Prey Lukas, Hu Yuncai, Schmidhalter Urs
Chair of Plant Nutrition, Technical University of Munich, Munich, Germany.
Front Plant Sci. 2020 Jan 17;10:1672. doi: 10.3389/fpls.2019.01672. eCollection 2019.
High-throughput, non-invasive phenotyping is promising for evaluating crop nitrogen (N) use efficiency (NUE) and grain yield (GY) formation under field conditions, but its application for genotypes differing in morphology and phenology is still rarely addressed. This study therefore evaluates the spectral estimation of various dry matter (DM) and N traits, related to GY and grain N uptake (Nup) in high-yielding winter wheat breeding lines. From 2015 to 2017, hyperspectral canopy measurements were acquired on 26 measurement dates during vegetative and reproductive growth, and 48 vegetation indices from the visible (VIS), red edge (RE) and near-infrared (NIR) spectrum were tested in linear regression for assessing the influence of measurement stage and index selection. For most traits including GY and grain Nup, measurements at milk ripeness were the most reliable. Coefficients of determination (²) were generally higher for traits related to maturity than for those related to anthesis canopy status. For GY (² = 0.26-0.51 in the three years, < 0.001), and most DM traits, indices related to the water absorption band at 970 nm provided better relationships than the NIR/VIS indices, including the normalized difference vegetation index (NDVI), and the VIS indices. In addition, most indices including RE bands, notably NIR/RE combinations, ranked above the NIR/VIS group. Due to index saturation, the index differentiation was most apparent in the highest-yielding year. For grain Nup and total Nup, the RE/VIS index MSR_705_445 and the simple ratio R780_R740 ranked highest, followed by other RE indices. Among the vegetative organs, ² values were mostly highest and lowest for leaf and spike traits, respectively. For each trait, index and partial least squares regression (PLSR) models were validated across years at milk ripeness, confirming the suitability of optimized index selection. PLSR improved the prediction errors of some traits but not consistently the R² values. The results suggest the use of sensor-based phenotyping as a useful support tool for screening of yield potential and NUE and for identifying contributing plant traits-which, due to their expensive and cumbersome destructive determination are otherwise not readily available. Water band and RE indices should be preferred over NIR/VIS indices for DM traits and N-related traits, respectively, and milk ripeness is suggested as the most reliable stage.
高通量、非侵入性表型分析对于在田间条件下评估作物氮（N）利用效率（NUE）和籽粒产量（GY）形成很有前景，但它在形态和物候不同的基因型中的应用仍很少被涉及。因此，本研究评估了与高产冬小麦育种系的GY和籽粒氮吸收（Nup）相关的各种干物质（DM）和N性状的光谱估计。2015年至2017年，在营养生长和生殖生长期间的26个测量日期进行了高光谱冠层测量，并对来自可见光（VIS）、红边（RE）和近红外（NIR）光谱的48种植被指数进行了线性回归测试，以评估测量阶段和指数选择的影响。对于包括GY和籽粒Nup在内的大多数性状，乳熟期的测量最为可靠。与成熟相关的性状的决定系数（²）通常高于与开花期冠层状态相关的性状。对于GY（三年中² = 0.26 - 0.51，< 0.001）和大多数DM性状，与970 nm处吸水带相关的指数比包括归一化差异植被指数（NDVI）在内的NIR/VIS指数以及VIS指数提供了更好的关系。此外，包括RE波段在内的大多数指数，特别是NIR/RE组合，排名高于NIR/VIS组。由于指数饱和，指数差异在产量最高的年份最为明显。对于籽粒Nup和总Nup，RE/VIS指数MSR_705_445和简单比值R780_R740排名最高，其次是其他RE指数。在营养器官中，叶和穗性状的²值大多分别最高和最低。对于每个性状、指数和偏最小二乘回归（PLSR）模型，在乳熟期进行了多年验证，证实了优化指数选择的适用性。PLSR改善了一些性状的预测误差，但并非始终提高R²值。结果表明，基于传感器的表型分析可作为筛选产量潜力和NUE以及识别有贡献的植物性状的有用支持工具——由于这些性状的破坏性测定昂贵且繁琐，否则难以获得。对于DM性状和与N相关的性状，应分别优先选择水波段和RE指数而非NIR/VIS指数，并且建议乳熟期是最可靠的阶段。
