PROBABILISTIC MAPPING OF RHODOLITH BEDS USING ENSEMBLE MACHINE LEARNING: A Case Study on Rhodolith Distribution in Scandinavia

Abstract

Rhodolith beds are ecologically important benthic habitat that help support marine biodiversity and local pH buffering and long-term carbon storage, yet in Scandinavian water their distribution remains poorly mapped. This study applies a species distribution modelling (SDM) and machine learning to predict current and future habitat Rhodolith habitat sustainability across Norway, Sweden and Denmark; using presence and pseudo absence data from the Global Biodiversity Information Facility (GBIF), three models; Random Forest, XGBoost and Logistic Regression were trained and combined into a soft voting ensemble. Environmental variables were sourced from Bio oracle v3. Including present and SSP2- 4.5 future climate projection (2020-2100). Model performance was assessed using accuracy, ROC AUC, precision, recall, and F1-score. The ensemble model showed high overall performance, though Random Forest performed slightly better. Spatial predictions revealed high suitability in Skagerrak and upper Kattegat with some moderately suitable areas in the Baltic and North Sea. Future projections indicated a substantial decline in suitable habitat between 2020 and 2100. The results showcase both known Rhodolith location, but also potentially overlooked areas, offering support to further studies in Rhodolith in the Scandinavia.