Friday, August 16, 2019
M108, Kentucky International Convention Center
Global soil microbiome surveys have shown the distribution of microorganisms is linked to environmental factors. Yet, a central question remains: will these relationships generalize out of sample? Can we predict the forest microbiome? We assembled data on soil fungal composition from 130 sites across temperate latitudes, fit models, and then used models to predict fungal abundances out of sample across the NEON network. We discovered remarkable scale dependence, with fungi becoming more predictable as spatial scale increased (core < plot < site). Variance decomposition shows forecast uncertainty is dominated by process error. This implies that improving forecasts requires better models.