While the physical dimensions of climate change are now routinely assessed through multimodel intercomparisons, projected impacts on the global ocean ecosystem generally rely on individual models with a specific set of assumptions. To address these single-model limitations, we present standardized ensemble projections from six global marine ecosystem models forced with two Earth system models and four emission scenarios with and without fishing. We derive average biomass trends and associated uncertainties across the marine food web. Without fishing, mean global animal biomass decreased by 5% (±4% SD) under low emissions and 17% (±11% SD) under high emissions by 2100, with an average 5% decline for every 1 °C of warming. Projected biomass declines were primarily driven by increasing temperature and decreasing primary production, and were more pronounced at higher trophic levels, a process known as trophic amplification. Fishing did not substantially alter the effects of climate change. Considerable regional variation featured strong biomass increases at high latitudes and decreases at middle to low latitudes, with good model agreement on the direction of change but variable magnitude. Uncertainties due to variations in marine ecosystem and Earth system models were similar. Ensemble projections performed well compared with empirical data, emphasizing the benefits of multimodel inference to project future outcomes. Our results indicate that global ocean animal biomass consistently declines with climate change, and that these impacts are amplified at higher trophic levels. Next steps for model development include dynamic scenarios of fishing, cumulative human impacts, and the effects of management measures on future ocean biomass trends.

Abstract. Model intercomparison studies in the climate and Earth sciences communities have been crucial to building credibility and coherence for future projections. They have quantified variability among models, spurred model development, contrasted within- and among-model uncertainty, assessed model fits to historical data, and provided ensemble projections of future change under specified scenarios. Given the speed and magnitude of anthropogenic change in the marine environment and the consequent effects on food security, biodiversity, marine industries, and society, the time is ripe for similar comparisons among models of fisheries and marine ecosystems. Here, we describe the Fisheries and Marine Ecosystem Model Intercomparison Project protocol version 1.0 (Fish-MIP v1.0), part of the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP), which is a cross-sectoral network of climate impact modellers. Given the complexity of the marine ecosystem, this class of models has substantial heterogeneity of purpose, scope, theoretical underpinning, processes considered, parameterizations, resolution (grain size), and spatial extent. This heterogeneity reflects the lack of a unified understanding of the marine ecosystem and implies that the assemblage of all models is more likely to include a greater number of relevant processes than any single model. The current Fish-MIP protocol is designed to allow these heterogeneous models to be forced with common Earth System Model (ESM) Coupled Model Intercomparison Project Phase 5 (CMIP5) outputs under prescribed scenarios for historic (from the 1950s) and future (to 2100) time periods; it will be adapted to CMIP phase 6 (CMIP6) in future iterations. It also describes a standardized set of outputs for each participating Fish-MIP model to produce. This enables the broad characterization of differences between and uncertainties within models and projections when assessing climate and fisheries impacts on marine ecosystems and the services they provide. The systematic generation, collation, and comparison of results from Fish-MIP will inform an understanding of the range of plausible changes in marine ecosystems and improve our capacity to define and convey the strengths and weaknesses of model-based advice on future states of marine ecosystems and fisheries. Ultimately, Fish-MIP represents a step towards bringing together the marine ecosystem modelling community to produce consistent ensemble medium- and long-term projections of marine ecosystems.

Understanding long‐term, ecosystem‐level impacts of climate change is challenging because experimental research frequently focuses on short‐term, individual‐level impacts in isolation. We address this shortcoming first through an interdisciplinary ensemble of novel experimental techniques to investigate the impacts of 14‐month exposure to ocean acidification and warming (OAW) on the physiology, activity, predatory behaviour and susceptibility to predation of an important marine gastropod (Nucella lapillus). We simultaneously estimated the potential impacts of these global drivers on N. lapillus population dynamics and dispersal parameters. We then used these data to parameterize a dynamic bioclimatic envelope model, to investigate the consequences of OAW on the distribution of the species in the wider NE Atlantic region by 2100. The model accounts also for changes in the distribution of resources, suitable habitat and environment simulated by finely resolved biogeochemical models, under three IPCC global emissions scenarios. The experiments showed that temperature had the greatest impact on individual‐level responses, while acidification had a similarly important role in the mediation of predatory behaviour and susceptibility to predators. Changes in Nucella predatory behaviour appeared to serve as a strategy to mitigate individual‐level impacts of acidification, but the development of this response may be limited in the presence of predators. The model projected significant large‐scale changes in the distribution of Nucella by the year 2100 that were exacerbated by rising greenhouse gas emissions. These changes were spatially heterogeneous, as the degree of impact of OAW on the combination of responses considered by the model varied depending on local‐environmental conditions and resource availability. Such changes in macro‐scale distributions cannot be predicted by investigating individual‐level impacts in isolation, or by considering climate stressors separately. Scaling up the results of experimental climate change research requires approaches that account for long‐term, multiscale responses to multiple stressors, in an ecosystem context.