Leveraging Microbial Community Structure Data to Inform Ecosystem Modeling, an Approach Based on Microbial Community Segmentation

Emelia Chamberlain1, Hyewon Kim2, Scott Doney3,4 and Jeff Shovlowsky Bowman1,5, (1)University of California San Diego, Scripps Institution of Oceanography, La Jolla, CA, United States, (2)University of Virginia, Department of Environmental Sciences, Charlottesville, VA, United States, (3)University of Virginia, Charlottesville, VA, United States, (4)Woods Hole Oceanographic Institution, Department of Marine Chemistry and Geochemistry, Woods Hole, MA, United States, (5)Scripps Institution of Oceanography, University of California San Diego, La Jolla, United States
Numerical modeling is a critical method for understanding ecosystem processes. However, current approaches are typically not informed by microbial diversity data despite its growing availability. One reason for this is that diversity data is high dimensional, with hundreds to thousands of variables defined for relatively few time/space observations. This creates a discrepancy between the observed and modeled biological dimensionality, simplifying the diverse functionality of the microbial community. In Bowman et al. (2017) we introduced a technique to “segment” the microbial community into functionally coherent units (“modes”) that can be described by a single categorical variable. This categorical variable reflects key genetic traits of the microbial community from which we can make reasonable estimates of physiology (e.g., respiration or cell size) and provide a correlation between community structure and function. In our analysis of 16S rRNA gene data from a 5-year time-series we identified 8 recurrent modes in the bacterial community which were well correlated with bacterial production. Here we present a conceptual application with the Regional Test-bed Model, a 1-D data assimilation ecosystem model with parameters optimized for the pelagic coastal western Antarctic Peninsula. Through a literature search we can make reasonable estimates of 10 key bacterial parameters in the model for each mode. Using a model hind-cast of the time series, we can compare data assimilation parameter estimates with estimates from the observed mode. Once incorporated into the model framework, the bacterial model parameters used will vary with each mode, providing realistic initial model parameter guesses and therefore more robust parameter optimization. We expect this approach to improve the fidelity of predictive models, leading to a better understanding of complex ecosystem processes in changing climates.