Quantifying physiological trait variation with automated hyperspectral imaging in rice

Abstract

ABSTRACT Advancements in hyperspectral imaging (HSI) and establishment of dedicated plant phenotyping facilities have enabled researchers to gather large quantities of plant spectral images with the aim of inferring target phenotypes non-destructively. However, large volumes of data that result from HSI and corequisite specialized methods for analysis may prevent plant scientists from taking full advantage of these systems. Here, we explore estimation of physiological traits in 23 rice accessions using an automated HSI system. Under contrasting nitrogen conditions, HSI data are used to classify treatment groups with ≥ 83% accuracy by utilizing support vector machines. Out of the 14 physiological traits collected, leaf-level nitrogen content (N, %) and carbon to nitrogen ratio (C:N) could also be predicted from the hyperspectral imaging data with normalized root mean square error of predictions smaller than 14% (R 2 of 0.88 for N and 0.75 for C:N). This study demonstrates the potential of using an automated HSI system to analyze genotypic variation for physiological traits in a diverse panel of rice; to help lower barriers of application of hyperspectral imaging in the greater plant science research community, analysis scripts used in this study are carefully documented and made publicly available. HIGHLIGHT Data from an automated hyperspectral imaging system are used to classify nitrogen treatment and predict leaf-level nitrogen content and carbon to nitrogen ratio during vegetative growth in rice.

Plant phenotyping relevance

自動ハイパースペクトル画像を用いてイネの生理形質を非破壊推定し、予測精度を評価しているため、フェノタイピング手法が中心的です。

Code and data availability

The paper's collected/analyzed datasets (hyperspectral imaging and physiological trait data) are publicly deposited in the Purdue University Research Repository, and the authors' analysis code is publicly available on GitHub. Both are paper-specific, public, and actionable.