Progress in Oceanography 198 (2021) 102659 Contents lists available at ScienceDirect Progress in Oceanography journal homepage: www.elsevier.com/locate/pocean Disentangling diverse responses to climate change among global marine ecosystem models Ryan F. Heneghan a, b, *, Eric Galbraith a, c, Julia L. Blanchard d, Cheryl Harrison e, Nicolas Barrier f, Catherine Bulman g, William Cheung h, Marta Coll i, Tyler D. Eddy j, Maite Erauskin-Extramiana k, Jason D. Everett l, m, n, Jose A. Fernandes-Salvador k, Didier Gascuel o, Jerome Guiet p, Olivier Maury f, Juliano Palacios-Abrantes q, Colleen M. Petrik r, Hubert du Pontavice o, q, Anthony J. Richardson l, m, Jeroen Steenbeek i, Travis C. Tai q, Jan Volkholz s, Phoebe A. Woodworth-Jefcoats t, Derek P. Tittensor u a Institut de Ciència i Tecnologia Ambientals, Universitat Autònoma de Barcelona, Barcelona, Spain b School of Mathematical Sciences, Queensland University of Technology, Brisbane, QLD, Australia c Department of Earth and Planetary Sciences, McGill University, Montreal, QC, Canada d Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, Tas, Australia e School of Earth Environmental and Marine Science, University of Texas Rio Grande Valley, Port Isabel, TX, USA f MARBEC, IRD, Univ. Montpellier, Ifremer, CNRS, Sète, France g CSIRO Oceans and Atmosphere, Castray Esplanade, Hobart, Tas 7001, Australia h Nippon Foundation-Nereus Program, Institute for the Oceans and Fisheries, The University of British Columbia, Vancouver, BC V6 T 1Z4, Canada i Institute of Marie Science (ICM-CSIC), Passeig Marítim de la Barceloneta, n◦ 37-49 and Ecopath International Initiative Research Association, 08003, Barcelona, Spain j Centre for Fisheries Ecosystems Research, Fisheries & Marine Institute, Memorial University of Newfoundland, St. John’s, NL, Canada k AZTI, Marine Research, Basque Research and Technology Alliance (BRTA), Herrera Kaia, Portualdea z/g, 20110 Pasaia, San Sebastián, Spain l CSIRO Oceans and Atmosphere, Queensland BioSciences Precinct (QBP), St Lucia, Qld 4067, Australia m Centre for Applications in Natural Resource Mathematics, School of Mathematics and Physics, The University of Queensland, St Lucia, Qld 4072, Australia n Centre for Marine Science and Innovation, School of Biological, Earth and Environmental Science, The University of New South Wales, Sydney, Australia o ESE, Ecology and Ecosystem Health, Institut Agro, INRAE, Rennes, France p Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, CA, USA q The Institute for the Oceans and Fisheries, University of British Columbia, Vancouver, BC, Canada r Department of Oceanography, Texas A&M University, College Station, TX, USA s Potsdam Institute for Climate Impact Research, Potsdam, Germany t Pacific Islands Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Atmospheric Administration, Honolulu, HI, USA u Department of Biology, Dalhousie University, Halifax, NS, Canada A R T I C L E I N F O A B S T R A C T Keywords: Climate change is warming the ocean and impacting lower trophic level (LTL) organisms. Marine ecosystem Climatic change models can provide estimates of how these changes will propagate to larger animals and impact societal services Modelling such as fisheries, but at present these estimates vary widely. A better understanding of what drives this inter- Fishery oceanography model variation will improve our ability to project fisheries and other ecosystem services into the future, Marine ecology FishMIP while also helping to identify uncertainties in process understanding. Here, we explore the mechanisms that Structural uncertainty underlie the diversity of responses to changes in temperature and LTLs in eight global marine ecosystem models from the Fisheries and Marine Ecosystem Model Intercomparison Project (FishMIP). Temperature and LTL im­ pacts on total consumer biomass and ecosystem structure (defined as the relative change of small and large organism biomass) were isolated using a comparative experimental protocol. Total model biomass varied be­ tween − 35% to +3% in response to warming, and -17% to +15% in response to LTL changes. There was little consensus about the spatial redistribution of biomass or changes in the balance between small and large or­ ganisms (ecosystem structure) in response to warming, an LTL impacts on total consumer biomass varied depending on the choice of LTL forcing terms. Overall, climate change impacts on consumer biomass and ecosystem structure are well approximated by the sum of temperature and LTL impacts, indicating an absence of * Corresponding author at: School of Mathematical Sciences, Queensland University of Technology, Brisbane, QLD, Australia. E-mail address: ryan.heneghan@gmail.com (R.F. Heneghan). https://doi.org/10.1016/j.pocean.2021.102659 Received 14 January 2021; Received in revised form 19 July 2021; Accepted 3 August 2021 Available online 9 August 2021 0079-6611/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 nonlinear interaction between the models’ drivers. Our results highlight a lack of theoretical clarity about how to represent fundamental ecological mechanisms, most importantly how temperature impacts scale from individual to ecosystem level, and the need to better understand the two-way coupling between LTL organisms and con­ sumers. We finish by identifying future research needs to strengthen global marine ecosystem modelling and improve projections of climate change impacts. 1. Introduction across global marine models is greater than climate scenario uncer­ tainty, which is problematic for the goal of using these models to provide Water temperature and primary production play critical roles in assessments of climate impacts on marine ecosystems and the societal marine processes. Higher temperatures accelerate reaction rates, with services they provide. consequences ranging from the molecular to ecosystem scale, while For all global models in the FishMIP ensemble, temperature and primary production provides the fundamental source of energy for lower trophic level (LTL) forcings such as net primary production, almost all marine life (Brown et al., 2004; Chavez et al., 2011). Climate phytoplankton and zooplankton biomass, and export carbon are the two change impacts on both water temperature and primary production will main drivers of projected climate change impacts (Tittensor et al., thus alter marine ecosystems in fundamental ways (Pörtner et al., 2014). 2018), yet their implementations vary. Although there is some agree­ For example, a first-order expectation of these impacts is that acceler­ ment on how temperature impacts physiological processes in general (e. ated metabolic rates will consume energy more quickly in a warmer g. Kooijman, 2010), there is less agreement on how these impacts vary ocean, all else being equal, so that less biomass could be supported by a across functional groups, body sizes, and different processes such as given level of primary production (Heneghan et al., 2019). Yet, growth and metabolism (van Denderen et al., 2020). Similarly, although ecosystem-level effects emerge from individual-level processes and in­ it is universally understood that LTL biomass and production provide the teractions, which could lead to nonlinear effects and changes in source of energy that supports higher trophic levels, there is less un­ ecosystem structure, while shifting thermal habitats may influence the derstanding about how the physiology and structure of LTLs affects distribution of species, transforming food-webs to previously unknown transfer efficiency and ecosystem structure, and how to couple lower states (Coll et al., 2020; Pinsky et al., 2020; Poloczanska et al., 2016). and higher trophic levels (Eddy et al., 2020; Heneghan et al., 2016; There is a growing need to quantify and project climate change im­ Stock et al., 2017). Previous multi-model ensemble studies have pacts on marine ecosystems to motivate mitigation (Bryndum-Buchholz explored structural model uncertainty in projections of consumer et al., 2020), provide insight into potential future threats to food security biomass and species distribution shifts under climate change (e.g. Jones (Barange et al., 2014; Blanchard et al., 2017a; Boyce et al., 2020), and et al., 2012; Woodworth-Jefcoats et al., 2015), but these studies did not identify needs for biodiversity conservation (Brito-Morales et al., 2020; disentangle the effects of temperature and lower trophic level (LTL) Waldron et al., 2020). Thus, there has been a recent proliferation of changes, a strategy that can provide mechanistic insight on underlying spatially-explicit marine ecosystem models that simulate higher trophic processes (Carozza et al., 2018). level biomass and ecosystem structure at regional and global scales, Here, we identify sources of structural uncertainty in marine driven by output from climate-ocean-biogeochemical models (Tittensor ecosystem models, by disentangling the effects of temperature and LTL et al., 2018). These ecosystem models differ significantly in their design, changes on model projections using eight global models from the Fish­ level of complexity and implementation, reflecting different choices for MIP ensemble. We first summarise how temperature and LTL processes how to represent fundamental marine ecosystem processes, as well as a are incorporated in these models, highlighting common representations diversity of model purpose and scope. As a result, there is considerable and differences across the ensemble. We then isolate the impact of uncertainty in model projections of climate change impacts on higher changes in temperature and LTL processes on consumer biomass and trophic levels (e.g. Lotze et al., 2019), with projections from each model ecosystem structure (which we define as the relative change in small dependent upon decisions around the inclusion or simplification of <30 cm and large ≥30 cm consumer biomass) in a simulation protocol many candidate processes. Structural diversity in model projections is a involving a combination of pre-industrial, historical and RCP 8.5 forc­ strength for gaining a rich view of possible outcomes, given that each ings. By illuminating key sources of structural uncertainty in marine model reflects a different subset of established physiological and process model projections, we identify critical areas of future research necessary knowledge, implemented using different mathematical representations to improve not only climate impact projections but also our under­ (Knutti, 2010; Brander et al., 2013; Lefevre et al., 2017; Payne et al., standing of the marine ecosystem. 2016). At the same time, this diversity reflects fundamental uncertainty in our understanding of ecosystem processes. Thus, identifying sources 2. Methods of structural uncertainty in ensemble projections can point to critical weaknesses and thereby accelerate model improvement. We used projections from eight marine ecosystem models from the The Fisheries and Marine Ecosystem Model Intercomparison Project Fisheries and marine ecosystem Model Intercomparison Project (Fish­ (FishMIP) was created to explore this uncertainty and provide more MIP, www.fishmip.org; Tittensor et al., 2018). There are several model robust assessments of climate impacts on marine ecosystems through the types (see Tables 1 and 2 for a summary of each model and key refer­ analysis of multi-model ensembles (Tittensor et al., 2018). A recent ences). First, models that draw on the strongly size-structured nature of FishMIP study (Lotze et al., 2019) found that projections of mean marine ecosystem processes to represent the ecosystem purely by body changes in animal biomass from a model ensemble typically compared size (BOATS, Macroecological) or trophic level (EcoTroph). Second, better with empirical data than individual models, emphasising the trait-based size-structured models (APECOSM, DBPM, FEISTY, benefits of ensemble climate impact projections. However, uncertainty ZooMSS), which move beyond a purely size-based representation to in ensemble projections of higher trophic level biomass is significant: include different communities and groups using functional traits other Lotze et al. (2019) found that the spread of changes across the FishMIP than body size. Last, DBEM is a habitat suitability-based species-distri­ ensemble in 21st century marine consumer biomass under the high bution model that resolves the biomass and spatial distribution of emissions, representative concentration pathway 8.5 (RCP 8.5) climate >1200 fish and invertebrate species using observational data, and in­ change scenario (0 to − 35%) was larger than the multi-model mean cludes other mechanisms such as species ecophysiology and dispersal. consumer biomass change between the RCP 2.6 (low emissions) and RCP There is large variation in the structural complexity of the models, and a 8.5 scenarios (-5% to − 20%). This means that structural uncertainty detailed description of how each model incorporates temperature and 2 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 lower trophic level (LTL) impacts, including relevant equations and production remains constant. The BOATS model uses a similar frame­ temperature parameters, can be found in the Supplementary Informa­ work to Macroecological and EcoTroph to determine maximum sup­ tion S2. Here we summarise the key similarities and differences of each ported biomass at each body size class. However, in BOATS individual model as they pertain to temperature, LTLs and other drivers in Sections mortality is resolved separately and the growth of individuals from one 2.1–2.3 and Tables 1 and 2. We then explain the experimental protocol size class to the next is explicitly resolved. As temperatures rise, indi­ and model outputs in Section 2.4. vidual growth rates in BOATS increase, increasing the speed of biomass flow from small to large size classes, but also increasing mortality and reducing the maximum biomass that can be supported at each body size. 2.1. How do models incorporate temperature impacts? Taken together, these processes mean that warming causes total biomass to decrease in BOATS. Across all models, individuals gain mass through anabolic processes Within BOATS, Macroecological and EcoTroph, ecological in­ such as food uptake and assimilation, while they lose mass through teractions such as predator-prey encounters or predator-predator catabolic processes such as respiration. Populations can also gain in­ competition are not explicitly resolved. Thus, temperature and LTL dividuals through reproduction, and lose individuals through mortality drivers do not explicitly change interactions among individuals. How­ (Table 2). These processes are all influenced by temperature. As a result, ever, in BOATS and Macroecological, all primary producers are repre­ changes in ecosystem structure depend on how models resolve: (i) sented by a single body size, which is inversely related to temperature; temperature effects on individual anabolic and catabolic processes as temperature increases, the single representative body size of primary across different functional groups, body sizes or trophic levels; and (ii) producers decreases according to an empirical equation. This in turn how these variations drive changes in ecological interactions (Table 2). decreases the production of higher trophic level organisms, as the Temperature effects on these processes are represented in all models as number of trophic steps that net primary production must be transferred an exponential scaling, with parameters varying widely between models through to reach any given body size increases. However, since trophic (Supplementary Information S2). However, within models the same transfer efficiency in these models is not temperature-dependent, an temperature scaling parameters are used across all functional groups increase in the number of trophic levels is not expected to change the and ecosystem components, excluding EcoTroph, which uses different ratio of small and large organism biomass. In contrast, transfer efficiency scalings depending on the ecosystem’s biome. decreases with warming in Ecotroph. This means that warmer waters in The representation of anabolic and catabolic processes varies across Ecotroph will support relatively less biomass at high trophic levels (large models (Table 2). Macroecological and EcoTroph have the simplest body sizes) than what they will at low trophic levels (small body sizes). representations, with individual mass changes resolved implicitly in For trait-based models (APECOSM, DBPM, FEISTY and ZooMSS), each model by a single individual metabolic rate that scales with tem­ individual growth is fuelled by ingesting smaller organisms, with indi­ perature and body size (for Macroecological) or trophic level (for Eco­ vidual ingestion rates scaling with temperature and body size. For Troph). For these models, total biomass at a given body size/trophic APECOSM, DBPM and FEISTY this scaling is also modulated with the level is determined by the metabolic carrying capacity of that size/tro­ density of prey. Thus, food uptake for individuals at one size is fuelled by phic level, divided by the metabolic rate of individuals. In these two predation of smaller size individuals, and in some cases predators can models, individual metabolic rates increase with temperature while compete with each other for the same prey. These models also include total metabolic carrying capacity at a given body size/trophic level is other sources of mortality (destruction of population biomass). APE­ determined by net primary production. Thus, as warming drives an in­ COSM, DBPM and ZooMSS incorporate at least one size-dependent crease in individual metabolism, total biomass decreases even if primary Table 1 Summary of temperature, lower trophic level (LTL) and other drivers sourced from Earth system models, used by each model in the FishMIP ensemble, as well as the ecosystem representation of each model. All drivers used by the models in this experiment had a monthly temporal resolution. Model and key references Temperature drivers LTL drivers Other drivers Taxonomic scope APECOSM Maury et al. 3D water temperature 3D small and large phytoplankton, 3D oxygen concentration, 3D All epipelagic, mesopelagic and (2007a, 2007b), Maury 3D small and large zooplankton photosynthetically active radiation, migratory heterotrophic marine (2010), Maury and Poggiale biomass*, 3D export carbon flux 3D current velocities animals in the pelagic ecosystem (2013) between 15 μg and 120 kg. BOATS Carozza et al. (2016, 2D water temperature 2D depth-integrated net primary NA All commercial animal biomass from 2017) (averaged over top 75 m) production 10 g to 100 kg. DBEM Cheung et al. (2008, 2D sea surface 2D depth-integrated net primary 2D surface and bottom oxygen >1200 fish and invertebrate species. 2010, 2011, 2016) temperature production concentration, salinity and pH, sea ice, mixed layer depth, 3D current velocities DBPM Blanchard et al. (2009, 2D sea surface and 2D depth-integrated small and NA All benthic and pelagic marine 2012) bottom water large phytoplankton biomass animals, weighing between 1 mg temperature and 1 tonne. EcoTroph Gascuel and Pauly 2D sea surface 2D depth-integrated net primary NA All marine animals with trophic (2009), du Pontavice et al. temperature production level ≥ 2. (2020) FEISTY Petrik et al. (2019) 2D upper pelagic 2D depth-integrated (top 100 m) NA Forage, large pelagic and demersal (averaged over 100 m) small and large zooplankton fish, as well as benthic invertebrates, and bottom water biomass*, 2D export carbon flux to between 1 mg and 125 kg. temperature the sea floor Macroecological Jennings 2D sea surface 2D depth-integrated net primary NA All marine animals between 1 mg and Collingridge (2015) temperature production and 1 tonne. ZooMSS Heneghan et al. 2D sea surface 2D sea surface phytoplankton NA Nine zooplankton groups, from (2020) temperature biomass flagellates to jellyfish and all marine animals between 1 mg and 10 tonnes. * Where small and large zooplankton biomass are not provided by an Earth system model (as is the case with CESM1-BGC, the Earth system model used in this study) FishMIP splits total zooplankton biomass using the fraction of total phytoplankton biomass from small and large phytoplankton. 3 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 mortality term, and FEISTY includes a single natural mortality term that FEISTY also explicitly resolve size and temperature-dependent costs of is independent of body size. These additional mortality sources increase maintaining existing biomass (metabolism). In these models, as tem­ with temperature (except for senescence mortality, which increases with perature increases, maintenance costs also increase, reducing the body size, in DBPM and natural mortality, which is independent of body available energy for growth and reproduction. If maintenance costs of size, in FEISTY), causing population biomass to decrease with increasing existing biomass exceed energy intake from ingestion, biomass de­ temperature. In FEISTY, maintenance costs increase faster with both creases. As food becomes limited in APECOSM, ingestion rates scale body size and temperature than do ingestion rates. APECOSM and more slowly with temperature than maintenance costs, limiting the Table 2 Summary of temperature and lower trophic level impacts in the FishMIP model ensemble. Model Temperature effect on: Lower trophic level effect on: Individual anabolic and catabolic Ecosystem structure Individual anabolic and catabolic Ecosystem structure processes processes APECOSM Ingestion and thus predatory Growth and mortality rates increase Small and large plankton biomass is the More plankton biomass supports mortality scale with temperature with temperature. In food limited primary food source of small consumer more ecosystem biomass and and vary with predator size as well situations, this leads to less biomass, organisms. More plankton biomass reduces the trophic amplification of as the density and size of prey. especially for large organisms. In increases satiation and maximizes food limitation with size. This leads Assimilation, maintenance, and prey-rich regions, temperature does individual growth and reproduction, to biomass increase of large non-predation mortality rates also not drive biomass down but drives a thus driving increases in biomass. organisms and the presence of larger scale with temperature. faster transfer toward large sizes species in the communities. Temperature effects are stronger causing an increase in large where prey density is high. In food- organisms and a decrease of small limited areas, catabolic processes organisms due to top-down control. increase faster than anabolic processes, causing individual mass to decrease. In food-rich areas, catabolic and anabolic processes increase in the same proportion, accelerating life-cycles. BOATS Warming drives higher individual Phytoplankton size decreases with Net primary production sets the limits Phytoplankton size decreases with growth and mortality rates, which warming. Smaller phytoplankton to growth across all body size classes. decreasing production. Smaller reduces the maximum biomass that means longer food chains causing Higher production means more phytoplankton mean longer food can be supported by a given level of biomass declines for all sizes. biomass. chains causing biomass declines for primary production. all sizes. DBEM Biomass creation occurs after Catabolism increases with size faster In all regions, net primary production is Lower net primary production catabolism is deducted from than anabolism, so warming affects a key part of what sets the limits to means less consumer biomass can be anabolism. Catabolism increases large species more and drives shifts maximum biomass across all higher supported. faster with warming than in spatial distribution of species. trophic levels. anabolism. Thus, biomass decreases with warming. DBPM Ingestion-driven growth, and Natural mortality costs scale with Small and large phytoplankton biomass Relatively more small mortality rates from predation and temperature but decrease with body set the slope and intercept of the phytoplankton with less natural sources scale with size. Thus, warming increases phytoplankton size-spectrum, which is phytoplankton biomass, which temperature at the same rate. Thus, mortality relatively more for small the primary food source of small pelagic reduces food for small organisms and temperature effects largely organisms compared to large, organisms. More phytoplankton means increases food chain length. This balance, except in low food regions potentially causing their biomass to more biomass. should decrease overall biomass, where natural mortality is decrease faster. especially for larger sizes, as relatively large and causes biomass senescence increases with size. to decrease. EcoTroph Warming drives higher individual Trophic transfer efficiency Net primary production is a driver of Lower net primary production turnover rates, and lower trophic decreases with warming, causing total biomass across all trophic levels. means less biomass can be supported transfer efficiency, which means higher trophic level biomass to Higher production means more across all trophic levels. fewer individuals can be decrease more than lower trophic biomass. supported, causing biomass to level biomass. decrease. FEISTY Maintenance costs, ingestion- Maintenance costs increase faster Zooplankton is food for all small Less zooplankton biomass supports driven growth, and mortality rates than ingestion-driven growth with consumers and medium pelagic lower pelagic biomass, and more from predation scale with body size and temperature. Thus, consumers. Export production fuels small zooplankton biomass may temperature. Maintenance costs warming will reduce the scope for benthic growth. More zooplankton reduce large fish biomass due to an increase faster with warming large organism growth more than biomass and export production mean increase in the number of trophic compared to ingestion, so warming small organisms. more ecosystem biomass overall. steps between zooplankton and a reduces the scope for growth, narrower scope for growth than causing biomass to decrease. smaller sizes. Macroecological Warming drives higher individual Phytoplankton size decreases with Net primary production is a key Phytoplankton size decreases with metabolic rates, which means warming, lengthening food chains determinant of total biomass. Higher decreasing production. Smaller fewer individuals can be supported and reducing how much energy is net primary production means more phytoplankton support longer food by a given level of primary transferred to higher trophic levels. biomass. chains, thus less biomass across all production, causing total biomass sizes. to decrease. ZooMSS Ingestion-driven growth and Warming negatively impacts large The phytoplankton spectrum—set by Less phytoplankton biomass means mortality rates from predation and organisms more than small by total phytoplankton biomass—is the less food for small organisms, and senescence scale with temperature increasing senescence. If large main food of microzooplankton. More relatively more small at the same rate. Thus, temperature organism biomass declines more phytoplankton means more consumer phytoplankton. Drives shifts in effects largely balance, except than small, small biomass will biomass. zooplankton composition, which where senescence mortality is increase from reduced predation. stabilise food chain length. large, causing biomass to decrease. 4 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 scope for new growth and potentially inducing biomass to decrease as predation mortality in the plankton. maintenance costs outpace ingestion. The correlation of mean phytoplankton size with total primary pro­ In APECOSM, DBPM, FEISTY and ZooMSS, temperature affects duction is an important driver of ecosystem structure (Boyce et al., anabolic and catabolic processes differently across ecosystems, which 2015). Phytoplankton are generally larger in more productive waters has cascading effects on how the different components of ecosystems (e. (Barnes et al., 2011; Finkel et al., 2010). Given the size-structured nature g. predators and prey) interact. In APECOSM, FEISTY and ZooMSS for of the marine ecosystem (Trebilco et al., 2013), smaller phytoplankton example, the scaling of maintenance costs (in APECOSM and FEISTY) support longer food chains, which are thought to support relatively less and senescence mortality (in APECOSM and ZooMSS) with body size and consumer biomass (Eddy et al., 2020; Ryther, 1969). All models temperature mean that large organisms are more vulnerable to warming explicitly represent this phenomenon with the exception of EcoTroph compared to small organisms. Everything else being equal, a warming- and DBEM. EcoTroph uses trophic level instead of body size to represent induced decrease in large organism biomass would reduce predation the marine ecosystem. In DBEM, changes in net primary production mortality on smaller organisms, thus favouring small organisms in these affect the carrying capacity of modelled species disregarding the size of models. primary producers. In BOATS and Macroecological, changes in food Unlike what happens in the size and trait-based models, anabolic and chain length are represented by a varying representative size of phyto­ catabolic processes in DBEM are not driven explicitly by net primary plankton, the size increasing with net primary production according to production or by the ingestion of smaller organisms. Instead, individual empirical equations. In DBPM and ZooMSS, the phytoplankton size- mass increases in DBEM when anabolism exceeds catabolism, both of spectrum, which is the relationship between primary producer abun­ which are affected by temperature and other drivers (see Section 2.3). dance N and body size w, N = awb , is continuous, with the intercept a Similar to APECOSM, FEISTY and BOATS, the explicit balance between and slope b set by phytoplankton biomass. In these two models, the anabolic and catabolic processes drives an organism’s scope for plankton size-spectrum intercept is lower and the slope is steeper in less growth–if catabolism outpaces anabolism, an individual’s mass will productive waters, meaning relatively more small producers but less decline. In DBEM, anabolism accelerates more slowly with warming biomass overall. APECOSM and FEISTY use size-fractionated phyto­ compared to catabolism. Thus, as waters warm, an organism’s potential plankton and zooplankton biomass inputs from earth system models to for growth becomes increasingly limited, and their maximum size directly set the biomass of small and large phytoplankton and decreases. zooplankton groups, with a fixed size-spectrum slope assigned to each Organisms do not interact in DBEM. Rather, temperature and other LTL group in APECOSM. APECOSM and FEISTY also use export carbon forcings drive the spatial distribution of species across the ocean, with to represent detrital flux across the entire water column (in APECOSM) species’ relative abundance in a region changing with respect to tem­ or to the seafloor to fuel the growth of benthic invertebrates (in FEISTY). perature depending on their thermal preference, and the prevailing water temperature. Thus, as waters warm, ecosystem structure changes 2.3. How do models incorporate other impacts? by individual organisms becoming smaller on average, and by different species shifting their spatial boundaries to follow their thermal All models in the FishMIP ensemble are driven solely by temperature preferences. and LTL drivers, with the exception of APECOSM and DBEM (Table 1). Finally, energy transfer from small to large organisms through size- In these two models, movement of organisms between adjacent grid cells based predation is not the only way that different parts of the is resolved, so both models incorporate current speeds. Since APECOSM ecosystem interact; in APECOSM, BOATS, DBPM, DBEM and FEISTY, resolves the 3D density of animal biomass, the model also uses 3D energy moves from large to the smallest size classes through reproduc­ photosynthetically active radiation to resolve water clarity and light tion. In these models, the flux of small organism biomass entering the penetration across the water column. Thus, in APECOSM areas with the population through reproduction can increase or decrease, depending highest consumer biomass are not necessarily regions with the highest on the relative impacts of warming on large organisms. In FEISTY for LTL biomass, due to active and passive horizontal movements in example, if large organisms are more adversely affected by warming response to temperature, light, food availability and the strength of than small organisms, the reproduction rate in larger size classes would currents. Both APECOSM and DBEM also incorporate oxygen concen­ also decline, leading to less biomass overall. tration, which impacts anabolic processes; lower oxygen concentration reduces the scope for organism growth in both models, and thus reduces 2.2. How do models incorporate lower trophic level processes? total biomass. DBEM also resolves the negative impacts of acidification on catabolic processes, by incorporating pH forcings. DBEM also uses Net primary production sustains essentially all non-photosynthetic salinity, sea ice and mixed layer depth forcings, alongside temperature, life in the oceans, and limits the biomass of higher trophic levels to establish the spatial extent of each of the >1200 fish and invertebrate (Ryther, 1969; Friedland et al., 2012). Solar energy captured and species the model resolves. organic matter synthesized by primary producers flow through food webs, primarily by larger organisms preying on smaller organisms. 2.4. Experimental protocol FishMIP models focus on higher trophic levels, so lower trophic level processes are driven by a range of Earth system model forcings (Table 1). To isolate the impact of temperature and LTL processes on the The role of lower trophic levels in setting the limits to growth for higher FishMIP ensemble, we conducted four simulations (Table 3) following trophic levels is represented across the eight FishMIP models in two the general approach of Carozza et al. (2018). In each simulation, all ways. First, for BOATS, DBEM, Macroecological and EcoTroph, net models were forced with different combinations of temperature, LTL and primary production is used to determine limits of consumer growth rates other (for APECOSM and DBEM) drivers from pre-industrial, historical and total biomass according to trophic transfer functions. Second, in the and high emissions scenarios (RCP 8.5; IPCC, 2014) from the CESM1- trait-based models (APECOSM, DBPM, FEISTY and ZooMSS), plankton BGC earth system model (Moore et al., 2013). simulation submitted biomass and export production are consumed by the size classes or for the Coupled Model Intercomparison Project 5 (CMIP5; IPCC, 2014). functional groups that feed on them. This energy is then transferred to For RCP 8.5 in the CMIP5 multi-model ensemble, CESM-BGC is average higher trophic levels through size-based predation. However, all eight in temperature sensitivity and less than average in global mean NPP and models considered here are one-way forced (run offline), so there is no export production decline (Bopp et al., 2013). All forcings were provided feedback from higher trophic levels to lower trophic level biomass or to modellers with a monthly temporal resolution. We do not use a range production. This means that for the trait-based models, ingestion-fuelled (from low to high) of emission scenarios for the future, or source forcings growth of higher trophic level predators is not explicitly matched by from multiple Earth system models, as our purpose here is to isolate 5 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Table 3 Summary of the experimental simulations and corresponding environmental driver combinations. Temperature: all temperature-related drivers (e.g., sea surface temperature); LTL: all lower trophic level drivers (e.g., phytoplankton biomass); Other: any drivers that are not related to temperature or lower trophic levels (e.g., pH). The abbreviations for forcings are: PI = pre-industrial control, H = historical, RCP 8.5 = RCP 8.5. Simulation Control Temperature Change LTL Change All (Climate) Change Drivers 1950-2005 2006-2100 1950-2005 2006-2100 1950-2005 2006-2100 1950-2005 2006-2100 Temperature PI PI H RCP 8.5 PI PI H RCP 8.5 LTL PI PI PI PI H RCP 8.5 H RCP 8.5 Other PI PI PI PI PI PI H RCP 8.5 sources of structural uncertainty within the FishMIP model ensemble except in polar regions (Fig. 1f, h). The mean change in sea surface itself (Payne et al., 2016). Under the RCP 8.5 scenario, the CESM1-BGC temperature across the global ocean from 1950 to 2100 under historical model projects a global sea surface temperature increase, which is (averaged over 1950–1960) and RCP 8.5 (averaged over 2090–2100) particularly marked at high latitudes (Fig. 1b); net primary production scenarios is +3.2 ◦ C, and for net primary production, phytoplankton and declines across most of the tropics and mid-latitudes, but increases at zooplankton carbon the mean change was − 14%, − 8% and − 21%, high latitudes and in the eastern South Pacific (Fig. 1d); phytoplankton respectively. and zooplankton biomass declines across most of the world’s oceans, To enable the model comparison, two standardized outputs - total Fig. 1. Control (historical averaged over 1950–1960) forcing variables and the change in those variables from climate change (RCP 8.5) from the CESM1-BGC earth system model; a, b) Sea surface temperature, c, d) Net primary production, e, f) Phytoplankton carbon, g, h) Zooplankton carbon. The change in each variable is measured as the mean over 2090–2100 under the RCP 8.5 scenario minus the mean over 1950–1960 (for sea surface temperature), or the percentage change between the mean in 1950–1960 and 2090–2100 (for net primary production, phytoplankton carbon and zooplankton carbon). 6 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 consumer biomass (which broadly includes all consumer with trophic change in response to warming ranged from around − 35% for Macro­ level >1, see Table 1) and the biomass of large consumers (≥30 cm; see ecological and BOATS, to +3% for APECOSM by the end of the 21st Tittensor et al., 2018 for details) - were calculated from each ecosystem century. EcoTroph produced the third largest change after BOATS and model. All models supplied both outputs, except DBEM which did not Macroecological of around − 13%. The remaining four models (DBEM, provide the biomass of large consumers. Outputs were reported as depth DBPM, FEISTY, ZooMSS) simulated modest changes in global consumer integrated carbon biomass (g m− 2) and aggregated to a spatial grid with biomass of between − 2% (FEISTY) to − 7% (DBPM) in response to a resolution of 1◦ on a monthly or annual time step, depending on model changes in temperature alone. capability. Owing to differences in model formulation total consumer The LTL Change simulation also showed globally averaged biomass biomass varies widely amongst models, all else being equal (Tittensor decreases for most models, except BOATS and Macroecological, which et al., 2018). Since our focus was not on explaining these differences in projected global biomass increases (Fig. 2b). For these two models, the total biomass, but rather the differences in the responses of the models to trajectory of global biomass change was switched in the LTL Change temperature and LTL changes, we compared model outputs using simulation from negative change to positive in comparison with the biomass change relative to biomass levels under the preindustrial con­ warming only simulation. In contrast, APECOSM projected global con­ trol. Further, as our focus was isolating impacts of temperature and LTL sumer biomass to increase slightly with warming, but decrease with LTL processes, simulations were run in the absence of fishing. changes. APECOSM projected a 7% decrease in total consumer biomass globally, while BOATS and Macroecological projected increases of 3. Results 10–15% in response to LTL changes in isolation. Maximum decreases of biomass in LTL simulations are half the magnitude (up to 15%) of the 3.1. Global changes in total consumer biomass decreases in warming simulations. The smallest response to LTL changes was from EcoTroph, which projected a total consumer biomass change All models projected a decline of globally averaged consumer of <− 1%. Trends in total consumer biomass from the other five models biomass in the Temperature Change simulation, with the exception of (DBEM, DBPM, FEISTY and ZooMSS) were grouped within a range be­ APECOSM (Fig. 2a). The spread of total global consumer biomass tween − 5% (DBEM) and − 15% (ZooMSS). Fig. 2. Model projections of percentage change in global consumer biomass, relative to the Control, from 1950 to 2100 for the: a) Temperature Change simulation, b) Lower Trophic Level (LTL) Change simulation, c) All (Climate) Change simulation and d) the non-additive impacts of temperature and LTL changes. Non-additive impacts are calculated by taking the difference between the All Change and the sum of the Temperature and LTL Change simulations. 7 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 The combined temperature and LTL changes led to a decline in When both temperature and LTL drivers changed simultaneously in globally averaged consumer biomass across all models except DBEM the All (Climate) Change simulation, shifts in the distribution of con­ (Fig. 2c). By the end of the 21st century, changes in global consumer sumer biomass for each model were a combination of the shifts driven by biomass in the All (Climate) Change simulation ranged from around separate temperature and LTL effects (Fig. 3, right column; Supple­ − 30% for BOATS and Macroecological, to +3% for DBEM. The other mentary Fig. S1). Across all models, temperature-induced declines in five models (APECOSM, DBPM, EcoTroph, FEISTY and ZooMSS) had consumer biomass were generally exacerbated in regions where LTL changes in total global consumer biomass of between − 5% (for APE­ changes negatively impacted consumer biomass. Overall, consumer COSM) and − 17% (for ZooMSS). For all models except BOATS, DBEM biomass generally increased in polar waters, where all LTL variables and Macroecological, climate change impacts at the global scale were increased but temperature changed relatively little. Increases in con­ largely the sum of the separate global impacts of warming and LTL sumer biomass in DBEM were greater in polar regions under climate change, with almost no non-additive impact (Fig. 2d). For BOATS and change, compared to the sum of the separate impacts of warming and Macroecological, climate change impacts caused total consumer LTL shifts (Supplementary Fig. S1e). Outside of polar regions, the biomass to decline by about 4% more than the sum of separate warming magnitude and direction of change in consumer biomass varied among and LTL impacts. In DBEM, total consumer biomass under climate models, depending on their individual responses to temperature and LTL change was ~15% higher than under the combined, separate impacts of changes. For BOATS and Macroecological, the magnitude of positive and warming and LTL impacts, indicating some non-additive impact of cu­ negative changes in consumer biomass from LTL shifts in isolation were mulative temperature and LTL changes. Non-additive impacts in DBEM attenuated when combined with the impacts of warming in the Climate may also be caused by additional impacts from changes in pH and ox­ Change simulation (Supplementary Fig. S1g, h), however these non- ygen levels. APECOSM, the only other model to incorporate non- additive effects largely cancelled at the global scale (Fig. 2d). temperature or LTL drivers, had negligible non-additive impacts, indi­ cating these other drivers had little effect compared to warming and LTL 3.3. Disentangling temperature and lower trophic level impacts on total shifts. consumer biomass 3.2. Spatial changes in total consumer biomass Fig. 4 compares the forced changes in sea surface temperature (SST) with the co-located simulated changes in biomass for all grid cells in the Globally averaged time-series of total consumer biomass change global ocean. Regressions give negative exponential slopes for all conceal considerable spatial variation across regions within each model, models, but with substantial variation (Supplementary Table S1). and between models in each experiment. Temperature-induced shifts in Globally, consumer biomass changed between − 0.5% and − 2.0% for the spatial distribution of total consumer biomass (Fig. 3, left column) every 1 ◦ C of sea surface warming for APECOSM, FEISTY, DBPM and varied from increases in many regions for APECOSM, to decreases across ZooMSS, and between − 4.8% and − 15.4% per 1 ◦ C across EcoTroph, the global ocean in DBPM, ZooMSS, EcoTroph, BOATS and Macro­ BOATS and Macroecological (Supplementary Table S4). The models ecological. The magnitude of the total consumer biomass changes vary in their degree of linearity, with DBEM projecting the greatest generally followed the magnitude of change in temperature (Fig. 1b); nonlinearity in the impacts of warming between cold and warm waters temperate regions that experienced the strongest warming (Fig. 1b) (Fig. 4e; Supplementary Table S4). DBEM consumer biomass increased exhibited the largest decreases in biomass for these five models. FEISTY by ~50% in cold waters (<15 ◦ C SST) in response to warming (Fig. 4e), and ZooMSS consumer biomass also decreased with increased temper­ and decreased on average by >27% for each 1 ◦ C warming in warm ature in many of the regions with the greatest warming. However, in (≥15 ◦ C SST) waters. warm regions (Fig. 1a) with relatively small temperature increases such Fig. 5 shows the corresponding plots for LTL forcing. For all models, as the eastern Pacific or northern Indian Ocean, FEISTY consumer changes in total consumer biomass were positively correlated with biomass increased, and small increases in ZooMSS consumer biomass changes in their respective aggregated lower trophic level (LTL) forcing occurred almost entirely in very high latitude polar regions where (Fig. 5). A 1% change in LTL forcings caused a change in total consumer temperature change was relatively small (Fig. 1b). In contrast, APE­ biomass of between 0.6% in DBPM to 1.7% in BOATS (Supplementary COSM consumer biomass increased across most of the global ocean in Table S4). Positive correlations between consumer biomass and LTL response to warming. The exception to this pattern was in patches where changes ranged from r = 0.39 for DBPM, to r = 0.98 for EcoTroph. For phytoplankton biomass was highest (Fig. 1c) such as the North Atlantic, all models except DBPM, the greatest correlation was between change in the Bering Strait or the South Pacific around New Zealand. In DBEM, total consumer biomass and change in total LTL production, or biomass, temperature-induced changes in consumer biomass were greatest in the of the model’s chosen LTL forcing (Supplementary Table S3). In models warmest waters around the equator, where DBEM consumer biomass that used size-fractionated LTL inputs, or additional secondary LTL in­ decreased by 60–100%. In cold high latitude waters, DBEM consumer puts, changes in consumer biomass were less correlated with changes in biomass increased by ≥60% in response to warming. their main aggregated LTL forcing (APECOSM, DBPM, FEISTY) For all models, lower trophic level (LTL) induced shifts in the dis­ compared to models that did not use size-fractionated or multiple LTL tribution of consumer biomass (Fig. 3, centre column) show more forcings (BOATS, DBEM, EcoTroph, Macroecological, ZooMSS). agreement in their patterns of change; most models show biomass de­ creases in equatorial regions, and increases towards the poles. The ex­ 3.4. Impacts of warming and lower trophic level change on ecosystem ceptions here are APECOSM, FEISTY and ZooMSS which show a mix of structure positive and negative consumer biomass toward the north pole. Con­ sumer biomass shifts generally followed changes in the distribution of In response to warming, there was little consensus in the relative the main LTL forcings used by each model (Fig. 1d, f, h). APECOSM, change of small (<30 cm) and large (≥30 cm) mean global consumer DBPM, FEISTY and ZooMSS use plankton biomass inputs (Table 2), and biomass (Fig. 6a), with four models (BOATS, EcoTroph, Macro­ for these models, consumer biomass generally decreased with ecological, ZooMSS) showing a decrease of both and the other three decreasing phytoplankton carbon (Fig. 1f) and increases were isolated to models (APECOSM, DBPM, FEISTY) showing a mixture of responses. polar regions. DBEM, EcoTroph, BOATS and Macroecological use net Small consumer biomass increased by ~2% in both APECOSM and primary production as their LTL forcing and the spatial distribution of FEISTY in response to warming, but large consumer biomass increased changes in consumer biomass followed spatial shifts in net primary in APECOSM by 5% while decreasing in FEISTY by >10%. Similarly, production (Fig. 1d), with increases in biomass not only in polar regions, although small consumer biomass in DBPM and ZooMSS decreased by but also in the North Pacific and in the South East Pacific. 3% and 6% respectively, these models disagreed on the direction of 8 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Fig. 3. Maps of relative total consumer biomass averaged over 2090–2100, compared to the Control (over 2090–2100), for the Temperature (left column), Lower Trophic Level (LTL) and All Change simulations for a-c) APECOSM, d-f) FEISTY, g-i) ZooMSS, j-l) DBPM, m-o) DBEM, p-r) EcoTroph, s-u) BOATS, v-x) Macro­ ecological. Maps are ordered by the magnitude (from smallest to greatest) of the negative warming impact on consumer biomass. 9 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Fig. 4. Change in total consumer biomass (%) against the mean change in sea surface temperature (SST) over 2090–2100, for individual 1◦ grid squares, under the Temperature Change simulation, compared to the Control simulation, for a) APECOSM, b) FEISTY, c) ZooMSS, d) DBPM, e) DBEM, f) EcoTroph, g) BOATS, h) Macroecological. Each point is coloured according to the mean 1950–1960 historical SST in its grid cell. Dotted horizontal and vertical black lines indicate where % change in total consumer biomass and change in temperature are zero, respectively. The green line is the fitted regression (ΔTotal Consumer Biomass = exp(β0 + β1 ΔSST) + ε) for the change in consumer biomass with warming. We use exponential regression to calculate the line of best fit here since all models incorporate temperature effects using an exponential function (see Supplementary Information). Information about the fitted regression is in Supplementary Table S1 and S4. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) change for large consumer biomass. In response to warming total large (Supplementary Fig. S2; Fig. 1f-h). However, in DBPM, total large con­ consumer biomass in DBPM increased by 15%, and over 60% in some sumer biomass declined by 40%, while small consumer biomass regions (Supplementary Fig. S2n), but in ZooMSS total large consumer declined by only 10%. This was reflected in the spatial patterns of biomass declined by ~2% overall. Finally, small and large consumer biomass change for DBPM, with large consumer biomass varying by over biomass declined in EcoTroph, BOATS and Macroecological, and the ±60% and small consumer biomass varying by less than ±30% across spatial pattern of decline across was similar both across models and non-polar regions of the global ocean (Supplementary Fig. S2 o,p). across small and large consumer biomass (Supplementary Fig. S2). Small and large consumer biomass declined for all models (except There was no difference in the magnitude of the decline of small and DBEM, which was excluded from this part of the analysis since it did not large consumer biomass in Macroecological, however in EcoTroph and provide size-fractionated biomass) in response to climate change BOATS the magnitude of the decline in large consumer biomass was (simultaneous temperature and LTL changes) impacts (Fig. 6c). Large greater than the decline in small consumer biomass. consumer biomass declined more than small consumer biomass in BOATS, Changes in total small and large consumer biomass in response to DBPM, EcoTroph and FEISTY. In contrast, small consumer biomass lower trophic level (LTL) changes show more agreement (Fig. 6b). The declined more than large consumer biomass in APECOSM and ZooMSS, change in total small and large consumer biomass was similar in APE­ and there was no difference between small and large consumer biomass COSM, FEISTY, ZooMSS, EcoTroph, BOATS and Macroecological, and change in Macroecological. For all models, the impacts of climate change again the spatial pattern of change in small and large consumer biomass on small and large consumers were largely the sum of temperature and generally followed each model’s respective LTL forcings LTL impacts, with relatively small non-additive impacts (Fig. 6d). 10 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Fig. 5. Change in total consumer biomass (%) against change in aggregated lower trophic level forcings (LTL), from 2090 to 2100 under the LTL Change simulation, against the Control, for individual 1◦ grid squares, compared to the Control simulation for a) APECOSM, b) FEISTY, c) ZooMSS, d) DBPM, e) DBEM, f) EcoTroph, g) BOATS, h) Macroecological, with Pearson’s correlation coefficient (r) reported for each. Each point is coloured according to the average 1950–1960 historical sea surface temperature (SST) in its corresponding grid cell. The black solid line is the 1:1 line, and the dotted horizontal and vertical black lines indicate where % change in total consumer biomass and % change in LTL are zero, respectively. The green line is the fitted regression (ΔTotalConsumerBiomass = β0 + β1 ΔLTL + ε) for the change in consumer biomass with warming. For models that use more than one LTL variable (APECOSM), or size-fractionated LTL (FEISTY and DBPM), ΔLTL is calculated from the sum of all LTL forcings. Information about the fitted regression is in Supplementary Table S2 and S4. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) 4. Discussion them (APECOSM and DBEM). By separating the marine ecosystem model responses to climate-driven warming versus LTL shifts, our results The results of our experimental protocol reveal commonalities, as point toward the processes that need to be clarified to reduce the un­ well as contrasts among the FishMIP models. All models agreed that the certainty of how these two dominant drivers impact marine ecosystems. combination of warming and lower trophic level (LTL) shifts will cause substantial regional changes in consumer biomass. Furthermore, no 4.1. Warming impacts are complex model projected a significant increase in global biomass in response to climate change. However, the impacts of warming varied markedly One straightforward expectation might be that the different re­ between models, leading to large inter-model disagreements. Changes in sponses to warming reflect differences in the temperature scalings used LTL drivers were more directly correlated with the outcomes on con­ in each model. However, the differences in temperature scalings do not sumer biomass, but with substantial variation among models, and strong readily explain the variation in the results. For instance, DBPM and dependence on each model’s choice of LTL driver. For almost all models, Macroecological use identical temperature scalings (see Supplementary the combined impacts of warming and LTL changes were largely addi­ Information S2.4 and S2.7), yet DBPM’s projections of warming-induced tive at the global scale, showing little nonlinear interaction, and addi­ biomass decline are almost an order of magnitude smaller than those of tional climate change drivers (e.g., oxygen, acidification, current Macroecological. This does not mean the temperature scalings are speeds) were not significant global drivers in the models that included irrelevant, but rather that the emergent results depend on the 11 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Fig. 6. Change in total small (<30 cm) consumer biomass versus change in global large (>30 cm) consumer biomass averaged between 2090 and 2100 relative to Control simulation for each model (excluding DBEM, which did not provide small and large consumer biomass) in the a) Temperature Change simulation b) Lower Trophic Level (LTL) Change simulation, c) All Change simulation and d) the non-additive impacts of temperature and LTL changes, calculated by taking the difference between the All Change and the sum of the Temperature and LTL Change simulations. The red solid line is the 1:1 line, and the dotted horizontal and vertical black lines indicate where the percentage change in global large and small consumer biomass are zero, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) interactions of multiple temperature-dependent processes, operating species follow their thermal niche poleward and are able to take within the structural context of each model. advantage of increased primary production in high latitude regions, For example, DBEM attempts to resolve preferred temperature compared to the regions they left behind. ranges for different species, while the other models consider the effect of DBEM aside, four of the models included here (APECOSM, DBPM, temperature on generalized physiological processes, implicitly assuming FEISTY and ZooMSS) project much smaller warming impacts on con­ that species moving out of a region are replaced by species moving in sumer biomass than the remaining three models (BOATS, EcoTroph and with no change in ecosystem function. Although food web processes Macroecological). Although there are many differences between these such as predator-prey interactions are not explicitly included in the models, one particularly salient feature is that the low-sensitivity models DBEM species-distribution model, it projects an emptying of tropical all use LTL biomass as Earth-system model (ESM) drivers for the pro­ waters and a corresponding build-up of biomass in polar waters, as jections, together with temperature-dependent feeding rates. In species move poleward to follow their thermal preferences. This redis­ contrast, the high-sensitivity models use ESM net primary production to tribution of the biomass of >1200 recorded commercial species included directly limit the growth rates of upper trophic levels. We suggest that in the model reflects the absence of very warm water fish that can the discrepancy in temperature sensitivity between the model groups repopulate the tropics, and the small number of cold water fish in the can be attributed, at least in part, to an inconsistency that arises from the initial state (Cheung et al., 2010). It also largely explains the model’s 1-way forcing of marine models with LTL biomass. The relationship combined impacts of warming, LTL shifts and other drivers being between LTL biomass (BLTL ), LTL production (PLTL ) and higher trophic nonlinear: relative to extant species in polar waters, a larger number of level predation (PredHTL ) through time t can be represented as: 12 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 dBLTL in large consumer biomass in response to phytoplankton biomass de­ = PLTL (t) − PredHTL (t) dt clines is a result of biomass destruction through senescence mortality, In reality, if warming accelerates predation rates, but lower trophic which increases with body size but does not depend on food density, level production remains constant or does not increase as much, such outpacing ingestion-fuelled biomass creation. The other predation- that PLTL (t) < PredHTL (t), LTL biomass would decrease. However in the explicit models—including ZooMSS and FEISTY, which also include 1-way forcing used here, LTL biomass is determined externally by the biomass destruction processes independent of food density that increase Earth system model and is not affected by predation from higher trophic with body size—did not exhibit similar declines in large organism levels. Thus, increased predation rates from warming on fixed LTL biomass. This is because in these models, ingestion-fuelled growth out­ biomass causes an increase in the flux of biomass energy into higher paces biomass destruction from these processes, highlighting the sensi­ trophic levels that is decoupled from lower trophic level production. tivity of model outputs to the parameterisation of these rates. In fact, This increased energy input counters the increased metabolic rates and across all models except DBPM, the change in large organism biomass associated respiratory losses, dampening biomass declines from warm­ with LTL change was equal to or slightly less than the change in small ing. This inconsistency in coupling between LTLs and higher trophic organism biomass. level consumers would tend towards an underestimate of warming im­ pacts on consumer biomass. In contrast, in the production-driven models 4.3. Cumulative warming and lower trophic level impacts are largely there is no spurious energy input under warming, so that warming- additive driven increases in consumer respiration costs and decreases in repre­ sentative phytoplankton size act to drive biomass down strongly. Across the model ensemble, climate change impacts on total con­ Our results also explored the warming impacts on ecosystem struc­ sumer biomass and ecosystem structure were generally well- ture, defined as the relative biomass of small versus large organisms. approximated by the sum of separate warming and LTL impacts. This Here, there was little consensus between models. DBPM and FEISTY lack of non-linearity is perhaps less surprising for the majority of models provide a striking example of divergent projections of ecosystem struc­ that only use temperature and LTL drivers to force their models (Tit­ ture with warming. In DBPM, ingestion-fuelled anabolism outpaces tensor et al., 2018), but remarkably it also holds for APECOSM, which senescence-induced mortality in large organisms as waters warm incorporates other drivers such as oxygen, pH and current velocity. The (Blanchard et al., 2012), causing their biomass to increase. This raises fact that the overall climate change impact on consumer biomass in predation pressure on smaller organisms, which when coupled with APECOSM was close to the sum of temperature and LTL impacts in­ warming-induced increases in natural mortality, causes their biomass to dicates that the additional forcings have a comparatively small effect. decline. By contrast, in FEISTY, biomass respiration increases faster with DBEM, which also includes additional environmental drivers, did show a both body size and temperature compared to ingestion-fuelled anabo­ much stronger non-additive impact of climate change on overall con­ lism (Petrik et al., 2019) reducing the scope for growth and causing large sumer biomass, but this appeared to be driven primarily by the reloca­ organism biomass to decline with warming. Declines in large consumer tion of species niches in DBEM in response to warming, rather than the biomass in FEISTY with increasing temperature relieve predation pres­ other drivers. DBEM aside, only BOATS and Macroecological show sig­ sure on small consumers, resulting in an increase in their biomass, nificant non-linear interactions between temperature and LTL drivers. especially in tropical waters. The divergent impacts of warming on in­ This can be attributed to the fact that, in BOATS and Macroecological, dividual processes and ecosystem structure reflects the lack of consensus the representative size of phytoplankton used to force the models scales among modellers of how temperature impacts on individuals translate with both net primary production and temperature, increasing in cooler into ecosystem impacts. waters or regions with high net primary production (Dunne et al., 2005). For these two models, the spatial pattern of attenuation follows shifts in net primary production, indicating that warming attenuates the in­ 4.2. Lower trophic level impacts are influenced by choice of forcing creases and decreases in biomass from shifts in net primary production. It may be tempting to assume that the lack of nonlinear interactions The choice of LTL forcings differed between models, with each model in the models means that such nonlinearities are unlikely to exist in the using either biomass or production variables at the phytoplankton or ocean. However, an increasing number of experimental and observa­ zooplankton level, with significant impacts on the results. Generally, tional studies indicate that cumulative impacts from climate change spatial changes in consumer biomass were most correlated with changes stressors such as warming, deoxygenation and acidification are likely to in the distribution of the LTL forcing used. The sensitivity of models to be nonlinear and amplifying (Sampaio and Rosa, 2020). Rather, given the choice of LTL forcing again indicates a lack of common under­ the rudimentary representation of many ecosystem processes in the standing of how to link lower trophic levels production with higher models (e.g., no phenological or diversity-related mechanisms, trophic levels, with no consensus on whether production rates or simplistic or absent predation relationships), we suggest that it is more standing-stock biomass should be used. We believe this problem appropriate to ascribe the lack of nonlinear interactions in marine fundamentally arises out of practical necessity because of each model’s climate change projections to our present lack of ability to resolve them one-way, offline coupling with the Earth system model—were higher in the models. trophic levels and LTLs to be fully coupled, and predation feedbacks on LTLs resolved, there should theoretically be no disagreement between 4.4. Improving marine ecosystem models with observational constraints models that use production or biomass variables, everything else being equal. However, in the absence of two-way coupled models in the In this study, we have identified key sources of structural uncertainty FishMIP ensemble, the development of which is a tremendous technical that drive disparate projections of climate change impacts on the global challenge (see Aumont et al., 2018), this problem remains to be marine ecosystem. As a first step, the marine modelling community can addressed. As mentioned above, this problem also leads to inconsistency work to reduce this structural uncertainty and increase the credibility of in the temperature response when plankton biomass versus net primary ecosystem projections by constraining models with independent obser­ production rates are used. vations. An increasingly popular approach to confront model projections Ecosystem structure did not change substantially in response to LTL with observations is to use emergent constraints, which relate the long- changes, except in DBPM. Large organism biomass in DBPM declined by term climate sensitivity of an observable ecosystem feature - such as 40% and small organisms declined by <10% in response to decreases in total biomass change (Free et al., 2019) or size-spectrum slope (Blan­ phytoplankton biomass and resultant shifts in the size structure of the chard et al., 2017b; Heneghan et al., 2019) - to its short-term, observed phytoplankton abundance spectrum. DBPM’s relatively large decrease variability (Allen & Ingram, 2002; Eyring et al., 2019). Models that give 13 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 a closer fit to short-term observed variability of an ecosystem feature are 6. Code and data availability hypothesised to provide more reliable projections of its long-term variability from climate change (Kwiatkowski et al., 2017; Veytia The experimental protocol in this paper has no code associated with et al., 2020). Moreover, within a model ensemble, each model’s it. Forcing data from CMIP5 used for the protocol, and the FishMIP weighting can be linked to its ability to capture the emergent constraint model outputs presented in this paper are available on the ISIMIP (Eyring et al., 2019). This provides a more sophisticated and credible servers (https://www.isimip.org/). way to weight model projections within an ensemble, over the standard approach where all models are given equal weighting (known as model Author contributions democracy), irrespective of performance (Knutti, 2010). Emergent constraints do not require or necessarily reward any particular JLB, TDE, EDG and DPT led the conceptualisation and development ecosystem representation. This is important as differing representations of the protocol for this study, with contributions from the other authors. of the marine ecosystem across the FishMIP ensemble not only represent CH and RFH obtained and processed forcings for the modellers to our present uncertainty of the most important drivers structuring marine complete the protocol, with assistance from JV. Model simulations were ecosystems, but also the diversity of purpose and scope for which models conducted by RFH, NB, CB, WC, MC, JDE, TDE, ME, JAG, DF, JG, OM, have been built. JP, CMP, HdP, JS, TCT, PAW. RFH conducted the analysis, with assis­ Finally, it is possible for models to perform well against whole- tance from EDG and JLB. RFH led the writing of the text, with feedback ecosystem emergent constraints, while neglecting fundamental physio­ and contributions from all authors. logical or ecosystem processes (Knutti, 2010). Therefore, if we are to improve marine models, it is also necessary to consider observational Declaration of Competing Interest constraints on physiological processes such as the balance between growth and respiration with temperature, or ecosystem processes such The authors declare that they have no known competing financial as the coupling of lower and higher trophic levels. Improving our un­ interests or personal relationships that could have appeared to influence derstanding of how physiological processes such as ingestion and the work reported in this paper. metabolism respond to warming, and how changes in LTL processes propagate through marine ecosystems, are critical steps towards model Acknowledgements improvement and more robust climate impact projections. JDE was funded by Australian Research Council Discovery Projects 5. Concluding remarks DP150102656 and DP190102293. MC, JS, NB and OM received finan­ cial support by the European Union’s Horizon 2020 research and Projecting the global impact of climate change on marine ecosystems innovation programme under grant agreement No 817578 (Triatlas and fisheries is an important and challenging task. Marine ecosystem project). CH received funding from the Open Philanthropy Project. NB models represent the current understanding of how climate change and OM also acknowledge the support of the French ANR project could impact the food web and fisheries globally in the future. Yet, CIGOEF (grant ANR-17-CE32-0008-01). DPT acknowledges funding although these models have made great strides in recent years, our re­ from the ISI-MIP project to support a workshop on this topic, and the sults show that the current understanding falls short in many respects. Jarislowsky Foundation. EDG received funding from the European Our harmonized experimental protocol clearly showed that the re­ Research Council (ERC) under the European Union’s Horizon 2020 sponses to the two most important drivers of change – warming and LTL research and innovation programme (grant agreement No 682602, shifts – differ widely among models. Uncertainty in the temperature BIGSEA). RFH was funded by the Spanish Ministry of Science, Innova­ sensitivities of competing processes, including both physiology and tion and Universities through the Acciones de Programación Conjunta ecological interactions, undermine confidence in the emergent sensi­ Internacional (PCIN-2017-115). MC acknowledges the ‘Severo Ochoa tivities, and can only be improved with better observational constraints. Centre of Excellence’ accreditation (CEX2019-000928-S) to the Institute Meanwhile, the outcome of changes in both water temperature and LTL of Marine Science (ICM-CSIC). TDE acknowledges funding from the production depends strongly on the feedback of consumers on the LTL ISIMIP project to support a workshop on this topic and the Fisheries and biomass itself, a process which is not captured by any of the one-way Oceans Canada Atlantic Fisheries Fund. All authors declare no conflict of forcings available at present, and can only be rectified with fully two- interest with respect to this study. JAFS received funding from the Eu­ way coupling, which is itself sure to raise many new questions. ropean Union’s Horizon 2020 FutureMARES project (#869300). What are the implications of our results for single ecosystem model studies? The eight models used here differ significantly in their design Appendix A. Supplementary data and ecosystem representation, having been built for different purposes (Tittensor et al., 2018). Although using common outputs across models Supplementary data to this article can be found online at https://doi. has been useful here to identify shared weaknesses, this approach con­ org/10.1016/j.pocean.2021.102659. ceals the strengths of individual models to resolve certain processes and ecosystem components that other models do not. Thus, studies that References explore the unique strengths and weaknesses of individual models remain important, in order to explore questions that each model has Allen, M.R., Ingram, W.J., 2002. Constraints on future changes in climate and the been designed to address. However, results of these single model studies hydrologic cycle. Nature 419 (6903), 228–232. https://doi.org/10.1038/ should be interpreted within the greater context of sources of structural nature01092. Aumont, O., Maury, O., Lefort, S., Bopp, L., 2018. Evaluating the potential impacts of the uncertainty shared across models identified here. diurnal vertical migration by marine organisms on marine biogeochemistry. Glob. Attempting to summarise the vast complexity of the global marine Biogeochem. Cycles 32, 1622–1643. https://doi.org/10.1029/2018GB005886. ecosystem in a handful of equations is enormously difficult. The fact that Barange, M., Merino, G., Blanchard, J.L., Scholtens, J., Harle, J., et al., 2014. Impacts of climate change on marine ecosystem production in societies dependent on fisheries. independently constructed models with contrasting architectures have Nat. Clim. Change 4, 211–216. https://doi.org/10.1038/nclimate2119. arrived at many similar conclusions is encouraging, while their diversity Barnes, C., Irigoien, X., De Oliveira, J.A.A., Maxwell, D., Jennings, S., 2011. Predicting is useful to identify common weaknesses. These initial results from the marine phytoplankton community size structure from empirical relationships with FishMIP ensemble provide a glimpse into the great promise of multi- remotely sensed variables. J. Plankton Res. 33 (1), 13–24. https://doi.org/10.1093/ plankt/fbq088. model comparisons to improve our understanding of the global marine Blanchard, J.L., Heneghan, R.F., Everett, J.D., Trebilco, R., Richardson, A.J., 2017b. ecosystem and its future under change. From Bacteria to Whales: Using Functional Size Spectra to Model Marine 14 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Ecosystems. Trends Ecol. Evol. 32 (3), 174–186. https://doi.org/10.1016/j. Friedland, K.D., Stock, C., Drinkwater, K.F., Link, J.S., Leaf, R.T., et al., 2012. Pathways tree.2016.12.003. between Primary Production and Fisheries Yields of Large Marine Ecosystems. PLoS Blanchard, J.L., Jennings, S., Holmes, R., Harle, J., Merino, G., et al., 2012. Potential ONE 7, e28945. https://doi.org/10.1371/journal.pone.0028945. consequences of climate change for primary production and fish production in large Gascuel, D., Pauly, D., 2009. EcoTroph : Modelling marine ecosystem functioning and marine ecosystems. Philos. Trans. Roy. Soc. B: Biol. Sci. 367 (1605), 2979–2989. impact of fishing. Ecol. Model. 220, 2885–2898. https://doi.org/10.1016/j. https://doi.org/10.1098/rstb.2012.0231. ecolmodel.2009.07.031. Blanchard, J.L., Jennings, S., Law, R., Castle, M.D., McCloghrie, P., et al., 2009. How Heneghan, R.F., Everett, J.D., Blanchard, J.L., Richardson, A.J., 2016. Zooplankton Are does abundance scale with body size in coupled size-structured food webs? J. Anim. Not Fish: Improving Zooplankton Realism in Size-Spectrum Models Mediates Energy Ecol. 78 (1), 270–280. https://doi.org/10.1111/j.1365-2656.2008.01466.x. Transfer in Food Webs. Front. Mar. Sci. 3 (October), 1–15. https://doi.org/10.3389/ Blanchard, J.L., Watson, R.A., Fulton, E.A., Cottrell, R.S., Nash, K.L., Bryndum- fmars.2016.00201. Buchholz, A., Müller, C., 2017a. Linked sustainability challenges and trade-offs Heneghan, R.F., Everett, J.D., Sykes, P., Batten, S.D., Edwards, M., et al., 2020. among fisheries, aquaculture and agriculture. Nat. Ecol. Evol. 1, 1240–1249. https:// A functional size-spectrum model of the global marine ecosystem that resolves doi.org/10.1038/s41559-017-0258-8. zooplankton composition. Ecol. Model. 435 (August), 109265 https://doi.org/ Bopp, L., Resplandy, L., Orr, J.C., Doney, S.C., Dunne, J.P., et al., 2013. Multiple stressors 10.1016/j.ecolmodel.2020.109265. of ocean ecosystems in the 21st century: projections with CMIP5 models. Heneghan, R.F., Hatton, I.A., Galbraith, E.D., 2019. Climate change impacts on marine Biogeosciences 10, 6225–6245. https://doi.org/10.5194/bg-10-6225-2013. ecosystems through the lens of the size spectrum. Emerging Topics in Life Sciences 3 Boyce, D.G., Frank, K.T., Leggett, W.C., 2015. From mice to elephants: overturning the (2), 233–243. https://doi.org/10.0142/etls20190042. ‘one size fits all’ paradigm in marine plankton food chains. Ecol. Lett. 18 (6), IPCC, 2014. Climate Change 2014: Synthesis Report. Contribution of Working Groups I, 504–515. https://doi.org/10.1111/ele.12434. II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Boyce, D.G., Lotze, H.K., Tittensor, D.P., Carozza, D.A., Worm, B., 2020. Future ocean Change [Core Writing Team, R.K. Pachauri and L.A. Meyer (eds.)]. IPCC, Geneva, biomass losses may widen socioeconomic equity gaps. Nat. Commun. 1–11 https:// Switzerland, 151 pp. doi.org/10.1038/s41467-020-15708-9. Jennings, S., Collingridge, K., 2015. Predicting consumer biomass, size-structure, Brander, K., Neuheimer, A., Andersen, K.H., Hartvig, M., 2013. Overconfidence in model production, catch potential, responses to fishing and associated uncertainties in the projections. ICES J. Mar. Sci. 70, 1065–1068. world’s marine ecosystems. PLoS ONE 10 (7), e0133794. https://doi.org/10.1371/ Brito-Morales, I., Schoeman, D.S., Molinos, J.G., Burrows, M.T., Klein, C.J., et al., 2020. journal.pone.0133794. Climate velocity reveals increasing exposure of deep-ocean biodiversity to future Jones, M.C., Dye, S.R., Pinnegar, J.K., Warren, R., Cheung, W.W.L., 2012. Modelling warming. Nat. Clim. Change 10 (June), 576–581. https://doi.org/10.1038/s41558- commercial fish distributions: Prediction and assessment using different approaches. 020-0773-5. Ecol. Model. 225, 133–145. https://doi.org/10.1016/j.ecolmodel.2011.11.003. Brown, J.H., Gillooly, J.F., Allen, A.P., Savage, V.M., West, G.B., 2004. Toward a Knutti, R., 2010. The end of model democracy ?: An editorial comment. Clim. Change metabolic theory of ecology. Ecology 85 (7), 1771–1789. https://doi.org/10.1890/ 102, 395–404. https://doi.org/10.1007/s10584-010-9800-2. 03-9000. Kooijman, S., 2010. Dynamic Energy Budget theory for metabolic organisation, third ed. Bryndum-Buchholz, A., Prentice, F., Tittensor, D.P., Blanchard, J.L., Cheung, W.W.L., Cambridge University Press, Cambridge. et al., 2020. Differing marine animal biomass shifts under 21st century climate Kwiatkowski, L., Bopp, L., Aumont, O., Ciais, P., Cox, P.M., et al., 2017. Emergent change between Canada’s three oceans. Facets 5, 105–122. https://doi.org/ constraints on projections of declining primary production in the tropical oceans. 10.1139/facets-2019-0035. Nat. Clim. Change 7 (April), 355–359. https://doi.org/10.1038/NCLIMATE3265. Carozza, D.A., Bianchi, D., Galbraith, E.D., 2016. The ecological module of BOATS-1.0: a Lefevre, S., McKenzie, D. J., & Nilsson, G. E. (2017). Models projecting the fate of fish bioenergetically constrained model of marine upper trophic levels suitable for populations under climate change need to be based on valid physiological studies of fisheries and ocean biogeochemistry. Geosci. Model Dev. 9 (4), mechanisms. 1545–1565. https://doi.org/10.5194/gmd-9-1545-2016. Lotze, H.K., Tittensor, D.P., Bryndum-Buchholz, A., Eddy, T.D., Cheung, W.W.L., et al., Carozza, D.A., Bianchi, D., Galbraith, E.D., 2017. Formulation, General Features and 2019. Global ensemble projections reveal trophic amplification of ocean biomass Global Calibration of a Bioenergetically-Constrained Fishery Model. PLoS ONE 12 declines with climate change. PNAS 116 (26), 12907–12912. https://doi.org/ (1), e0169763. https://doi.org/10.1371/journal.pone.0169763. 10.1073/pnas.1900194116. Carozza, D.A., Bianchi, D., Galbraith, E.D., 2018. Metabolic impacts of climate change on Maury, O., 2010. An overview of APECOSM, a spatialized mass balanced “Apex Predators marine ecosystems: Implications for fish communities and fisheries. Glob. Ecol. ECOSystem Model” to study physiologically structured tuna population dynamics in Biogeogr. 28, 158–169. https://doi.org/10.1111/geb.12832. their ecosystem. Prog. Oceanogr. 84 (1–2), 113–117. https://doi.org/10.1016/j. Chavez, F.P., Messi, M., Pennington, J.T., 2011. Marine Primary Production in Relation pocean.2009.09.013. to Climate Variability and Change. Ann. Rev. Mar. Sci. 3, 227–260. https://doi.org/ Maury, O., Faugeras, B., Shin, Y.J., Poggiale, J.C., Ari, T.B., Marsac, F., 2007a. Modeling 10.1146/annurev.marine.010908.163917. environmental effects on the size-structured energy flow through marine ecosystems. Cheung, W.W.L., Dunne, J., Sarmiento, J.L., Pauly, D., 2011. Integrating ecophysiology Part 1: The model. Prog. Oceanogr. 74 (4), 479–499. https://doi.org/10.1016/j. and plankton dynamics into projected maximum fisheries catch potential under pocean.2007.05.002. climate change in the Northeast Atlantic. ICES J. Mar. Sci. 68 (6), 1008–1018. Maury, O., Poggiale, J.-C., 2013. From individuals to populations to communities: a https://doi.org/10.1093/icesjms/fsr012. Dynamic Energy Budget model of marine ecosystem size-spectrum including life Cheung, W.W.L., Frolicher, T.L., Asch, R.G., Jones, M.C., Pinsky, M.L., et al., 2016. history diversity. J. Theor. Biol. 324, 52–71. https://doi.org/10.1016/j. Building confidence in projections of the responses of living marine resources to jtbi.2013.01.018. climate change. ICES J. Mar. Sci. 73, 1283–1296. Maury, O., Shin, Y.J., Faugeras, B., Ari, T.B., Marsac, F., 2007b. Modeling environmental Cheung, W.W.L., Lam, V.W.Y., Pauly, D., 2008. Modelling Present and Climate-Shifted effects on the size-structured energy flow through marine ecosystems. Part 2: Distribution of Marine Fishes and Invertebrates. Fisheries Centre Research Reports Simulations. Prog. Oceanogr. 74 (4), 500–514. https://doi.org/10.1016/j. 16 (3), 76 https://doi.org/papers3://publication/uuid/4F260971-0169-4716- pocean.2007.05.001. AFAA-5F970130A95C. Moore, J.K., Lindsay, K., Doney, S.C., Long, M.C., Misumi, K., 2013. Marine Ecosystem Cheung, W.W.L., Lam, V.W.Y., Sarmiento, J.L., Kearney, K., Watson, R., et al., 2010. Dynamics and Biogeochemical Cycling in the Community Earth System Model Large-scale redistribution of maximum fisheries catch potential in the global ocean CESM1(BGC): Comparison of the 1990s with the 2090s under the RCP4.5 and RCP under climate change. Glob. Change Biol. 16 (1), 24–35. https://doi.org/10.1111/ 8.5 Scenarios. J. Clim. 26, 9291–9312. https://doi.org/10.1175/JCLI-D-12-00566.1. j.1365-2486.2009.01995.x. Payne, M.R., Barange, M., Cheung, W.W.L., Mackenzie, B.R., Batchelder, H.P., et al., Coll, M., Steenbeek, J., Pennino, M.G., Buszowski, J., Kaschner, K., et al., 2020. 2016. Uncertainties in projecting climate-change impacts in marine ecosystems. Advancing Global Ecological Modeling Capabilities to Simulate Future Trajectories ICES J. Mar. Sci. 73 (5), 1272–1282. https://doi.org/10.1093/icesjms/fsv231. of Change in Marine Ecosystems. Front. Mar. Sci. 7, 1–23. https://doi.org/10.3389/ Petrik, C.M., Stock, C.A., Andersen, K.H., van Denderen, P.D., Watson, J.R., 2019. fmars.2020.567877. Bottom-up drivers of global patterns of demersal, forage, and pelagic fishes. Prog. du Pontavice, H., Gascuel, D., Reygondeau, G., Maureaud, A., Cheung, W.W.L., 2020. Oceanogr. 176, 102124 https://doi.org/10.1016/j.pocean.2019.102124. Climate change undermines the global functioning of marine food webs. Glob. Pinsky, M.L., Selden, R.L., Kitchel, Z.J., 2020. Climate-Driven Shifts in Marine Species Change Biol. 26 (3), 1306–1318. https://doi.org/10.1111/gcb.14944. Ranges: Scaling from Organisms to Communities. Ann. Rev. Mar. Sci. 12, 153–179. Dunne, J.P., Armstrong, R.A., Gnanadesikan, A., Sarmiento, J.L., 2005. Empirical and https://doi.org/10.1146/annurev-marine-010419-010916. mechanistic models for the particle export ratio. Glob. Biogeochem. Cycles 19 Poloczanska, E.S., Burrows, M.T., Brown, C.J., Molinos, J.G., Halpern, B.S., et al., 2016. (December), 1–16. https://doi.org/10.1029/2004GB002390. Responses of marine organisms to climate change across oceans. Front. Mar. Sci. 3 Eddy, T.D., Bernhardt, J.R., Blanchard, J.L., Cheung, W.W.L., Colléter, M., et al., 2020. (MAY), 1–21. https://doi.org/10.3389/fmars.2016.00062. Energy Flow Through Marine Ecosystems: Confronting Transfer Efficiency. Trends Pörtner, H.-O., Karl, D.M., Boyd, P.W., Cheung, W.W.L., Lluch-Cota, S.E., et al., 2014. Ecol. Evol. 36(1), 76–86. https://doi.org/10.1016/j.tree.2020.09.006. Ocean systems. In: Field, C.B., Barros, V.R., Dokken, D.J., Mach, K.J., Eyring, V., Cox, P.M., Flato, G.M., Gleckler, P.J., Abramowitz, G., et al., 2019. Taking Mastrandrea, M.D., Bilir, T.E., Chatterjee, M., Ebi, K.L., Estrada, Y.O., Genova, R.C., climate model evaluation to the next level. Nat. Clim. Change 9 (2), 102–110. Girma, B., Kissel, E.S., Levy, A.N., MacCracken, S., Mastrandrea, P.R., White, L.L. https://doi.org/10.1038/s41558-018-0355-y. (Eds.), Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global Finkel, Z.V., Beardall, J., Flynn, K.J., Quigg, A., Rees, T.A.V., et al., 2010. Phytoplankton and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment in a changing world: Cell size and elemental stoichiometry. J. Plankton Res. 32 (1), Report of the Intergovernmental Panel on Climate Change. Cambridge University 119–137. https://doi.org/10.1093/plankt/fbp098. Press, Cambridge, United Kingdom and New York, NY, USA, pp. 411–484. Free, C.M., Thorson, J.T., Pinsky, M.L., Oken, K.L., Wiedenmann, J., et al., 2019. Impacts Ryther, J.H., 1969. Photosynthesis and Fish Production in the Sea. Science 166 (3901), of historical warming on marine fisheries production. Science 363 (March), 72–76. https://doi.org/10.1126/science.166.3901.72. 979–983. https://doi.org/10.1126/science.aau1758. Sampaio, E., Rosa, R., 2020. Climate Change, Multiple Stressors, and Responses of Marine Biota. In: Leal, F.W., Azul, A.M., Brandli, L., Özuyar, P.G., Wall, T. (Eds.), 15 R.F. Heneghan et al. Progress in Oceanography 198 (2021) 102659 Climate Action. Encyclopedia of the UN Sustainable Development Goals. Springer, van Denderen, D., Gislason, H., van den Heuvel, J., Andersen, K.H., 2020. Global analysis Cham. https://doi.org/10.1007/978-3-319-95885-9_90. of fish growth rates shows weaker responses to temperature than metabolic Stock, C.A., John, J.G., Rykaczewski, R.R., Asch, R.G., Cheung, W.W.L., et al., 2017. predictions. Glob. Ecol. Biogeogr. https://doi.org/10.1111/geb.13189. Reconciling fisheries catch and ocean productivity. Proc. Natl. Acad. Sci. 114 (8), Veytia, D., Corney, S., Meiners, K.M., Kawaguchi, S., Murphy, E.J., et al., 2020. E1441–E1449. https://doi.org/10.1073/pnas.1610238114. Circumpolar projections of Antarctic krill growth potential. Nat. Clim. Change 10 Tittensor, D.P., Eddy, T.D., Lotze, H.K., Galbraith, E.D., Cheung, W.W.L., et al., 2018. (June), 568–575. https://doi.org/10.1038/s41558-020-0758-4. A protocol for the intercomparison of marine fishery and ecosystem models: Fish- Waldron, A., Adams, V., Allan, J., Arnell, A., Asner, G., et al., (2020). Protecting 30% of MIP v1.0. Geosci. Model Dev. 11, 1421–1442. https://www.geosci-model-dev-dis the planet for nature: costs: benefits and economic implications. Campaign for cuss.net/gmd-2017-209/. Nature. Available at: https://www.conservation.cam.ac.uk/files/waldron_report_3 Trebilco, R., Baum, J.K., Salomon, A.K., Dulvy, N.K., 2013. Ecosystem ecology: Size- 0_by_30_publish.pdf. based constraints on the pyramids of life. Trends Ecol. Evol. 28 (7), 423–431. Woodworth-Jefcoats, P.A., Polovina, J.J., Howell, E.A., Blanchard, J.L., 2015. Two takes https://doi.org/10.1016/j.tree.2013.03.008. on the ecosystem impacts of climate change and fishing: Comparing a size-based and a species-based ecosystem model in the central North Pacific. Prog. Oceanogr. 138, 533–545. https://doi.org/10.1016/j.pocean.2015.04.004. 16