How the projections are made (delta method). The climate-model ensemble (FutureMares, CMIP6, ~9 km) is only used for how much temperature, salinity and oxygen change from today to each future year. That change is added to the real, high-resolution Copernicus ocean data (~2 km, 39 depth levels), re-using real years so that natural year-to-year variability is kept. Scenarios: SSP2-4.5 (intermediate) and SSP5-8.5 (high emissions). How each species' suitability is scored is explained in the Methods tab.
Animation: one frame per month; columns = observed 2014-2024 and four future periods, rows = scenario. Seasonal cycle: dashed = observed periods, solid = future decades.
Monthly suitability, observed 2014-2024 vs projected periods (2026-2099), both scenarios.
Seasonal cycle: domain-mean suitability by month, observed periods and projected decades.
Winter gains, summer losses -- in both trophic roles. The 70 modelled species are grouped into Large Predators (17 piscivorous fish: fish with trophic level ≥ 3.5 and ≥ 60 cm, plus sea trout) and Mesopredators (53 species: all other fish and the predatory decapods). By 2090-2099 under SSP5-8.5, mean winter (December-March) suitability rises by 69% in Large Predators and by 137% in Mesopredators relative to 1994-2022, because warmer winters lift conditions above the species' cold limits. Mean summer (June-September) suitability falls by 24% and 38%. By 2080-2099, 68 of the 70 species lose summer suitability and 67 gain in winter.
The whole-year average therefore rises slightly (Large Predators +6%, Mesopredators +11%): winter gains outweigh summer losses, which is not an overall improvement. The two roles follow the same pattern and do not differ statistically (Kruskal-Wallis p = 0.41, SSP5-8.5); which species win or lose depends on their biogeographic affinity (next tab).
1. Trophic roles over time. Change in mean habitat suitability per decade, relative to 1994-2022, by season. Mean of the species' log response ratios, back-transformed to %; bars = 95% CI across species. Grey background = observed decades.
2. Example species. One boreal (cold-water, top row) and one non-boreal species (bottom row) per trophic group. Change in whole-year mean suitability relative to 1994-2022; thin = single years, bold = 10-year rolling mean; labels = change by 2080-2099.
3. Every species: winter vs summer. Change in winter (y) versus summer (x) mean suitability from 1994-2022, for mid-century (2041-2060, top) and end of century (2080-2099, bottom) under both scenarios. Dot size = catch in the Swedish coastal survey (2000-2024) relative to the most-caught species of the same trophic group; the most-caught species are named. Most species sit in the upper left: they gain in winter but lose in summer. Below the dotted line the summer loss is larger than the winter gain.
4. Every species: whole-year change. % change in whole-year mean suitability from 1994-2022, per trophic role, for mid-century (2041-2060) and end of century (2080-2099).
Boreal species lose, non-boreal species gain, in both trophic roles. Each species was classified from the literature, independently of the habitat model, as boreal (B, cold-water, 31 species), non-boreal (NB, Lusitanian or wide Atlantic range, 37 species) or non-native (NN: blue crab, round goby; shown but not tested). By 2080-2099 under SSP5-8.5, the catch-weighted median boreal species loses 13% of its whole-year habitat suitability, while the catch-weighted median non-boreal species gains 30% (unweighted medians: -2% and +23%; Wilcoxon p < 0.001). The same split appears within Large Predators (p = 0.001) and Mesopredators (p < 0.001, Holm-corrected), whereas the two trophic roles themselves do not differ (Kruskal-Wallis p = 0.41). Exceptions: 12 boreal species gain (most clearly Norwegian topknot, longspined bullhead and pollack; ling, three-spined stickleback and plaice stay at about 0%), and one non-boreal species loses (rock cook).
Affinity sources: Engelhard et al. 2011 (ICES J Mar Sci 68: 580-591; boreal / Lusitanian after Yang 1982) for 61 species; Rogers & Millner 1996 and Vaughan et al. 2026 where Engelhard has no entry; great spider crab after Zimina et al. 2015. Shore crab, brown crab, European lobster and brown shrimp are classified from their geographic range (no published classification exists for these decapods); Baltic whitefish by association with its boreal congener.
Change in whole-year mean suitability per species, observed 1994-2022 to 2080-2099 (SSP5-8.5), as log response ratio (right axis: % change). A: all species; B: within each trophic role. Dot size = catch in the Swedish coastal survey (2000-2024) relative to the most-caught species of the same trophic group; violin shape and median line are weighted by this catch. Stars: *** p < 0.001, ** p < 0.01 (unweighted Wilcoxon tests).
What the envelope model does. Every species has a range of water conditions it can live in, taken from AquaMaps / FishBase / SeaLifeBase: a tolerated range (minimum to maximum) and, inside it, a preferred range. For every 2 km grid cell, every depth level and every month, the model turns the local temperature, salinity and oxygen into a suitability score between 0 (unsuitable) and 1 (ideal), and checks whether the depth is within the species' depth range. The curves below show how each variable is turned into a score, using Atlantic cod as the example.
Solid black = curve used in the model. Dashed grey = simple AquaMaps trapezoid (straight lines), shown for comparison. Light grey band = AquaMaps preferred range.
A. Temperature: a thermal performance curve
Inside the preferred range (Tpmin to Tpmax) the score is 1. On the cold side it follows a smooth S-shape down to 0 at the minimum temperature Tmin: a little colder than preferred costs little, close to the limit the score drops quickly. On the warm side the score stays high just above Tpmax and then falls ever faster, reaching 0 at Tmax. This is the typical shape of a fish's thermal performance curve: a slow rise with warming, then a sharp collapse near the upper limit (Deutsch et al. 2008; Ern et al. 2023).
s = 3f2 − 2f3, f = (T − Tmin) / (Tpmin − Tmin) (cold side)
s = 1 − g2, g = (T − Tpmax) / (Tmax − Tpmax) (warm side)
B. Salinity: trapezoid
1 inside the preferred range, falling in a straight line to 0 at the minimum and maximum salinity. Salinity mainly decides where a species can live (for example how far into the brackish Baltic it reaches), so the simple AquaMaps shape is kept.
C. Oxygen: % air saturation, lower limit only
Bottom oxygen is expressed as % of the oxygen the water would hold in equilibrium with air at its own temperature and salinity (Garcia & Gordon 1992). Below 20% (Ocrit, the level at which fish can no longer sustain their resting metabolism; Claireaux & Chabot 2016) the score is 0. Above it the score rises quickly at first and then levels off, reaching 1 at the species' preferred minimum (Opref, converted to % saturation at 10 °C / 30 psu, at most 100%). There is no upper limit: more oxygen never makes water less suitable. Bottom oxygen is applied to the whole water column, which is conservative.
s = 1 − (1 − f)2, f = (O − Ocrit) / (Opref − Ocrit)
D. Depth: in or out
Depth is a yes/no filter: a depth level (and the seafloor of the cell) deeper than the species' maximum depth zmax is excluded; everything shallower counts fully. There is no reason a fish should be "half suitable" at 300 m if it is regularly found at 600 m.
From four curves to one score. The temperature, salinity and oxygen scores are combined as their geometric mean, then multiplied by the depth filter: S = (sT · sS · sO)1/3 × depth (0 or 1). A geometric mean means one poor condition pulls the score down strongly, and a 0 in any variable makes the place unsuitable.
Where in the water column. Fish that move through the water column get the score of their best depth level in each cell and month. Seafloor species (17 of the 70: crabs, lobster, brown shrimp, flatfish) are scored with the conditions at the seafloor only. The maps show monthly scores; seasonal values are means over December-March (winter) and June-September (summer).
What the model does not include. The ranges describe where species have been found (their realised niche), not laboratory physiology. Food, seabed type, fishing and species interactions are not part of the score.
References: Claireaux G, Chabot D (2016) J Fish Biol 88: 232-251, doi:10.1111/jfb.12833 · Deutsch CA et al. (2008) PNAS 105: 6668-6672, doi:10.1073/pnas.0709472105 · Ern R et al. (2023) Physiology, doi:10.1152/physiol.00027.2022 · Garcia HE, Gordon LI (1992) Limnol Oceanogr 37: 1307-1312, doi:10.4319/lo.1992.37.6.1307.