COS 269-3 - Forecasting the soil microbiome across the United States

Abstract: Background/Question/

Methods: The soil microbiome regulates many of the Earth's biogeochemical processes, but it is unclear how belowground communities will shift in the near future. Our capacity to predict changes has been limited by the sheer complexity of soil communities, which contain thousands of taxa and vary dramatically between patches of soil. Furthermore, soil microbes have demonstrated some variation between seasons, but the magnitude of this variability remains uncertain. To identify relationships between spatial heterogeneity, seasonality, and predictability, we developed iterative Bayesian forecasts for soil fungal and bacterial groups across a wide range of biomes. We used 16S and ITS amplicon sequencing data to calibrate models for 236 taxa and 40 functional groups using two years of microbial time-series data (2016-2017) at 27 sites monitored by the National Ecological Observatory Network (NEON). Abundances were modeled at 264 plots using soil moisture, pH, percent carbon, leaf area index (LAI), and percent ectomycorrhizal trees. We then created forecasts for subsequent years (2018-2019) at 11 held-out NEON sites to evaluate whether the soil microbiome could be accurately forecasted in time and space.



Results/

Conclusions: We find that most microbes (over 70% of modeled groups) have seasonal trends that are not fully explained by seasonally-varying environmental predictors (e.g., LAI, soil temperature, and soil moisture). The amplitude of seasonal trends is larger for fungi than bacteria (p< 0.01), as is spatial heterogeneity within sampling sites (p< 0.01). This variability over space and time makes fungi more difficult to accurately forecast than bacteria, which are much more temporally stable at any given site. On average, bacterial groups are more sensitive than fungal groups to soil pH and percent carbon, whereas fungal groups are more sensitive to soil temperature and moisture. When forecasts were made at new sites based on minimal site information (e.g. soil chemistry and historical climate), we found that functional groupings of soil taxa retained more accuracy than taxonomic groupings, with the predicted~observed relationship R2 showing a mean decrease of 7% for functional groups and 34% for taxonomic groups. These results demonstrate that seasonality is a key force shaping the soil microbiome, yet it is poorly captured by commonly-measured environmental variables. The broad spatiotemporal scale of this analysis also serves to establish baseline trends in soil microbial ecology, which will prove especially valuable as the effects of climate change increase in magnitude and complexity.