Estimating photosynthetic characteristics of forage rape by fusing the sensitive spectral bands to combined stresses of nitrogen and salt.

Abstract

Leaf gas exchange and chlorophyll fluorescence parameters (PGE-CFPs), which respond significantly and quickly to environmental stresses, have been used to assess the early responses of crop physiology to stresses. Most spectral estimations only focus on crop photosynthetic characteristics under a single environmental stress. Thus, the methods proposed previously are not suitable for the estimations under combined stresses (i.e., nitrogen and salt). In this research, the leaf spectral features of forage rape ( Brassica napus L.) under nitrogen stress (NSpe) and salt stress (SSpe) were fused to increase the accuracy of the spectral estimation of photosynthetic characteristics of forage rape under combined stresses in arid region of Xinjiang, China. The results showed that PGE-CFPs' spectral features were extracted with SPA (successive projections algorithm) after preprocessing. Among the SSpe- and NSpe-based models, the RF (random forest) models had higher estimation accuracy than the PLSR (partial least squares regression) and BPNN (backpropagation neural network) models. Specifically, the RF models had a PGE-CFPs estimation accuracy of 0.597-0.712, 0.640-0.715, and 0.377-0.461 under nitrogen stress (NS), salt stress (SS), and NS*SS, respectively. After fusing NSpe and SSpe, the accuracy in estimating PGE-CFPs of forage rape under NS, SS, and NS*SS were 0.729-0.755, 0.667-0.768, and 0.621-0.689, respectively. Then, the constructed models were further validated using field data, and the accuracy obtained was in the range of 0.585-0.711. Therefore, the feature fusion modeling method proposed has strong transferability and applicability. This research will offer a technical reference for crop photosynthesis monitoring at the early stage of environmental stresses.

Plant phenotyping relevance

分光特徴量の融合と機械学習により、複合ストレス下の植物光合成特性を推定する手法を開発・検証しており、表現型取得・推定が研究の中心である。

Code and data availability

The supplied blocks describe hyperspectral and PGE-CFP measurements and PLSR/RF/BPNN modeling, but contain no public dataset, image, code, or model deposit. No data availability statement text is present, and no authors' public URL for assets is given.

No evidence-backed public reproduction asset is currently recorded.