Scholarlink

Purpose <p>Photosynthesis is a fundamental process in plants that directly contributes to net primary productivity. High-throughput estimation of photosynthesis is crucial for assessing crop growth and predicting yield. This study aims to establish an effective framework for high-throughput estimation of canopy-scale photosynthesis in winter wheat under drought stress by optimizing spectral resolution, screening spectral indices, and comparing machine learning algorithms.</p> Methods <p>Photosynthetic phenotypic parameters and hyperspectral reflectance data of winter wheat at different growth stages under drought stress were collected. Spectral characteristics of the photosynthetic parameters were analyzed to identify the optimal spectral resolution, followed by extraction of hyperspectral bands sensitive to key photosynthetic parameters (canopy chlorophyll content, CCC; net photosynthetic rate, Pn; maximum photochemical efficiency, Fv/Fm). Estimation models for the three parameters were subsequently constructed using partial least squares (PLS), support vector machine (SVM), random forest (RF), and back propagation neural network (BPNN) algorithms, respectively, with their performances evaluated.</p> Results <p>The optimal spectral resolution for estimating photosynthetic phenotypic parameters was determined as 10 nm. At this resolution, CCC was highly sensitive to hyperspectral reflectance in the ranges of 725–835 nm, 995–1095 nm, 1145–1175 nm, and 1245–1295 nm. Pn was associated with bands of 795–945 nm and 1035–1135 nm, while 615–695 nm was identified as the characteristic band for Fv/Fm. Model performance evaluation results indicated that the best model for CCC estimation was BPNN-NDSI-CCC (R<sup>2</sup>= 0.639, RMSE = 0.461, RPD = 1.672); for Pn, BPNN-RSI-Pn (R<sup>2</sup>= 0.883, RMSE = 4.757, RPD = 2.933) performed the best; and the PLS-RSI-Fv/Fm model (R<sup>2</sup>= 0.406, RMSE = 0.028, RPD = 1.286) was significant for Fv/Fm. Compared to baseline models (Raw-PLS), the R<sup>2</sup> values improved by 7.9%, 6.8%, and 9.7%, respectively.</p> Conclusion <p>This study established an integrated framework combining spectral resolution optimization, multi-type spectral index screening, and multi-machine learning algorithm comparison to achieve high-throughput estimation of canopy-scale photosynthesis in winter wheat under drought stress.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hyperspectral estimation on the photosynthetic phenotype of winter wheat under drought stress using machine learning algorithms

  • 期刊论文

  • 发表日期：

    2026年1月31日

查看全文
  • Yanxia Chen,
  • Xiaobin Yan,
  • Tingyue Huo,
  • Yuchao Yang,
  • Bei Tie,
  • Chenhao Qin,
  • Meichen Feng,
  • Xingxing Qiao,
  • Xiaokai Chen,
  • Guangxin Li,
  • Chao Wang,
  • Wude Yang

摘要

Purpose

Photosynthesis is a fundamental process in plants that directly contributes to net primary productivity. High-throughput estimation of photosynthesis is crucial for assessing crop growth and predicting yield. This study aims to establish an effective framework for high-throughput estimation of canopy-scale photosynthesis in winter wheat under drought stress by optimizing spectral resolution, screening spectral indices, and comparing machine learning algorithms.

Methods

Photosynthetic phenotypic parameters and hyperspectral reflectance data of winter wheat at different growth stages under drought stress were collected. Spectral characteristics of the photosynthetic parameters were analyzed to identify the optimal spectral resolution, followed by extraction of hyperspectral bands sensitive to key photosynthetic parameters (canopy chlorophyll content, CCC; net photosynthetic rate, Pn; maximum photochemical efficiency, Fv/Fm). Estimation models for the three parameters were subsequently constructed using partial least squares (PLS), support vector machine (SVM), random forest (RF), and back propagation neural network (BPNN) algorithms, respectively, with their performances evaluated.

Results

The optimal spectral resolution for estimating photosynthetic phenotypic parameters was determined as 10 nm. At this resolution, CCC was highly sensitive to hyperspectral reflectance in the ranges of 725–835 nm, 995–1095 nm, 1145–1175 nm, and 1245–1295 nm. Pn was associated with bands of 795–945 nm and 1035–1135 nm, while 615–695 nm was identified as the characteristic band for Fv/Fm. Model performance evaluation results indicated that the best model for CCC estimation was BPNN-NDSI-CCC (R2= 0.639, RMSE = 0.461, RPD = 1.672); for Pn, BPNN-RSI-Pn (R2= 0.883, RMSE = 4.757, RPD = 2.933) performed the best; and the PLS-RSI-Fv/Fm model (R2= 0.406, RMSE = 0.028, RPD = 1.286) was significant for Fv/Fm. Compared to baseline models (Raw-PLS), the R2 values improved by 7.9%, 6.8%, and 9.7%, respectively.

Conclusion

This study established an integrated framework combining spectral resolution optimization, multi-type spectral index screening, and multi-machine learning algorithm comparison to achieve high-throughput estimation of canopy-scale photosynthesis in winter wheat under drought stress.