Monitoring vertical SPAD distribution of winter wheat using UAV cross-circle oblique photography

Abstract

Timely acquisition of crop chlorophyll contents is essential for effective field management decisions and comprehensive crop nutritional monitoring. Unmanned Aerial Vehicle (UAV) has been extensively utilized for canopy chlorophyll content monitoring. However, prior studies predominantly concentrated on the horizontal variability of SPAD, with few researches dedicated to monitoring the vertical distribution of SPAD. This study aimed to develop a high-resolution vertical SPAD distribution model for winter wheat by integrating UAV-based Cross-Circle Oblique (CCO) photography with the Structure from Motion (SFM) and Multi-View Stereo (MVS) methods. The canopy was divided into upper, middle, and lower layers based on plant height. Nadir and CCO photography were used to capture images, with Nadir photography constructing the upper canopy SPAD model and CCO data used for the vertical distribution model (including upper, middle and lower layers). Shapley values were applied to evaluate feature importance in different machine learning models (GBR, gradient boosting regression; RF, random forest; SVM, support vector machine; RR, ridge regression). Finally, a vertical SPAD distribution model for winter wheat was created with a 0.1-meter gradient. The results demonstrated that the accuracy of SPAD vertical distribution inversion using single machine learning algorithms (KNN: k-nearest neighbor, RR: ridge regression) was lower than that achieved by ensemble learning methods (stacking, RF: random forest). Ensemble learning enhanced R 2 and RMSE by 0.13 and 1.23 for the training set, and by 0.09 and 0.56 for the test set, respectively. CCO photography exhibited high accuracy in capturing the SPAD spatiotemporal distribution of winter wheat, with R 2 values for the training sets across all three growth stages surpassing 0.8, and R 2 values for the test sets ranging from 0.6 to 0.81. The highest accuracy in SPAD vertical distribution inversion was achieved during the heading stage, with R 2 and RMSE values for the training and test sets of 0.89, 2.58, and 0.80, 3.34, respectively. Therefore, UAV-based CCO photography combined with the SFM-MVS algorithm demonstrates preliminary potential for vertical SPAD phenotyping and precision nutrient management; however, further validation across different growing seasons and wheat varieties is necessary to confirm its broad generalizability.

Plant phenotyping relevance

UAV斜視画像とSfM-MVS、機械学習を統合し、コムギの垂直SPAD分布を推定する方法の開発・検証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.