Unsupervised ecosystem risk assessment in the Baltic Sea reveals bottom-oxygen depletion and fishing pressure as key threats in species-abundant hotspots

Abstract Marine ecosystems are increasingly shaped by the combined effects of multiple environmental and anthropogenic stressors, generating non-linear, spatially heterogeneous patterns of risks to ecosystem functioning. The Baltic Sea represents a paradigmatic system where geographical constraints intersect with intense human pressures. This study presents a data-driven framework for ecosystem risk assessment, conceptualising risk as an emergent property of interacting environmental, anthropogenic, and biological stressors. The approach integrates complementary unsupervised models: a clustering-based technique (Multi K-means) and a deep learning model based on a Variational Autoencoder, to capture recurrent and anomalous stressor patterns. Model-generated risk maps are statistically analysed within an ensembled risk characterisation to ensure ecological interpretability and comparability. The framework is applied to 2020 data from the central and western Baltic Sea, integrating multi-source data with abundance information on 145 marine species, commercially relevant species, and cod. The models identify spatially concentrated ecosystem risk hotspots in species-abundant areas characterised by the concurrence of bottom-oxygen depletion, depth-related constraints, and fishing pressure, consistent with expert studies. Cross-model concordance analysis reveals consistency and complementarity between the models. By reducing subjectivity while maintaining ecological interpretability, the framework provides a scalable and transferable tool for ecosystem risk assessment and ecosystem-based management.