Future trends of marine fish biomass distributions from the North Sea to the Barents Sea

The study addresses how climate warming will reshape marine fish distributions, richness, dominance, and biomass across the Northeast Atlantic from the North Sea to the Barents Sea. Prior observations show poleward range expansions of warmer-water species and contractions of colder-water species, with the Arctic warming nearly four times faster than the global average and already experiencing compositional changes and richness increases. These shifts have significant implications for ecosystems, fisheries, conservation planning, and potential transboundary conflicts. Existing future projections often omit abundance/biomass, model species independently, or focus on few common species. To overcome these limitations, the authors apply a joint species distribution model (HMSC J-SDM) to jointly model 107 species (61 with biomass), capture co-occurrence patterns and phylogenetic structure, and project changes under SSP1-2.6, SSP2-4.5, and SSP5-8.5 to 2050 and 2100. The study aims to quantify future trends in richness, relative dominance, range size, biomass (including core areas), community centroid shifts, and fragmentation, and to assess the role of species traits and zoogeographic groups, with special attention to Arctic species’ risks given limited scope for northward shifts due to shelf constraints.

Background literature documents widespread poleward shifts in marine taxa driven by local climate velocities and uneven warming patterns. In the North Sea and Barents Sea, past decades have seen northward displacements across plankton, pelagic and demersal fish, increased species richness, arrival of boreal species, and declines in Arctic species. Prior SDM-based projections informed policy but commonly lacked biomass considerations, modelled species independently, or focused on data-rich commercial species. Joint species distribution models (notably HMSC) improve on traditional SDMs by leveraging shared environmental responses, phylogenetic niche conservatism, and latent co-occurrence structures, enhancing predictions for rare species. Previous studies also identified bottom temperature as a key driver of fish community structure in the region. However, taxonomic resolution and uncertainty limit global mechanistic projections, and species’ ability to track shifting habitats may be constrained by bathymetry (e.g., Arctic shelf limits), elevating extinction risks for polar species.

Study area and data: Fish biomass (relative abundance as CPUE, fish/min) and occurrences were compiled from FishGlob and three fishery-independent bottom trawl surveys (NS-IBTS, Norwegian Sea coastal survey, Nor-BTS) from 2004–2022 across the continental shelf (≤500 m) from the North Sea to the Barents Sea. Only Campelen and GOV bottom trawls with 20 mm mesh were used; catches were standardized by effort. Rare species (<100 hauls or <10 years) and hauls without environmental data (<1%) were excluded, yielding 16,345 unique hauls and 107 fish species (biomass modelling subset later reduced to 61). Environmental variables: Candidate predictors included bottom and surface temperature, sea ice concentration (where relevant), eastward and northward currents, bottom dissolved oxygen, phytoplankton concentration, and depth. Data sources: Copernicus Global Ocean Physics Reanalysis (0.08°) for temperature and currents; Copernicus Global Ocean Biogeochemistry Hindcast (0.25°) for oxygen and primary productivity; Bio-ORACLE for depth (0.08°). Monthly means were used (annual mean for sea ice). Collinearity was addressed using VIF>4; surface temperature was removed (highly correlated with bottom temperature). Quadratic terms were included for depth and bottom temperature. Future environmental layers: Future mean annual layers were from IPSL-CMIP6 (the only CMIP6 model with all predictors), for SSP1-2.6 (+1.6 °C), SSP2-4.5 (+2.6 °C), and SSP5-8.5 (+4.5 °C) in 10-year increments from 2030–2100; present-day was approximated by 2010–2013 means. Total of 25 time periods (3 scenarios × 8 future periods + present-day). Statistical modelling: A two-part hurdle J-SDM under the HMSC framework was fitted: (1) occurrence using a binomial probit, and (2) biomass using Gaussian models on log(CPUE). Spatial autocorrelation was modelled via spatial random effects using Gaussian Predictive Processes with 183 knots. The model incorporated phylogenetic structure (NCBI Common Taxonomy Tree) and species traits (maximum length, age at maturity, fecundity, habitat, trophic level, preferred temperature, maximum depth, and a zoogeographic class: Arctic, Arctic-Boreal, Boreal, Deep-water, Subtropical, Temperate). MCMC and validation: Four MCMC chains, thinning 500, 250 samples per chain after burn-in of 62,500. Convergence diagnosed with Gelman–Rubin PSRF (<1.1) and effective sample sizes. Model fit metrics: mean AUC 0.97 (occurrence), R^2 0.54 (biomass). Five-fold cross-validation yielded AUC 0.88 and R^2 0.12. Present-day projected richness correlated with surveyed richness (Pearson r=0.4, p<0.01). Discrepancies between Copernicus (fit) and IPSL (projections) bottom temperature were quantified (overall correlation significant; North Sea weaker: 0.61; excluding North Sea: 0.85). A MESS analysis identified poorly sampled environmental spaces requiring caution. Species inclusion for biomass analyses: Species with mean cross-validated R^2>0.05 in the biomass model were retained (61 species; mean R^2=0.22, range 0.05–0.53). Final biomass projections were the product of binomial occurrence and Gaussian log-biomass components, with projected biomass below half the minimum observed set to zero to impose a biologically meaningful presence threshold. Projections and metrics: Spatial projections were generated for present-day (2010–2013) and decadal steps to 2100 for each SSP, excluding the spatial random effect in future projections to avoid extrapolating fixed spatial structures. Presence-absence thresholds maximizing TSS were computed per species for range calculations. Range metrics included: total geographic range (km^2), total biomass (sum log-CPUE), core range (cells ≥ present-day 90th percentile CPUE), and core biomass (CPUE within core range). Rates of change (2010–2100) were estimated via linear regression. Geographic range fragmentation was quantified as number of polygons, mean polygon area, and mean inter-polygon distance. Species richness was the sum of probabilities of occurrence across species (n=107), and relative dominance was the highest species’ CPUE as a percentage of total CPUE per cell; dominant species were identified per cell.

Environmental drivers and structure: Depth explained the largest share of deviance (occurrence: mean 58%; biomass: 49%), followed by bottom temperature (32% and 19%, respectively) and spatial random effects (8% and 29%). Other variables contributed <1% on average but varied by species (e.g., phytoplankton explained 20% for Amblyraja radiata occurrence; sea ice 15% for Clupea harengus CPUE). Strong phylogenetic niche conservatism was detected (rho=0.58 [0.41–0.73] for occurrence; rho=0.96 [0.89–0.95] for CPUE) and significant residual co-occurrence patterns persisted after accounting for fixed effects. Richness and dominance: Species richness is projected to increase markedly in the northern Barents Sea, roughly doubling around Svalbard and Norway’s north coast by 2100, with no increases or slight decreases in the deep Bear Island Trench. In the North Sea, modest richness increases are projected centrally and small declines elsewhere. Relative dominance was weakly inversely related to richness (Pearson r=−0.09, p<0.01) and declined in the northern/eastern Barents Sea. Polar cod (Boreogadus saida) lost dominance from 29% of the study area at present to 0% by 2100 under SSP5-8.5; increases in dominance share were projected for Norway pout (Trisopterus esmarkii), blue whiting (Micromesistius poutassou), and whiting (Merlangius merlangus). The number of dominant species across the study area declined from 10 (present) to 6 (SSP5-8.5, 2100), indicating homogenization. Range and biomass responses: Across scenarios, most species increased their geographic range and biomass, especially under higher emissions, but responses diverged by zoogeographic group. Arctic (n=3 for biomass) and Arctic-Boreal (n=8) species showed strong declines in range and biomass in all scenarios, while Boreal (n=40) and warmer-water (temperate/subtropical; n=5) species expanded ranges and increased biomass. Despite expansions, losses of currently abundant Arctic species (notably polar cod and capelin Mallotus villosus) were not offset by gains in Boreal species, producing an overall decline in fish biomass in Arctic latitudes by 2100 (consistent across scenarios; spatial maps show lower total log-CPUE). Centroid shifts: Community-wide northward and eastward shifts accelerated with emissions. Mean range centroids shifted 0.9 km/yr N and 0.3 km/yr E (SSP1-2.6) up to 3.2 km/yr N and 1.1 km/yr E (SSP5-8.5). Biomass-weighted shifts were 1.0/0.8 km/yr (N/E) under SSP1-2.6 and 3.7/2.1 km/yr under SSP5-8.5. Core range (top 10% biomass) shifted fastest: 1.1/0.5 km/yr (SSP1-2.6) to 4.8/3.2 km/yr (SSP5-8.5). Fragmentation: No clear community-wide trend. Increases in number of polygons and decreases in polygon area with higher SSPs were driven by declines in Arctic species’ polygon areas and increases in polygon number for warmer-water species, yielding no consistent net fragmentation signal. Species at risk: Projected local extirpations include Atlantic poacher (Leptagonus decagonus; Boreal-Arctic) under SSP5-8.5; bigeye sculpin (Triglops nybelini) and pale eelpout (Lycodes pallidus) under SSP2-4.5 and SSP5-8.5. Polar cod loses dominance everywhere and is projected to nearly disappear from the study area by 2100 under SSP5-8.5, with >50% declines in area and biomass under all scenarios. Trait effects: Zoogeographic class best explained variation in rates of change: Arctic/Arctic-Boreal negative, Boreal/Temperate/Subtropical/Deep-water positive (p<0.05). Higher trophic level associated with larger range extent; maximum length with higher biomass; maximum depth with larger core range (p<0.05).

The findings show that ongoing borealization will intensify: richness increases and dominance declines accompany warming, particularly in the Barents Sea. However, the loss of Arctic and Boreal-Arctic keystone fishes (e.g., polar cod, capelin) leads to net biomass declines in Arctic latitudes, raising concerns about food web stability and dependent predators (e.g., ringed seals). Limited continental shelf at higher latitudes restricts northward shifts of Arctic demersal fish, elevating regional and potentially global extinction risks without eastern refugia. While warmer-water and some boreal species are projected to expand (creating potential new fisheries opportunities), iconic boreal species like Atlantic cod may see biomass reductions by century’s end, indicating that warming may ultimately surpass their thermal optima. Community homogenization (fewer dominant species, broader sharing of biomass) suggests altered ecosystem functioning and may influence predictability of community properties. The study underscores the need for adaptive fisheries management anticipating transboundary stock redistributions and potential conflicts, and for conservation strategies that consider climate-driven displacement from existing protected areas. Caution is advised when interpreting projections in the North Sea due to environmental data mismatches and unaccounted immigrants from outside the study area.

Using a joint species distribution modelling framework that incorporates co-occurrence, phylogeny, and traits, the study projects that by 2050–2100 warming will increase species richness and expand the ranges and biomass of boreal and warmer-water fishes, but cause severe declines and local extirpations of Arctic and Boreal-Arctic fishes, resulting in overall biomass reductions in Arctic latitudes and a homogenization of dominant species. Key contributions include quantifying environmental drivers (depth and bottom temperature), demonstrating strong phylogenetic niche structure, providing rates and directions of centroid and core-range shifts, and identifying species and zoogeographic groups at highest risk. Future research should: integrate fishing pressure and fleet behavior into projections; improve environmental-data harmonization and resolution; collect finer-scale biomass data, especially for Arctic demersal species and range edges; develop validation frameworks targeting distributional change rather than static patterns; and explore potential climate refugia pathways (eastern routes) for vulnerable Arctic taxa.

- Fishing impacts were not included; fishing can interact synergistically with climate change and alter future biomass and distributions. - Predictive environmental data came from IPSL-CMIP6 while model fitting used Copernicus datasets; despite significant correlations, discrepancies (notably in the North Sea) reduce confidence. - MESS analysis indicates portions of future environmental space were poorly sampled during training, warranting caution for those projections. - Limited occurrence/biomass data for several Arctic species reduced predictive skill; borrowing information via phylogeny/traits was insufficient for some species. - SDMs generally predict static distributions better than temporal changes; robust validation of change requires temporally independent, high-resolution datasets that are scarce. - Environmental predictors (e.g., phytoplankton, oxygen) had coarser resolution than others, potentially underestimating their effects. - Spatial random effects were excluded from future projections to avoid extrapolating spatial structures, which may omit persistent spatial processes. - Southern boundary effects: species immigrating from outside the study area (especially into the North Sea) are not accounted for, potentially biasing richness and dominance estimates there.