COS 74-8 - Ecological forecasting of soil bacteria: Predicting taxonomic and functional groups across the United States

Thursday, August 15, 2019: 10:30 AM
M109/110, Kentucky International Convention Center
Zoey R. Werbin, Biology, Boston University, Boston, MA, Colin Averill, Environmental Systems Sciences, ETH Zürich, Zürich, Switzerland, Michael C. Dietze, Earth and Environment, Boston University, Boston, MA and Jennifer M. Bhatnagar, Department of Biology, Boston University, Boston, MA
Background/Question/Methods: Soil microorganisms regulate many of the Earth’s biogeochemical processes, but we have little understanding of what drives microbial biogeography. Understanding microbial spatial patterns and environmental niche is essential for incorporating microbial dynamics into biogeochemical models, but many believe that the heterogeneity of soil microbial communities precludes any meaningful predictive capacity. To test this hypothesis, and explore the limits of prediction and uncertainty in the soil bacterial microbiome, we created Bayesian statistical models of soil bacterial groups based on abiotic predictors using 68 sites from a globally-sampled dataset. We then validated these models out of sample by predicting soil bacteria to the core, plot, and site scale at 11 locations across the United States, using 16S amplicon sequencing data collected by the National Ecological Observatory Network (NEON). We forecasted 120 cosmopolitan soil taxonomic groups and 11 functional groups involved in soil biogeochemical cycling, as well as taxa with hypothesized copiotrophic or oligotrophic life-strategies.

Results/Conclusions: In-sample model fits for functional groups within our global calibration dataset were remarkably accurate, with a mean of 49% of the variation in functional group abundance captured by our models. Accuracy of in-sample model fits for cosmopolitan taxa varied by taxonomic rank, with lowest accuracy at the phylum level. The accuracy of forecasts to new sites generally reflected in-sample trends, but forecast accuracy also varied by spatial scale, with highest accuracy in site-level forecasts. We decomposed sources of variance within forecasts and found that process uncertainty was dominant over parameter error and covariate uncertainty. Our results suggest that the biogeography of major taxonomic and functional groups of bacteria within the soil microbiome are fundamentally predictable at high spatial scales, and we highlight data that may improve future forecasts.