The Potential of Hyperspectral Patterns of Winter Wheat to Detect Changes in Soil Microbial Community Composition

Abstract

Reliable information on soil status and crop health is crucial for detecting and mitigating disasters like pollution or minimizing impact from soil-borne diseases. While infestation with an aggressive soil pathogen can be detected via reflected light spectra, it is unknown to what extent hyperspectral reflectance could be used to detect overall changes in soil biodiversity. We tested the hypotheses that spectra can be used to (1) separate plants growing with microbial communities from different farms; (2) to separate plants growing in different microbial communities due to different land use; and (3) separate plants according to microbial species loss. We measured hyperspectral reflectance patterns of winter wheat plants growing in sterilized soils inoculated with microbial suspensions under controlled conditions. Microbial communities varied due to geographical distance, land use and microbial species loss caused by serial dilution. After 3 months of growth in the presence of microbes from the two different farms plant hyperspectral reflectance patterns differed significantly from each other, while within farms the effects of land use via microbes on plant reflectance spectra were weak. Species loss via dilution on the other hand affected a number of spectral indices for some of the soils. Spectral reflectance can be indicative of differences in microbial communities, with the Renormalized Difference Vegetation Index the most common responding index. Also, a positive correlation was found between the Normalized Difference Vegetation Index and the bacterial species richness, which suggests that plants perform better with higher microbial diversity. There is considerable variation between the soil origins and currently it is not possible yet to make sufficient reliable predictions about the soil microbial community based on the spectral reflectance. We conclude that measuring plant hyperspectral reflectance has potential for detecting changes in microbial communities yet due to its sensitivity high replication is necessary and a strict sampling design to exclude other ‘noise’ factors.

Removal of soil biota alters soil feedback effects on plant growth and defense chemistry

“…plant size or N content). Hyperspectral reflectance patterns in leaves have been used to detect the sensitivity of J. vulgaris to soils (Carvalho et al ., ). Indices related to plant stress, such as PPR, NRI and PSa, were higher in plants exposed to whole soil or 1000‐ and 20‐μm watery inocula, in which larger sized soil organisms, such as fungi, were present, than in plants exposed to watery inocula without these organisms (Fig.…”

Evaluating Hyperspectral Vegetation Indices for Leaf Area Index Estimation of Oryza sativa L. at Diverse Phenological Stages

“…Elevated N application generates taller plants with higher numbers of tillers and a greater leaf area, which requires more carbohydrates in the plant canopy ( Pan et al, 2011 ). Not only the biophysical characteristics of vegetation, canopy architecture, atmospheric absorption, and scattering affect the canopy hyperspectral reflectance but so do the direction of incidence radiation and soil backgrounds ( Carvalho et al, 2016 ). During early phenological periods of our rice crop, variations in the maximum spectral reflectance in the visible region are likely to be small, which might be due to the soil water background and nitrogen contents in the canopy under varying N application.…”

Approaches, applications, and future directions for hyperspectral vegetation studies: An emphasis on yield‐limiting factors in wheat

“…The algorithm then learns the mapping function from the input data to the corresponding output variables. Common supervised classification techniques for hyperspectral data include linear discriminant analysis (LDA) (Arafat, Aboelghar, & Ahmed, 2013; Carvalho, van der Putten, & Hol, 2016), support vector machines (SVM) (Axelsson et al., 2013; Xiaming et al., 2015), partial least squares‐discriminant analysis (PLS‐DA) (Liu et al., 2014; Matzrafi et al., 2017), and artificial neural networks (ANNs) (Goel et al., 2003; Yi, Huang, Wang, & Liu, 2007). Unsupervised techniques are less common than supervised techniques but include k ‐means clustering (Behmann et al., 2014; Bergsträsser et al., 2015) and PCA (Kalacska, Bohlman, Sanchez‐Azofeifa, Castro‐Esau, & Caelli, 2007; Liu et al., 2014).…”