Twenty-first-century climate change impacts on marine animal biomass and ecosystem structure across ocean basins

2.1 Data sources

Historical (1970–2005) and future (2006–2100) projections of unfished global marine animal biomass (total animal biomass, biomass >10 cm, biomass 10–30 cm, and biomass >30 cm; vertebrates and invertebrates of trophic level >1, except for zooplankton) under different climate change scenarios were extracted from the Fisheries and Marine Ecosystem Model Intercomparison Project (Fish-MIP v1.0; Tittensor, Eddy, et al., 2018; Data access: http://doi.org/10.5880/PIK.2018.005). The projections included outputs from six different global marine ecosystem models (Supporting Information Table S1): BOATS (Carozza et al., 2016; Carozza, Bianchi, & Galbraith, 2017), Macroecological (Jennings & Collingridge, 2015), DPBM (Blanchard et al., 2012), DBEM (Cheung et al., 2011), EcoOcean (Christensen et al., 2015), and APECOSM (Maury, 2010). Each marine ecosystem model was forced with standardized output from two Earth system models (ESMs; Supporting Information Table S2) and greenhouse gas emissions scenarios (Representative Concentration Pathways, RCPs) following the Fish-MIP simulation protocol (Tittensor, Eddy, et al., 2018). ESM outputs were derived from the CMIP5 database (https://esgf-node.llnl.gov/search/cmip5/) and bracketed a wide range of projected climate system changes, with GFDL-ESM2M representing moderate and IPSL-CM5A-LR strong changes in, for example, sea surface temperature (SST) and oceanic net primary productivity (NPP) (Bopp et al., 2013). ESM outputs were postprocessed to provide forcing inputs at the temporal and spatial resolution required by the ecosystem models (typically 1-degree spatial resolution and 1-month or 1-year temporal resolution, and vertically integrated or vertically specific variables; Supporting Information Tables S1 and S2). Which specific ESM output variables were used and how each was implemented depended on the respective ecosystem model. For example, DBEM used SST, NPP, zooplankton carbon concentration, current speed, dissolved oxygen, pH, and salinity to model changes in species’ habitat suitability (Supporting Information Tables S1 and S2). In contrast, the Macroecological model used changes in NPP and water temperature to model changes in production of size-structured pelagic communities. Specific details for each ecosystem model, including the spatial, vertical, and temporal resolution of forcing variables are given in Supporting Information Tables S1 and S2.

For this study, we selected two contrasting emissions scenarios: RCP2.6 characterizes a low emissions or high climate change mitigation scenario, assuming that greenhouse gas emissions peak at 2010–2020 and decline substantially until 2100 (van Vuuren et al., 2011); RCP8.5 characterizes a high emissions pathway assuming a continuous increase in emissions until 2100, while not including specific climate change mitigation targets (Riahi et al., 2011).

As projections including fishing impacts were only available for three marine ecosystem models, and spatially explicit future fisheries projections are as yet unavailable, we chose to focus on runs under a no-fishing scenario, thus isolating climate change effects on marine animal biomass (Tittensor, Eddy, et al., 2018).

2.2 Data analysis

Projected time series for historical and future marine animal biomass (g C m⁻²) for BOATS, Macroecological, DPBM, DBEM, EcoOcean, and APECOSM were extracted on a 1 × 1 degree spatial grid for seven ocean basins: North Atlantic Ocean, South Atlantic Ocean, North Pacific Ocean, South Pacific Ocean, Indian Ocean, Arctic Ocean, and Southern Ocean (Figure 1). The forcing variables of SST (°C) and net primary production (NPP; mol m⁻³ s⁻¹) from GFDL-ESM2M and IPSL-CM5A-LR were extracted for the same ocean basins over the same timescales. For the ocean basin data subsetting, we selected each grid cell centroid located within the respective ocean basin boundaries using arcmap 10.5 (ESRI, Redlands, & USA, 2017), and combined the individual cells into an ocean basin annual mean using the statistical software r 3.4.3 (R Core Team, 2017).

2.3 Temporal changes in marine animal biomass, SST, and NPP

The marine ecosystem models included in this study account for different species, size-classes, or trophic groups of marine animals (Supporting Information Table S1; Tittensor, Eddy, et al., 2018). Hence, for each ocean basin and individual marine ecosystem model–ESM combination we calculated proportional biomass change time series by deriving annual mean changes in total marine animal biomass relative to the average of the 1990–1999 (as a historical reference period). These individual time series of relative change were then averaged to derive an ensemble mean change. We also calculated proportional biomass changes for each ocean basin in the 2090s relative to the 1990s. A similar approach was used for SST and NPP forcing data. As our measure of variability around the ensemble mean of marine animal biomass projections, we used a one intermodel standard deviation.

2.4 Model agreement in projected biomass changes

Model agreement in projected total biomass changes was assessed for the complete ensemble of all ecosystem model–ESM combinations. As measures of model agreement, we used a robustness index (ensemble mean/intermodel standard deviation; Bopp et al., 2013) as well as the percent model agreement in the direction (increase or decrease) of projected changes in the 2090s relative to 1990s. A robustness index >1 indicated high robustness (SD < mean) and an index <1 low robustness (SD > mean) in marine animal biomass projections across ecosystem models (Bopp et al., 2013). For the percent model agreement, 80%–100% of models agreeing on the direction of change in marine animal biomass change was assumed to represent high agreement in the ensemble projections (Bopp et al., 2013).

2.5 Sources of variability in ensemble projections

We compared the relative variability or intermodel spread in projected total marine animal biomass changes due to variability in (a) the different ESMs and (b) the different marine ecosystem models under the low and high emissions scenarios (RCP2.6 and RCP8.5). For (a), we calculated the mean standard deviation between individual ecosystem model results forced by GFDL-ESM2M and IPSL-CM5A-LR (n = 4: for marine ecosystem models forced by both ESMs). Next, we calculated the mean standard deviation across ecosystem models to derive the mean variability in our ensemble projections due to ESMs. For (b), we calculated the intermodel standard deviation of marine ecosystem models for GFDL-ESM2M and IPSL-CM5A-LR separately. Then, we calculated the mean of the standard deviations from both ESMs for each ocean basin to derive the variability due to marine ecosystem models.

2.6 Changes in animal size structure

To examine climate change impacts on ecosystem structure, we analyzed differences in climate change impacts under RCP2.6 and RCP8.5 on the biomass of marine animals in small (0–10 cm), medium (10–30 cm), and large (>30 cm) size-classes. Because DBEM did not distinguish between size-classes, this model was excluded from this analysis. Ecosystem models that modeled different biomass size-classes account for growth and movement between the size-classes (BOATS: Carozza et al. (2016); EcoOcean: Christensen et al. (2015); DPBM: Blanchard, Law, Castle, & Jennings, 2011; APECOSM: Maury (2010)). The only exception being the Macroecological model, in which movements of individuals between size-classes was not considered as it is a static representation of the system (Jennings & Collingridge, 2015). Moreover, since BOATS did not have any size-classes <10 cm, we could not calculate a small size-class for this model but included it in the medium and large size-classes. Note that excluding BOATS from the small size-class data set did not alter the overall results. For each distinct size-class, we calculated the percent relative change in biomass in the 2090s relative to the 1990s for each ecosystem–ESM combination and used box plots to derive the ensemble mean, median, and intermodel variation.

2.7 Climate change mitigation effect on biomass changes

Finally, we assessed climate change mitigation effects on projected changes in total marine animal biomass for the model ensemble and individual ecosystem model-ESM means by subtracting the annual mean biomass change under RCP8.5 from RCP2.6. The obtained values represent the climate change mitigation effect in terms of the difference between the projected percentage changes in total biomass under the high mitigation scenario (RCP2.6) and the no-mitigation/high emissions scenario (RCP8.5).

3.1 Temporal changes in marine animal biomass, SST, and NPP

Our ensemble projections suggest that, in an unfished ocean and hence all impacts due entirely to changes in climate, total animal biomass in all basins except the polar oceans would be consistently lower by the end of the 21st century than at the beginning of the time series under both low (RCP2.6; Figure 2) and high (RCP8.5; Figure 3) emissions scenarios (Table 1). Under RCP2.6, ensemble projections of total animal biomass in the North Atlantic and North Pacific Ocean projected sharp declines until 2040 (North Atlantic: 13%; North Pacific: 10%) and leveled off afterward until 2100 (Figure 2). In the South Pacific, South Atlantic, and Indian Ocean, lower rates of decline in total animal biomass were projected under RCP2.6 (Figure 2). In contrast, under RCP8.5, projected changes in total animal biomass reached >20% declines in the North Atlantic and North Pacific, and 10%–20% declines in the South Atlantic, South Pacific, and Indian Ocean until 2100 relative to the 1990s (Figure 3).

In the polar ocean basins, trends in ensemble biomass projections differed. In the Arctic Ocean, projected total animal biomass increased until the 2040s under both emissions scenarios (Figures 2 and 3). In subsequent years, biomass changes stabilized under RCP2.6 (Figure 2) but started to decrease under RCP8.5 (Figure 3). Given the rate of increase until the 2040s, total animal biomass in the Arctic Ocean was projected to increase by 45% (±94% standard deviation) under RCP2.6% and 80% (±200%) under RCP8.5 in the 2090s relative to the 1990s (Table 1). While all 10 ecosystem–ESM combinations projected increases in animal biomass in the Arctic Ocean by the end of the 21st century under RCP2.6, only half did so under RCP8.5 (Supporting Information Figures S3 and S4; Supporting Information Table S3). However, the magnitude of projected biomass changes in the Arctic varied across models as indicated by the high intermodel standard deviation. In particular, DBEM projected substantially higher increases in animal biomass in the Arctic relative to the other models (Supporting Information Figure S4), while the variability of projections among the other models was smaller. In the Southern Ocean, projections of total animal biomass showed relatively high variability throughout most of the time series under both emissions scenarios; however, toward the end of the 21st century, ensemble projections indicated a 10% decline under RCP2.6 (Figure 2) and an 15% increase under RCP8.5 (Figure 3).

The temporal trends in projected total animal biomass generally corresponded to a combination of historical and future changes in NPP and SST generated by the ESMs. Under RCP2.6, SST in all basins except the polar oceans was projected to increase by ~1°C until the 2040s and level off until 2100, with total animal biomass showing a corresponding 5%–10% decline (Figure 2). NPP was projected to initially decrease by 3%–10% until 2030 and either leveled off or increased in the North Atlantic and Pacific, the South Pacific, and Indian Ocean until 2100 (Figure 2). NPP projections under RCP2.6 in the South Atlantic Ocean did not show a clear trend throughout most of the 21st century; however, increased by 1%–2% toward the end of the 21st century (Figure 2).

Under RCP8.5, ensemble projections of total animal biomass declined continuously until 2100 in the Atlantic Ocean (North Atlantic: 29%; South Atlantic: 13%), Pacific Ocean (North Pacific: 25%; South Pacific: 18.5%), and Indian Ocean (19%). Over the same period, SST was projected to continuously increase and NPP to continuously decrease, except in the South Atlantic for the latter (North Atlantic: +3.5°C and 13% decline in NPP; North Pacific: +4.1°C and 9% decline in NPP; South Pacific: +3.2°C and 5% decline in NPP; Indian Ocean: +3.3°C and 6% decline in NPP; Figure 3). In the South Atlantic Ocean, total animal biomass was projected to decline by more than 10% until the end of the 21st century with no substantial concurrent decline in NPP (0.36%) yet an SST increase of +3.4°C (Figure 3).

In the Arctic Ocean under RCP2.6, SST was projected to increase by 0.5°C by the 2030s and level off until 2100. NPP projections were relatively variable interannually (Figure 2 and Supporting Information Figure S2) but correlated with the projected SST changes by the end of the 21st century (Figure 2; Supporting Information Figure S1). Under RCP8.5, projections of SST continuously increased up to 2°C by 2100 (Figure 3). NPP showed a projected 25% increase until the 2040s with a 6% decrease thereafter, which correlated with the projected trend in total animal biomass changes (Figure 3). In the Southern Ocean, projected trends in SST and NPP under RCP2.6 were highly variable with no evidence for an underlying trend. This was reflected in the projected trends in total animal biomass (Figure 2). Patterns in projected SST, NPP, and total animal biomass were similar under RCP8.5 (Figure 3).

3.2 Model agreement in projected biomass changes

Our metrics of model agreement within the model ensemble revealed high robustness (>1) and high percentage model agreement in the direction of projected biomass changes (>80%–100%) in all basins except the polar basins under both emissions scenarios (Table 1). For both RCP2.6 and RCP8.5, the highest robustness index of >2 was found in the North Atlantic Ocean (Table 1). In contrast, projections in the polar ocean basins under both emissions scenarios had low robustness indices (<1). Model agreement in the direction of change was high in both polar oceans under RCP2.6, but only in the Southern Ocean under RCP8.5 (Table 1). In the Arctic Ocean, under RCP8.5 only 50% of the included ecosystem models agreed on the direction of change in projected total animal biomass (Table 1 and Supporting Information Table S3; Supporting Information Figures S3 and S4).

Model spread, represented as one intermodel standard deviation of the ensemble mean, was lower under RCP2.6 than RCP8.5 across all ocean basins (Table 1; Supporting Information Figure S3). Model spread ranged from ±3% to 6% under RCP2.6 across all basins except for the Arctic Ocean with ±94% (Table 1). Under RCP8.5, model spread was higher, ranging from ±12% to 17% in all ocean basins except for ±35% for the Southern Ocean and ±200% for the Arctic Ocean (Table 1).

3.3 Sources of variability in ensemble projections

Projections forced by GFDL-ESM2M and IPSL-CM5A-LR differed between most ocean basins (i.e., North Atlantic, North Pacific, Southern Ocean, Arctic Ocean; Supporting Information Figure S3). However, the mean variability in total marine animal biomass projections under both emissions scenarios due to the ESMs was of similar magnitude to the mean variability due to the marine ecosystem models across all ocean basins (RCP2.6: 2%–7%; RCP8.5: 4%–10%) except in the Arctic Ocean (Supporting Information Figure S5). In the Arctic Ocean, mean variability of total animal biomass projections due to the marine ecosystem models was ~40% greater under RCP2.6 and ~70% greater under RCP8.5 than the mean variability due to the different ESMs (Supporting Information Figure S5).

3.4 Changes in animal size structure

Our analysis of relative changes in the projected biomass of animals in different size-classes (large-sized animals: >30 cm; medium-sized animals: 10–30 cm; and small-sized animals <10 cm) showed that projected biomass in all size-classes decreased in the Pacific, Atlantic, and Indian Ocean basins under RCP8.5 by the end of the 21st century (Figure 4). In the North and South Atlantic Ocean, a greater decrease in the mean biomass of medium-sized animals (North Atlantic: 29%; South Atlantic: 17%) was projected compared to small animals (North Atlantic: 24.5%; South Atlantic: 10%) and large animals (North Atlantic: 24%; South Atlantic: 12%) (Figure 4). The reverse was observed in the North Pacific Ocean, with a mean projected biomass decrease of 13.5% in medium-sized animals, while biomass of large and small animals decreased by 21% and 23%, respectively (Figure 4). The overall trends in the South Pacific Ocean did not change substantially across size-classes (small-sized animals: 13%; medium-sized animals: 12%; and large-sized animals: 12%; Figure 4). Similarly, in the Indian Ocean, projected trends in biomass under RCP8.5 did not differ substantially among size-classes (small-sized animals: 13.5%; medium-sized animals: 12%; large-sized animals: 14%; Figure 4). In contrast, biomass of animals in all size-classes in the Southern Ocean was projected to increase by the end of the 21st century, with mean biomass of large animals projected to increase by ~10%, medium-sized animals by 5%, and small animals by 3% (Figure 4). In the Arctic Ocean, only biomass of large animals was projected to increase (by 5%), while mean biomass of medium-sized animals decreased by 10% and biomass of small animals by 5% (Figure 4).

Under RCP2.6, biomass of all three size-classes in the Arctic Ocean was projected to increase, ranging from 15% for large animals to 5%–7% for the medium and small size-classes, while only large animal biomass was projected to increase (2%) in the Southern Ocean (Supporting Information Figure S6). The trends in projected biomass in different size-classes in the Pacific, Atlantic, and Indian Ocean basins under RCP2.6 were similar in direction but smaller in magnitude than trends under RCP8.5 (Figure 4 and Supporting Information Figure S6).

3.5 Climate change mitigation effect on biomass changes

The climate change mitigation effect on projected ensemble mean changes in total marine animal biomass was minor until 2050 in all ocean basins (mean mitigation effect over 2006–2049: +0.02%, ±1.87%; Figure 5) except for the Arctic Ocean, where climate change mitigation was projected to lead to greater increases in mean animal biomass under the high mitigation scenario (RCP2.6) than under the no-mitigation scenario (RCP8.5) until 2050 (mean mitigation effect over 2006–2049: +10.5%, ±7.4%; Figure 5). After 2050, climate change mitigation was projected to have a positive effect on biomass changes in most ocean basins, meaning climate change mitigation would dampen projected climate change-induced biomass decreases (mean mitigation effect over 2050–2100: North Atlantic: 13%, ±5%; South Atlantic: 5%, ±2%; North Pacific: 10%, ±4%; South Pacific: 7%, ±3%; Indian Ocean: 9%, ±4%; Figure 5). However, for the Arctic and Southern Ocean, climate change mitigation reduced the projected biomass increase toward the end of the 21st century (mean mitigation effect over 2050–2100: Arctic Ocean: −28%, ±13%; Southern Ocean: −7%, ±6%; Figure 5).

Climate change mitigation effects on total animal biomass projections from 2050 to 2100 differed notably in magnitude between individual ecosystem models in all ocean basins except in the North Atlantic and South Pacific Ocean (Supporting Information Figure S7). BOATS, Macroecological, and DBEM showed the largest climate change mitigation effects from 2050 to 2100 in all ocean basins except polar basins, ranging from 9%–13% (±3%–5%) for BOATs to 6%–20% (±4%–8%) for the Macroecological model, and 8%–17% (±3%–7%) for DBEM. In comparison, mean climate change mitigation effects from 2050 to 2100 reached 4%–5% (±2%) for APECOSM, 3%–4% (±1%–2%) for DPBM, and 2%–9% (±1%–3%) for EcoOcean. In the Arctic and Southern Ocean, climate change mitigation effects from 2050 to 2100 differed in magnitude and trend compared to the other basins (Supporting Information Figure S7). Notably, in the Arctic Ocean all models, except for DBEM, showed a mean positive climate change mitigation effect ranging from 4% to 10% (± 4%–7%), while that for DBEM was −172% (±54%). In the Southern Ocean, most models showed a negative climate change mitigation effect from 2050 to 2100, with DBEM showing a larger mitigation effect of −28% (±23%) than the other models (APECOSM −3% (±4%), BOATS −1.5% (±2), DPBM 0.2% (±2.5%), Macroecological model −3% (±4%), and EcoOcean −4% (±5%)).

4.1 Ensemble projections in different ocean basins

In the North Atlantic and North Pacific, projected total marine animal biomass declined less under the strong mitigation (RCP2.6, 10% decline) than the high emissions scenario (RCP8.5, 20% decline), in line with the lower magnitude of projected changes in SST (RCP2.6: ~1°C increase; RCP8.5:3–4°C increase) and NPP (RCP2.6: 3%–5% decrease; RCP8.5: 8%–13% decrease). In the South Atlantic, South Pacific, and Indian Ocean, projected biomass declines were similar under RCP8.5 (~20%), yet reached only ~5% under RCP2.6 by the mid-21st century and leveled off afterward. In these three ocean basins, projected NPP decreased less than in the North Atlantic and North Pacific, which is primarily due to differences in stratification and nutrient supply regimes in the Earth system models (ESMs) used to force the marine ecosystem models in this study (Capotondi, Alexander, Bond, Curchitser, & Scott, 2012; Doney, 2010; Dufresne et al., 2013; Dunne et al., 2012). Thus, the differences in projected biomass declines between the two emissions scenarios can be partially explained by differences in environmental drivers the modeled marine organisms experience in the simulated future ocean, such as effects on the physiology of marine organisms (e.g., metabolic rates, growth, survival, and trophic interactions) and availability of habitat (Cheung et al., 2009; Fu, Randerson, & Keith Moore, 2016; Worm & Lotze, 2016). If a habitat becomes unsuitable for a given population, for example, due to thermal stress, population size may decline as ecophysiological performance is negatively affected or species may shift their distribution to cooler waters (Cheung et al., 2013; Pörtner, 2001; Pörtner & Knust, 2007). These effects play out differently in the different ecosystem models due to their varying structures and characterization of processes (Tittensor, Eddy, et al., 2018), thus influencing the projected biomass trends. For example, projections by the species distribution model DBEM are strongly affected by changes in the availability of suitable habitat due to shifting temperature fields, ice cover, and primary production. In comparison, biomass dynamics in the size-structured models are driven by size-dependent processes such as production and energy transfer (Macroecological, BOATS) or detailed size-dependent feeding processes, growth, and mortality (DPBM) and size-dependent movement (APECOSM), which are all affected by changes in environmental forcing variables (Supporting Information Table S2).

Ensemble projections in the Arctic suggested a 60% increase in total animal biomass until the mid-21st century followed by a stabilization under RCP2.6 and a decrease of 80% under RCP8.5 toward the end of the 21st century. In the Southern Ocean, conversely, ensemble projections showed only a slight biomass decrease (~10%) toward the end of the 21st century under RCP2.6 yet a continuous increase to ~15% under RCP8.5. The projected biomass increases in polar oceans until the mid-21st century can be attributed to processes such as immigrating marine animals from warmer waters as new habitats become available (Cheung et al., 2009), increasing water temperatures and primary production enhancing growth and survival (Frainer et al., 2017), and longer growing seasons influencing phenology (Racault, Le Quéré, Buitenhuis, Sathyendranath, & Platt, 2012). In our ecosystem model ensemble, the magnitude of the mean biomass increases in both the Arctic and Southern Ocean was primarily influenced by DBEM, which models species-specific habitats for commercial fish and invertebrates (Cheung et al., 2011). In the 1990s (our historical reference period), DBEM has only a few commercial species with relatively low biomass levels in the Arctic and Antarctic; thus, any newly invading commercial species and increasing growth result in large proportional changes in biomass. Thus, these results can be partly explained by the specific focus of this model. In contrast, all other ecosystem models (Macroecological, BOATS, DPBM, EcoOcean, and APECOSM) project bulk changes in marine animal biomass across different size-classes, functional-, and trophic groups due to changes in environmental factors affecting metabolic rates, energy transfer, and trophic relationships (see Supporting Information Table S1) and can include commercial and non-commercial species. Therefore, these models generally start with higher initial biomass in polar oceans meaning that proportional changes in the future are lower.

In the second half of the 21st century, the projected stabilization of biomass changes in the Arctic and Southern Ocean under RCP2.6 can be explained by changes in the forcing variables driven by strong climate change mitigation policies (van Vuuren et al., 2011), as indicated by the leveling off in projected SST and NPP trends (Supporting Information Figures S1 and S2). In contrast, under the high emissions scenario (RCP8.5), in which greenhouse gas emissions are projected to increase until 2100 (Riahi et al., 2011), the decline in projected total marine animal biomass in the Arctic may be attributed to continuing changes in the physical and biogeochemical environment, with consequences for the entire trophic network (Hillebrand et al., 2018). Indeed, longer term projections of changes in ocean ecosystems until 2300 suggest a strong decline in ocean productivity in the Northern Hemisphere and its shift toward the Southern Ocean (Moore et al., 2018; (Figure 3)). In the Arctic, the projected late 21st-century biomass decline under RCP8.5 was concurrent with a projected 20% decline in NPP during that period, likely attributed to enhanced stratification due to changes in water temperature and salinity with melting sea ice and permafrost (Fu et al., 2016). Large decreases in sea ice cover could also enhance light levels, leading to higher seasonal NPP (Leung, Cabre, & Marinov, 2015). The loss of sea ice can also directly affect sea ice-dependent marine animals in both the Arctic and Southern Ocean, which rely on sea ice for reproduction, feeding, or survival, ranging from krill (Antarctic krill in the Southern Ocean, Calanus copepods in the Arctic) to Arctic cod (Boreogadus saida) to many whale species, such as narwhales (Monodon monoceros) and killer whales (Orcinus orca) (Hillebrand et al., 2018; Macias-Fauria & Post, 2018; Stenson & Hammill, 2014). With krill and copepods representing a significant link between phytoplankton and higher trophic levels, sea ice loss is expected to lead to substantial modifications in the existing Arctic and Antarctic ecosystems and associated commercial and subsistence fisheries (Mcbride et al., 2014; Moore et al., 2018). However, only half of the marine ecosystem models accounted for changing ice dynamics (Supporting Information Table S2), yet these did not necessarily agree on the direction of biomass change (Supporting Information Figures S3 and S4); consequently, it is difficult to determine how much the projected biomass changes in the polar basins are due to changing ice cover and its implications for polar food webs.

4.2 Variability around ensemble projections

We used an ensemble model approach, which included six marine ecosystem models forced with two different ESMs and different RCPs to project past and future marine animal biomass under different climate change scenarios (Tittensor, Eddy, et al., 2018; Tittensor, Lotze, et al., 2018). The ensemble approach has the advantage that ecosystem models characterized by different model structure, processes, and underlying assumptions are more likely to capture relevant features in complex ocean ecosystems than any single model (Spence et al., 2017; Tittensor, Eddy, et al., 2018). The ensemble approach also allows for the ability to quantify uncertainties due to marine ecosystem models, which remains important information for policy-makers and managers but is unavailable from single model projections. Here, we used metrics including variability around the ensemble mean, a robustness index, and model agreement in the direction of change (Bopp et al., 2013; Tittensor, Eddy, et al., 2018). Comparing results of different ecosystem models can also help to understand how projections are affected by different model structures and ecological processes. Thus, ensemble projections and model intercomparison projects have emerged as an extremely useful approach in climate impact sciences (Schellnhuber et al., 2013; Spence et al., 2017; Tittensor, Eddy, et al., 2018).

High variability and uncertainty of ensemble results were detected in the Arctic and Southern Ocean. As discussed above, this may be partly due to the fact that DBEM projections of changes in habitat and associated population dynamics specifically focused on commercial fish and invertebrates (Cheung et al., 2011; Cheung, Pinnegar, Merino, Jones, & Barange, 2012), which are currently very low in abundance and may therefore lead to proportionally larger relative biomass changes in these regions due to changes in projected SST, NPP, ice cover, and other environmental variables. However, the general trends in biomass change projected in the polar oceans by DBEM did not differ from most of the other ecosystem models, suggesting broad agreement in the direction of projected changes over the coming century despite varying magnitudes. Overall, a general projected increase in total marine animal biomass in the Arctic and Southern Ocean, yet a decrease in the North and South Atlantic and Pacific and Indian Ocean may occur by the end of the 21st century under both emissions scenarios, which corresponds with the IPCC's Fifth Assessment Report and other single- and multi-model studies (Blanchard et al., 2017; Pörtner et al., 2014).

4.3 Changes in ecosystem structure

In most ocean basins, the greater projected declines in biomass of medium-sized animals may be explained by the decline of their smaller sized prey. In comparison, the reduced relative declines in the projected biomass of large animals may result from model structures and parameterizations which result in larger animals having access to a larger pool of available food sources (both medium- and small-sized animals), slower turnover times which result in lagged responses to changing ecosystem dynamics, and food competition effects due to increasing competition for small-sizes animals with the medium-sized animals (Lefort et al., 2015; Perry, 2005).

4.4 Climate change mitigation

Based on our model ensemble, climate change mitigation that reduces greenhouse gas emissions in line with RCP2.6 (van Vuuren et al., 2011) was projected to lessen the decreases in total marine animal biomass by 10%–20% compared to the high emissions scenario (RCP8.5) in all non-polar oceans, but also dampen increases in polar oceans. Thus, with successful climate change mitigation, declines in marine animal biomass in the North and South Atlantic and Pacific and the Indian Ocean could potentially be alleviated, particularly after the 2050s. This result was consistent across all ecosystem models, only differing in the magnitude of the climate change mitigation effect. Along with recent projections of the mitigation effects on the timing of emergence of climate change impacts on environmental drivers (Henson et al., 2017), our results suggest that climate change impacts on marine ecosystems can be substantially reduced by successfully implementing mitigation measures. In the Arctic and Southern Ocean, climate change mitigation also reduced projected impacts and led to lower changes in biomass, which resulted in reduced proportional biomass increases or even declines. However, the individual ecosystem models showed contrasting trends, with only DBEM projecting substantially reduced biomass increases. As discussed above, this result is likely due to different model structures and taxonomic scope. By slowing the pace of climate change and reducing impacts, climate change mitigation would provide time and opportunity for adaptation and development of proactive ocean policies, such as in the context of marine conservation efforts and fisheries management strategies (Blanchard et al., 2017; Henson et al., 2017).

4.5 Caveats and future steps

The use of outputs from two ESMs (GFDL-ESM2M and IPSL-CM5A-LR) to force the marine ecosystem models represents a relatively small range of the set of ESMs available. However, as GFDL-ESM2M represents relatively weak and IPSL-CM5A-LR relatively strong changes in sea surface warming and NPP over the 21st century, they bracket the spread of projections reasonably well (Bopp et al., 2013). Furthermore, most other ESMs in the CMIP5 database do not provide or have not stored the necessary monthly, depth-resolved outputs of different size groups of phyto- and zooplankton required by several of the global marine ecosystem models within Fish-MIP (Tittensor, Eddy, et al., 2018; Supporting Information Table S2). By choosing two ESMs representing the high and low end of projected future climate change scenarios, our projected mean future change is comparable to the overall CMIP5 ensemble mean (Bopp et al., 2013). Future studies may have the possibility of including a larger range of ESMs and their outputs through the upcoming CMIP6, which will also provide higher resolution of biogeochemical variables (Ruane et al., 2016).

Another caveat is that coastal ecosystems and upwelling areas account for a large proportion of global primary production; however, the ESMs provide limited resolution of physical and biogeochemical processes in these systems (Bonan & Doney, 2018; Holt et al., 2017). To improve projections of biomass changes in these systems, regional downscaling of global ESMs is desirable to incorporate climate and ecosystem features at a higher resolution (Holt et al., 2017).

The selected ocean basins comprise areas that range from highly productive regions, for example nutrient-rich upwelling ecosystems (Canary and Benguela Current in the Atlantic Ocean, California, and Humboldt Current in the Pacific Ocean), to low productivity regions, for example warm, nutrient-poor subtropical gyres in the Atlantic, Pacific, and Indian Ocean (Hoegh-Guldberg & Poloczanska, 2017). We acknowledge that our analysis does not account for regionalization within each ocean basin, which might mask important regional variation in marine animal biomass under global change. Future research could focus on a region-by-region scale using the Fish-MIP data to further our understanding on regionalized climate change impacts on marine life.

While our ensemble model projections and analysis of model agreement contribute information on potential future changes in marine animal biomass and the spread of uncertainty around these changes, our study represents only the beginning of a systematic collaborative marine ecosystem model evaluation and intercomparison. To comprehensively improve ecosystem models participating in Fish-MIP, future effort should focus on improving our understanding of the mechanisms that drive individual model responses to forcing variables, such as by separating the forcing temperature and NPP (Carozza et al. in press), evaluating uncertainty within and across models, and attempting to refine individual model predictions under climate change.

4.6 Implications and conclusions

At present, trends in greenhouse gas emissions are consistent with those assumed in the high emissions scenario (RCP8.5; Peters et al., 2012), under which total marine animal biomass was projected to decline by at least 10%–20% in all but the polar ocean basins, where projected biomass increased by at least 15%–80% over the 21st century. Such changes would have socio-economic and food security impacts on regional and global scales (Blanchard et al., 2017; Pörtner et al., 2014). However, we have also demonstrated the level of these changes can be greatly reduced through climate mitigation efforts—such as adopting policies on national and global scales that reduce the sources and enhance the sinks of long-lived anthropogenic greenhouse gases (Bonan & Doney, 2018) and moving toward meeting international climate mitigation agreements, such as the Paris Agreement within the United Nations Framework Convention on Climate Change.