Use of near-infrared reflectance spectroscopy to predict species composition in tree fine-root mixtures

Abstract

Assessment of belowground interactions in mixed forests has been largely constrained by the ability to distinguish fine roots of different species. Here, we explored near infrared reflectance spectroscopy (NIRS) to predict the proportion of woody fine roots in mixed samples and analyzed whether the prediction quality of NIRS models is related to the complexity of the fine-root mixture. For model calibration and validation purposes, 11 series of artificial mixed species samples containing known amounts of fine roots of up to four temperate tree species and non-woody plants were prepared. Three types of models with different calibration/validation approaches were developed and tested against external independent data for additional validation. With these models the proportion of each species in root mixtures was predicted accurately with low standard error of prediction (RMSECV/RMSEP <6.5%) and high coefficient of determination (r² > 0.93) for all fine-root mixtures. In addition, NIRS models also provided satisfactory estimates for samples with low (<15%) or no content of particular components. The predictive power of the NIRS models did not decrease substantially with increasing complexity of the root samples. The approach presented here is a promising alternative to hand sorting of fine roots, which may be influenced substantially by operator variation, and it will facilitate investigating belowground interactions between woody species.