure. This behaviour of the network results from the respective characteristics and feeding ecology of the taxa. Indeed, in the Celtic Sea, the most sensitive taxa are medium size elasmobranchs which are not top predators and have medium trophic levels (circa TL=3). Medium trophic levels taxa often have a high structural importance, with usually the largest values of centrality (Scotti & Jordán 2010). However, we showed that this is not the case in the Celtic Sea were the most sensitive taxa are not the most central and are thus, if targeted, unlikely to spread a perturbation to the whole network. Considering both the sensitivity and the exposure to fishing pressure, we showed that none of the taxa considered in this study is highly vulnerable to fishing, which could be linked to the long history of exploitation of the area (Guénette & Gascuel 2012). The most vulnerable taxa include smooth hound, a sensitive and moderately exposed taxon, as well as large piscivorous taxa (cod, hake, anglerfishes, plaice, conger, ling) that are not very sensitive but are highly fished. Amongst these large piscivorous taxa, cod, anglerfish and hake are central taxa and are likely to propagate a perturbation to the whole network through top-down control. These species are important commercial stocks under quotas in the area, whose decrease should lead to significant cascading impacts in the trophic network.
The robustness of the Celtic Sea network to the removal of taxa with many preys and highly connected taxa was relatively high. The removal of taxa with many preys leads to the fastest collapse of the network, but only happens after removing 60% of the taxa, which is far from realistic conditions. Connectance decreases before the network collapses, making in a first instance the propagation of a perturbation less likely and the network more robust. Likewise, the network seems relatively robust to the removal of the most central taxa (decrease in connectance and a number of secondary extinctions similar to the one expected by chance). This finding contradicts the expected low robustness of a network facing the removal of its most connected taxa (Dunne & Williams 2009; Staniczenko et al. 2010; Curtsdotteret al. 2011; De Visser et al. 2011) but see (Allesina & Pascual 2009). Nevertheless, the robustness of the network facing these two removal sequences is in line with the observed increase in modularity and decrease in nestedness (Fig. S2). Indeed, the increased modularity indicates an increased compartmentalization, which is known to promote stability by restricting the spread of the perturbation outside the module (Thébault & Fontaine 2010). On the other hand, a decrease in nestedness implies the removal of the redundant trophic interactions first (Nordstrom et al. 2015), which translates here into the removal of whiting Merlangus merlangius , megrim and squid loligo sp. that are preyed upon by generalist taxa.
The removal of taxa with many predators leads to the lowest robustness of the trophic network. Taxa with the largest number of predators here belong to intermediate trophic levels (between 3 and 3.5) namely herringClupea harengus , sprat Sprattus sprattus , sardine Sardina pilchardus , dragonet Callionymus lyra , poutsTrisopterus esmarkii and Trisopterus minutus and horse mackerel Trachurus trachurus (Table S1). These forage species account for a large proportion of the biomass of the taxa considered in this study but also a large proportion in the catches (ICES 2018b; Hernvann & Gascuel 2020). These species are crucial for the network stability as their predators display medium to high trophic levels and are both benthic and pelagic. Hence, forage species allow the coupling between these two pathways, which has been shown to participate to trophic network stability and resilience (Blanchard et al. 2011). In addition, if affected by a perturbation, these taxa heavily destabilize the network of the Celtic Sea (Moullec et al. 2017).
The food-web topology reconstructed here integrates data from trophic studies covering a long time-span. Thus, the restructuration of the network due to changes in diets could not be investigated. Nonetheless, this makes the response of the network to removal-scenarios interpretable regarding the long-term history of the ecosystem. The relatively high robustness of the network to the removal of taxa with many preys could be one of the stability factors (with predation control of benthopelagic predator larvae by pelagic fish, Baum & Worm, 2009) of the more pelagic-dominated state of the ecosystem after the depletion of high trophic levels before 1980 (Hernvann & Gascuel 2020).
Finally, the network has low robustness to the removal of the most exposed taxa at early stage of perturbation (i.e. when removing the 7% of the most exposed taxa). These highly exploited taxa are queen scallopAequipecten opercularis , king scallop Pecten maximus , edible crab, European spider crab Maja brachydactyla , cod, sprat and hake. This increase in connectance originates from a faster decrease in the number of potential interactions than in the realized ones and is due to these removed taxa having in general fewer feeding links than the averaged species in the network. This raised one of the limitations of network reconstruction, since the taxa considered were sampled with a bottom trawl that is not adapted to sample the basal components of the network (phyto and zooplankton are missing while benthos is underrepresented), as well as the top predators. Nevertheless, taxa included here are megafauna with the highest occurrence and account for most part of the network, enabling notably the survival of commercial taxa. These taxa are thus considered to provide a representative picture of the Celtic Sea ecosystems. Trophic levels were computed from local isotopic data collected in the Celtic Sea whereas the trophic links were taken from the literature. Thus, there might be a mismatch between the trophic position and the trophic links of some taxa with taxon feeding on taxon at slightly higher trophic level. However, because the computation the centrality score of one taxon is not based on data specific to our study area, it could be applied to other North-East Atlantic studies. In addition, we did not consider the fluxes of biomass between taxa, which could modulate our findings. Indeed, it might influence the spread of a perturbation, with a larger spread between taxa linked by a large flux of biomass.
Network theory has been identified as a helpful tool to support ecosystem-based fisheries management (Gaichas & Francis 2008; Deeet al. 2017). Exposure and trait-based sensitivity metrics relative to fishing brought here a complementary information to the network analysis. Indeed, our study suggests that widely used mesoscale metrics such as centrality were not always adapted to prioritize species conservation to maintain the structure and the functioning of the network. On the contrary, our exposure metric highlighted that in some cases, the current fishing exploitation pressure should prevail on topology-based metrics, while sensitivity must be considered as it implies different abilities of species to tolerate various exposure levels. Such metrics are particularly promising in the context of exploring potential new fishing management strategies. In particular, integrating the sensitivity to fishing, they could be used to investigate the risk of exploiting new species regarding to their own productivity potential (Zhou et al. 2019).
Studying fishing perturbation, we showed that the trophic network of the Celtic Sea was the least robust to the simulated loss of taxa with many predators (i.e. forage taxa) and of the 7% taxa the most exposed to fishing pressure. Estimating the sensitivity to fishing of the 69 taxa of the network was insufficient to predict its robustness since the simulated removal of the taxa most sensitive led to a robustness level similar as that of a random removal sequence. This study focused on fishing since this variable has a documented impact on taxa’s biomass in the Celtic Sea, due to the long history of exploitation of this ecosystem (Hernvann & Gascuel 2020). However, climate change will likely become the main driver of this ecosystem in the coming years. The framework proposed in this work could easily be adapted to assess species sensitivity to temperature or pH tolerance by selecting traits known to respond to these pressures. Ultimately, such a framework could be used in complement of management tools to indicate which taxon could impair ecological network structure and ecosystem functioning under increasing global change. It could also unravel early warnings about the loss of certain taxa that could jeopardize a trophic network more than their sensitivity at the taxon’s level could suggest.