Partitioning spatio-temporal variability in forecasts of the soil microbiome

Background/Question/Methods: Soil contains a complex, diverse microbiome that carries out many of Earth's critical biogeochemical functions. Despite their global significance, we have little understanding of how stable these communities are at multi-year timescales - and consequently, we do not know how the soil microbiome will shift with climate change in the near-term future. We used time series data (2013-2017) from 5 National Ecological Observatory Network (NEON) sites to develop models and forecasts for hundreds of microbial taxa and functional groups using soil properties and weather conditions. Bayesian hierarchical models were constructed at multiple taxonomic resolutions to determine the effect of taxonomic scale on predictability of microbial communities over space and time. We investigated the degree to which space, time, and environmental conditions can explain variation in microbial taxonomic and functional group abundances. Data from 2013-2016 were used to calibrate models and hindcasts to 2017 were performed to test the predictive accuracy of models under realistic data constraints.

Results/Conclusions: We found that spatial variability was larger than temporal variability for the majority of groups modeled. The abundance of microbial taxa is generally stable over time at the phylum level, but fluctuates at finer (genus and family) taxonomic resolutions. Temperature and precipitation explained the majority of temporal variation in group abundances, but bacterial taxonomic and functional groups were more sensitive to these environmental conditions. Obstacles to accurate forecasts included low temporal sampling resolution and uncertainty in forecasts of environmental drivers. Our forecasts will regularly assimilate new data as it is released by NEON and we expect the forecasts to provide a transparent record of our developing forecast capacity. Our approach for linking large spatio-temporal bioinformatics and ecological data has the potential to disentangle macroecological processes that shape the soil microbiome.