Simulating Microbial Functional Composition Across Diverse Environments Through Machine Learning
Presenter: Changpeng Fan1
Co-Author(s): -
Advisor(s): Dr. Yang Song
1Department of Hydrology and Atmospheric Sciences, University of Arizona
Panapto Presentation Video
Poster PDF
Poster Session 1
Microbial communities are crucial in soil organic matter (SOM) decomposition, making them an inevitable part of the Earth system, especially in carbon cycling. However, simulations of microbial activities in the Earth System Model (ESM) have great uncertainty due to the incomprehension of the spatial distribution of microbial functional diversity and the parameterization approaches to integrate complex microbial taxonomic and functional information into ESM. We applied a series of machine learning (ML) models to simulate and predict the microbial functions based on data from microbial samples within CONUS to bridge these gaps. The target variables are clustered relative abundances of enzymes involved in SOM decomposition from JGI – IMG. These selected enzymes and corresponding abundances are classified into enzyme functional groups, with each group representing a kind of decomposition process in the ESM. The predictors are essential climate, soil physical and chemical variables involved in SOM decomposition. These ML models forecast the target variables with corresponding predictors after training with part of the dataset, estimating the spatial distribution of microbial functions for SOM decomposition across CONUS. The results indicate that general microbial functions involved in SOM decomposition can be simulated at a satisfactory level across diverse environments within the study region. The important factors that dominate the SOM decomposition are soil stoichiometry, soil carbon content, and the ecosystem type. Furthermore, the microbial functional dataset can help understand and simulate SOM decomposition in ESM.