Quantification of nitrogen concentration in perennial ryegrass and red fescue using near-infrared reflectance spectroscopy (NIRS) and chemometrics

Abstract

Measurements of nitrogen (N) concentrations in plant samples are increasingly being used to support the development of novel N application strategies, which are based on actual plant N concentration as well as introduction of N budgets in agriculture. In order to meet the increasing demands for N measurements, the development of a fast and cheap, but still reliable technique is required. In the present study it was accordingly investigated whether near-infrared spectroscopy (NIRS) can be implemented for measurement of N concentration in grass samples. From 2000 to 2002 a total of 837 plant samples were collected from different field trials on 12 sampling sites in Denmark. The sample set consisted of 17 cultivars of red fescue ( Festuca rubra L.) and perennial ryegrass ( Lolium perenne L.) with a range in N concentration from 0.6 to 6.26% N. Visual-NIRS measurements (400–2498 nm) were performed on the dried, ground samples and plant N concentrations were measured using the Dumas method. Partial least squares regression models were developed on the near-infrared (NIR) spectra (1100–2498 nm) and the N concentrations in the dry grass samples with the aim of predicting the N concentration in samples not contained in the models. Models on raw and scatter corrected spectra gave root mean square error of prediction, RMSEP=0.19–0.24% N and correlation coefficients, R=0.97–0.98, when tested on an independent test set of samples from all harvest years, whereas models tested on samples from a harvest year not included in the calibration gave RMSEP=0.23–0.35% N and R=0.95–0.99. The prediction error is higher than the reproducibility of the Dumas method, but the NIRS method developed can still be used for measuring the N concentration in samples of perennial ryegrass and red fescue with sufficient precision and accuracy for practical use. Studies of the year effect showed that samples from more years needs to be included in the calibration data in order to increase the robustness of the model.