Characterizing soil microbial properties using MIR spectra across 12 ecoclimatic zones (NEON sites)

Abstract

We investigated the use of mid-infrared (MIR) soil spectra to characterize soil microbial properties, including microbial community compositions based on PLFA biomarkers as well as bacterial abundance and diversity based on 16S ribosomal RNA gene sequencing. Samples (n = 70 for PLFAs, n = 86 for 16S samples) were collected from 12 ecoclimatic domains with eight soil orders, different land use, and topography across the US. PLSR models were developed to predict soil microbial properties using MIR spectral data. Two validation schemes were used: 1) randomly splitting the dataset into 70% for calibration and 30% for validation and the random sampling was executed 50 times; 2) leave-one-domain-out validation in which each of the 12 domains was used as the validation dataset and the remaining samples from 11 domains were used to train the PLSR models. When the validation samples were randomly selected, some PLFAs (G+ bacteria, G− bacteria, bacteria, Actinomycetales, fungi, total lipids), and the relative abundances of some bacterial phyla (Actinobacteria, Acidobacteria, Gemmatimonadetes), and bacterial diversity metrics (Shannon diversity, No. of OTUs, Pielou evenness) were moderately-well predicted from the MIR spectra with mean r2 > 0.60 in the validation. In the leave-one-domain-out validation, prediction accuracy varied for different NEON domains. Good prediction results (i.e., RMSE values similar to or smaller than mean validation RMSE in the random sampling case) were obtained for domains D10–17 in Alfisol, Mollisols, Ultisols, Entisols, Inceptisols, and Aridisols, whereas poor predictions (i.e., large RMSE values and samples more dispersed from 1:1 line) were obtained for D2, 3, and 5 in some Alfisols, Histosols, and Spodosols. The NMDS plots using PLFAs, relative abundances of bacterial phyla, and their corresponding MIR spectra showed that MIR spectra could be used to distinguish spatial patterns of microbial properties. Soil organic carbon, sand content, and pH affected the spatial distribution of microbial biomass and diversity and improved the prediction of microbial properties from MIR spectra. We conclude that MIR spectroscopy can be used to characterize soil microbial properties for large-scale ecological modeling and monitoring.