Decoding bacterial and fungal richness with autoencoders yields a unified ratio indicating soil health and ecological susceptibility

Soil microbial communities are central to ecosystem functioning, yet their large-scale diversity patterns and environmental sensitivities remain poorly understood. Here, we analysed a national soil microbiome dataset spanning all major Australian ecosystems to assess bacterial and fungal richness across diverse environmental gradients. Using supervised deep autoencoders—neural networks that compress complex data to identify key patterns—and structural equation models, we explained around 60% of the variance in richness. Bacteria and fungi showed contrasting spatial patterns, fungal richness was tightly constrained by moisture and organic carbon, while bacterial richness peaked in nitrogen-rich, topographically complex regions and persisted across broader conditions. We propose the bacterial-to-fungal richness ratio as a spatially explicit, scalable indicator of soil community composition and ecological condition. This ratio captures gradients of aridity, nutrient imbalance, and land-use intensification, and may help identify ecosystems vulnerable to environmental change, bridging microbial biogeography with applied soil health assessment across global biomes. The bacterial-to-fungal richness ratio serves as a unified, scalable indicator of microbial balance, soil health and ecological susceptibility, derived from spatial analysis, deep autoencoders, and structural equation modelling of Australian soil microbiomes.