The subdivision of the Earths terrestrial surface into different biomes, bio-climatic realms or ecological land units is a fundamental precondition for geographical and ecological research on a global level. Those classification schemes differ significantly according to the considered criteria as well as the underlying methodology of class assignments. Despite those dissimilarities, they are unsystematically chosen in current scientific publications.
Material and Methods
21 biome, climate and land-cover classifications were compared according to the most defining parameters out of 82 climatic and environmental variables (vegetation, soil, biodiversity, atmosphere, topography, human impact). A correlation analysis and subsequent PCA lead to their recognition. Data-driven clustering resulted in the identification of the ideal number of clusters to which all individual biomes were assigned in in a divisive hierarchical clustering approach. Areas with the highest spatial overlap within clusters were identified as core zones. Their environmental characteristics were used as training data in order to derive a complete global biome map with ensemble modeling.
Revised global terrestrial biome map.
The synthesis of this machine learning optimized global terrestrial biome map based on the consideration of 21 different land classification schemes and 82 environmental parameters constructs a novel biome concept. This constitutes a fundamental resource for future biogeographical research.