Hyperspectral indicators and characterization of glyphosate-induced stress in common lambsquarters (Chenopodium album L.)

Abstract

· Treatments induced both lethal and sub-lethal hormetic CHEAL response to glyphosate · 31 key wavelengths capture CHEAL spectral response to glyphosate · Four vegetation indices best described glyphosate-induced herbicide injury in CHEAL · Regression modeling failed to capture CHEAL's biphasic dose-response to glyphosate · Random forest minimized model residual and predictive errors Hyperspectral sensors are increasingly used to develop optimized vegetation indices (VIs) that capture plant spectral response to specific stressors. The project goal was to develop quantitative metrics for characterization of weed response to herbicide application. This work applied hyperspectral sensing to describe and predict the spectral response of common lambsquarters ( Chenopodium album L., CHEAL) to glyphosate application. Thirteen treatments, including one glyphosate rate used alone or in combination with eleven adjuvants plus one nontreated control, were applied to CHEAL seedlings cultivated in a greenhouse. Visible injury ratings and non-imaging hyperspectral data were collected 14 days after treatment application. Sensor data processing included cleaning, normalization, smoothing, and spectral reduction. The treatments resulted in a significant (P < 0.001) gradient of injury ranging from 0 to 98%, with visible differences in leaf spectral signatures. Thirty-one key wavelengths were identified using principal component analysis, relief-f feature selection, and Bayesian discriminant analysis and used to create 45,732 VIs. No single VI accurately described CHEAL injury (minimum mean absolute error (MAE) = 14.0%). A random forest algorithm developed using four VIs adequately described CHEAL injury with an MAE of 7.7%. Post-calibration was not needed to improve the random forest model performance (P ≥ 0.05). Therefore, hyperspectral sensing could be used to quantify weed response to herbicide application and overcome the limitations of visual methods current in use. Further development of this method and validation will allow development of a platform for high-throughput phenotyping of weed response to herbicide application and screening for herbicide resistance.

Plant phenotyping relevance

薬剤誘発ストレスによる植物傷害を、ハイパースペクトルセンシングと機械学習で定量化する指標・予測手法を開発・評価しており、表現型取得法が研究の中心である。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.