AI scientists use satellite imagery and self-supervised learning to map underground fungal diversity across Europe and Asia with unprecedented detail

Self-supervised learning models applied to satellite data can predict below-ground ectomycorrhizal fungal richness, explaining over 50% of variance across ~12,000 field samples

SSL-derived features proved more informative than traditional climate, soil, and land cover datasets for predicting fungal biodiversity

The approach achieves 10,000-fold improvement in spatial resolution, advancing from 1km landscape averages to 10m habitat-level mapping

Research reveals that approximately 90% of mycorrhizal diversity hotspots remain unprotected, highlighting the need for better monitoring techniques