Every marine protected area begins with a map. Before deciding where to protect the ocean, we need to know where its species live today and where they are likely to live as the climate changes. In a new study published in Scientific Data, developed within the MPA Europe project and led by colleagues at the Ocean Biodiversity Information System (OBIS, UNESCO), we present range maps for more than 12,000 marine species across pan-European seas, under present-day conditions and five climate futures.
The knowledge gap
Under the Kunming-Montreal Global Biodiversity Framework, countries have committed to protecting at least 30% of the planet by 2030. In the ocean, the gap remains wide, with only about 3% currently highly protected. Closing it effectively depends on a deceptively simple question: where should new marine protected areas be placed? Answering it requires spatially explicit knowledge of where species occur. Yet, even though the number of records in OBIS has doubled over the past four years to more than 141 million, marine distribution data remain patchy and biased towards certain regions and taxonomic groups. Climate change adds a further layer of complexity, shifting species ranges and potentially altering the very communities that protected areas are designed to safeguard. Existing large-scale products were either too coarse or offered too few climate scenarios to guide the design of protected area networks in European seas.
Our approach
We built an open-source species distribution modelling framework in R and applied it to all marine species recorded in pan-European waters. Occurrence data from OBIS and GBIF were combined with high-resolution environmental predictors from Bio-ORACLE v3.0, at 0.05° resolution (approximately 5.5 km). Predictor sets were tailored to seabirds, marine mammals, photosynthesizers and ectotherms, and to surface or seafloor conditions depending on each species’ mode of life. Models were fitted using the full global range of each species to avoid niche truncation, within a spatial point-process framework suited to presence-only data. Three machine learning algorithms (MAXENT, Random Forest and XGBoost) were applied to species with at least 30 records, while an Ensemble of Small Models was used for data-poor species with 10 to 30 records. Distributions were then projected to present-day conditions and to five Shared Socioeconomic Pathways (SSP1 to SSP5) for 2050 and 2100.
The technological challenge
Modelling tens of thousands of species cannot be done by hand. Starting from a list of 30,043 marine species, the pipeline processed nearly one billion occurrence records, so every step had to be automated, parallelized and computationally efficient. We developed an automated quality-control pipeline that removes records falling on land, environmental outliers and geographic outliers, the latter detected with an Isolation Forest algorithm applied to distances that treat land as a barrier to dispersal. Spatial autocorrelation was reduced by thinning records, model parameters were tuned with spatial block cross-validation, and only models reaching a Continuous Boyce Index of at least 0.3 were retained. Each species is delivered with ensemble predictions, bootstrap uncertainty layers, extrapolation metrics (MESS and SHAPE) and a set of spatial masks, and the framework was benchmarked against virtual species with fully known distributions. All outputs are stored as cloud-optimized GeoTIFFs, accessible through a cloud bucket, a STAC catalogue and an interactive online application.
What we found
The final dataset provides validated range maps for 12,039 species, approximately 40% of all marine species recorded in European seas, with 89% modelled using standard algorithms and 11% using methods designed for small sample sizes. Projections for 2050 and 2100 under five climate scenarios allow users to track how the range of each species may expand, contract or shift. Beyond individual species, the maps were stacked to estimate species richness for the whole assemblage and for specific groups, and to map the potential distribution of key habitat-forming species, including seagrasses, kelps, maerl, cold-water corals, reef-building polychaetes and bivalve beds. Among the species that could not be mapped, more than half lacked sufficient occurrence records, a clear reminder of how much of marine biodiversity remains undersampled.
Implications for conservation and management
These maps provide spatially consistent biodiversity inputs for systematic conservation planning across European seas:
Identifying priority areas for new marine protected areas that maximize the representation of marine biodiversity, in support of the 30x30 target of the Global Biodiversity Framework.
Designing climate-smart protected area networks, by locating areas where species are expected to persist and anticipating where protected communities may change by 2050 and 2100.
Informing marine spatial planning and the management of habitat-forming species that underpin blue carbon and ecosystem services.
Offering an open and transferable framework that can be applied to other regions of the world, contributing to Sustainable Development Goal 14 (Life Below Water).
Range maps describe where environmental conditions are suitable for a species, not where it is guaranteed to occur, and are most powerful when combined with local knowledge of habitats and human pressures.