New multi-angle drone model improves accuracy of wheat nitrogen assessment

By accounting for vertical differences in leaf nitrogen content using multispectral drone data and multi-angle observations, this model—especially in its random forest configuration—achieves up to 27% better accuracy than traditional methods. This breakthrough offers a promising tool for precision agriculture, enabling real-time and reliable nitrogen assessment across wheat fields.

Nitrogen is a critical nutrient for crop growth, influencing yield and grain quality. Yet, both excess and deficiency of nitrogen can hinder crop performance. Leaf nitrogen accumulation (LNA)—a product of leaf dry mass and nitrogen concentration—is widely regarded as an indicator of overall crop nitrogen status and photosynthetic capacity. Remote sensing technologies have enabled efficient, nondestructive monitoring of LNA using vegetation indices (VIs). However, conventional methods often overlook the vertical heterogeneity of nitrogen distribution within the crop canopy, potentially compromising estimation accuracy. To address this limitation, more advanced strategies incorporating machine learning and multi-angle sensing are being explored.

A study (DOI: 10.34133/plantphenomics.0276) published in Plant Phenomics on 5 December 2024 by Yongchao Tian’s team, Nanjing Agricultural University, demonstrates the benefits of stratified modeling and machine learning in refining crop nitrogen estimation, especially under varying environmental and nitrogen application conditions.

In this study, researchers employed a novel remote sensing strategy to estimate wheat leaf nitrogen accumulation (LNA) by incorporating vertical heterogeneity and multi-angle spectral data. Wheat canopies were stratified into distinct leaf layers, and agronomic parameters such as leaf nitrogen concentration (LNC), dry weight (LDW), and LNA were quantified across five layers throughout the growth stages. Spectral reflectance data were collected using unmanned aerial vehicle (UAV) multispectral sensors and near-ground hyperspectral sensors at three view zenith angles (0°, −30°, −45°). Researchers then established linear regression (LR) and random forest (RF) models using vegetation indices (VIs) optimized for each leaf layer and angle, such as EVI, RERI730, and NDRE. The results revealed that LNC, LDW, and LNA exhibited distinct vertical and temporal patterns, with values peaking at different growth stages depending on nitrogen levels. The best-performing estimation model—termed LNASum—combined data from the top three leaf layers, achieving the highest accuracy (RRMSE = 19.3%) using LR and further improved performance (RRMSE = 17.8%) with RF. The most suitable observation angles and VIs differed by layer: the upper layer was best captured at nadir using EVI, while the middle and lower layers required oblique angles (−30° and −45°) with red-edge indices for accurate LNA estimation. Compared to unstratified models (LNAnon), those incorporating vertical heterogeneity (LNASum) consistently demonstrated superior accuracy across both measured and simulated UAV datasets. Notably, the RF-based LNASum model achieved the highest correlation (R² = 0.88) and lowest error (RRMSE = 21.4%) at field scale. These findings confirm that incorporating vertical structure and multi-angle observations significantly enhances the precision of nitrogen monitoring in wheat, with strong potential for application in precision agriculture.

This study offers a powerful tool for precision agriculture, enhancing real-time decision-making in nitrogen fertilization. By accounting for the vertical structure of the crop canopy, the LNASum model enables more accurate detection of nitrogen stress, particularly in the less visible middle and lower leaves. These insights can lead to optimized fertilizer use, reduced environmental impact, and improved crop health. The successful application of UAV-based multispectral sensing and machine learning also paves the way for large-scale adoption of smart farming technologies.

###

References

DOI

10.34133/plantphenomics.0276

Original Source URL

https://doi.org/10.34133/plantphenomics.0276

Funding information

This work was supported by the National Natural Science Foundation of China (No. 32371990), the Jiangsu Provincial Key Research and Development Program (BE2023368), the Postdoctoral Fellowship Program of CPSF under Grant Number (GZC20240723), the Jiangsu Funding Program for Excellent Postdoctoral Talent, the Fundamental Research Funds for the Central Universities, and the Shandong Province Key Technologies R&D Program (2022SFGC0203).

About Plant Phenomics

Science Partner Journal Plant Phenomics is an online-only Open Access journal published in affiliation with the State Key Laboratory of Crop Genetics & Germplasm Enhancement, Nanjing Agricultural University (NAU) and distributed by the American Association for the Advancement of Science (AAAS). Like all partners participating in the Science Partner Journal program, Plant Phenomics is editorially independent from the Science family of journals. Editorial decisions and scientific activities pursued by the journal's Editorial Board are made independently, based on scientific merit and adhering to the highest standards for accurate and ethical promotion of science. These decisions and activities are in no way influenced by the financial support of NAU, NAU administration, or any other institutions and sponsors. The Editorial Board is solely responsible for all content published in the journal. To learn more about the Science Partner Journal program, visit the SPJ program homepage.