Use of NIR hyperspectral imaging and chemometrics to quantify roots and crop residues in soil

Introduction

Monitoring of root development and crop residues decomposition in crop soils is important to understand the effects of agricultural practices and to improve them; however their quantification in the soils is still a challenge. Currently used method based on soil coring allows multiple samplings but too many time and effort are needed to manually extract and separate roots and crop residues from the soil. The aim of this work is to present an alternative for this tedious manual sorting by using NIR hyperspectral imaging and chemometric tools, which should allow separation of roots and crop residues after washing on the basis of their respective NIR spectra.

Material and methods

This work has been developed for roots and crop residues of winter wheat. For that, soil cores taken under the crop were first washed and dried, then NIR images were acquired with a push broom NIR hyperspectral imaging camera collecting spectra in the range 1100-2500 nm and with a spatial resolution of 0.3 mm. Spectra of roots and crop residues were then discriminated using models developed with a Support Vector Machine (SVM) discriminant analysis. For calibration, 1000 NIR spectra were used for each class. To quantify, the number of pixels detected for each class on new NIR images was converted in terms of mass of dry matter.

Results and discussion

Applied on an independent set of 1000 spectra, SVM classified well 89.5% of crop residues spectra and 99.2 % of root spectra. When the method was applied on new NIR images containing only crop residues or roots, 79.4 % of crop residues pixels and 92.5 % of roots pixels were well classified. Confusion appeared mainly on the border of crop residues, probably due to effect of shadow on spectra, and with dead roots.