Using Massive Multivariate NIRS Data in Ryegrass

Publikation: KonferencebidragPosterForskning

  • Vahid Edriss, Danmark
  • Morten Greve-Pedersen, DLF-Trifolium, Danmark
  • Christian S Jensen, DLF-Trifolium A/S, Danmark
  • Just Jensen
Near infrared spectroscopy (NIRS) analytical techniques is a simple, fast and low cost method of high dimensional phenotyping compared to usual chemical techniques. To use this method there is no need for special sample preparation. The aim of this study is to use NIRS data to predict plant traits (e.g. dry matter, protein content, etc.) for the next generation. In total 1984 NIRS data from 995 ryegrass families (first cut) were used. The Absorption of radiation in the region of 960 – 1690 nm in every 2 nm distance produced 366 bins to represent the NIRS spectrum. The amount of genetic variance and the heritability for each bin were estimated using a mixed model. To use all the information for prediction, since we have 366 bins, a reduction in number of parameters is necessary. The usual method is to combine principal component analysis (PCA) and partial least square (PLS). Another method is random regression, which have the advantage that dimension reduction can be applied to different elements in the model. From initial results, heritabilities were between 0.15 and 0.24 for the 366 bins. Phenotypic correlations between bins ranged from 0.85 to 0.99. These correlations indicate that we can reduce the parameter dimension with random regression or PCA.
Udgivelsesår10 jan. 2015
Antal sider1
StatusUdgivet - 10 jan. 2015
BegivenhedPlant and Animal Genome Conference XIII - Californien, San Diego, USA
Varighed: 10 jan. 201514 jan. 2015


KonferencePlant and Animal Genome Conference XIII
BySan Diego

Se relationer på Aarhus Universitet Citationsformater

ID: 86137407