Abstract
Canopy reflectance captures plant traits related to ecological processes, which may reflect the composition of soil microbial communities. However, the extent to which canopy reflectance can help elucidate soil microbial community composition and diversity across biomes remains unclear. Using data from 14 National Ecological Observatory Network ecoregions (domains), we linked plant traits to soil microbial composition and diversity (characterised by phospholipid fatty acids and 16S rRNA gene sequencing) and built partial least squares regression models to predict soil microbial attributes from airborne imaging spectroscopy at the continental scale. The ability of remote sensing to predict soil microbial communities was mediated by plant attributes that both directly influence microbial communities and reflect shared responses to soil and climate gradients. Model validation accuracy varied with taxonomic resolution (normalised root mean squared error, 10.1-24%; coefficient of determination, 0.27-0.86), with models of broad soil microbial groups performing best, although bacterial community composition and diversity could also be modelled with moderate levels of accuracy (normalised root mean squared error, 12.5-18.6%; coefficient of determination, 0.43-0.61). Models using full-spectrum hyperspectral data consistently outperformed those based on simple vegetation indices, highlighting the value of imaging spectroscopy for soil microbial research.