Using time-series data collected across two growing seasons, the study accurately reconstructed canopy growth trajectories and identified key intermediate traits—particularly mid-season leaf area index (LAI) dynamics—that strongly predict yield performance under high planting density.
As global demand for food continues to rise, developing soybean varieties that thrive under dense planting is critical for achieving sustainable productivity. However, traditional field phenotyping methods are limited by low temporal resolution and discontinuous modeling, which fail to capture dynamic canopy development and yield stability. Existing machine learning models typically ignore temporal dependencies in crop growth, leading to poor biological interpretability. To address these challenges, scientists are exploring UAV-based phenotyping and time-series deep learning to quantify canopy traits such as LAI, plant height (PH), and canopy cover (CC). Yet, a comprehensive framework integrating temporal modeling with physiological interpretation for dense planting has remained elusive.
A study (DOI: 10.1016/j.plaphe.2025.100083) published in Plant Phenomics on 24 June 2025 by Yuntao Ma's team, China Agricultural University, offers a powerful tool for accelerating breeding of density-tolerant soybean varieties and advancing precision agriculture.
The study employed an integrated methodology combining unmanned aerial vehicle (UAV)–based high-throughput phenotyping, spatiotemporal deep learning, and dynamic modeling to assess soybean tolerance to high planting densities. Researchers conducted two-year field trials (2022–2023) in Heilongjiang, China, testing 208 soybean genotypes under high (50 × 10⁴ plants ha⁻¹) and low (30 × 10⁴ plants ha⁻¹) density treatments. Multispectral and RGB images were captured 15–18 times per season using a DJI Mavic 3M UAV. Ground-truth data for yield, leaf area index (LAI), plant height (PH), and canopy cover (CC) were collected to train and validate four predictive models—Random Forest (RF), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Spatiotemporal Residual Network (ST-ResNet). The ST-ResNet model achieved the highest accuracy (R² = 0.90; RMSE = 0.23 m² m⁻²), capturing continuous canopy growth through spatiotemporal feature fusion. The resulting LAI time-series data, along with UAV-derived PH and CC, were fitted using P-spline dynamic modeling to generate smooth growth curves and extract 15 intermediate traits that quantified canopy development rates at different stages. Mixed-effect models adjusted for genotype, density, and year effects, while correlation and SHAP analyses linked these intermediate traits to yield stability under dense planting. Results showed that UAV-based PH estimates aligned closely with field measurements (R² = 0.90; RMSE = 0.05 m), confirming the reliability of the remote sensing method. Canopy cover dynamics revealed significant varietal variation between 28 and 55 days after emergence, indicating genotypic differences in early vigor. Among all extracted traits, mid-season LAI (ΔMeanLAI-mid) exhibited the strongest correlation with yield response to density (r = 0.51), identifying it as the key indicator of density tolerance. Early LAI increase (ΔLAItPH15+14d) and canopy persistence at maturity (ΔMeanCC-maturity) also contributed to higher yields. This integrated UAV-deep-learning-dynamic-modeling framework accurately quantified canopy growth patterns, provided interpretable physiological indicators, and established a high-efficiency, scalable platform for breeding soybean varieties resilient to dense planting.
This study demonstrates how integrating UAV-based time-series phenotyping with deep learning and dynamic modeling enables high-precision, interpretable quantification of soybean growth under dense planting. By identifying stage-specific canopy traits linked to yield stability, the framework provides a valuable decision-support tool for breeders and agronomists aiming to develop high-yield, space-efficient cultivars. Moreover, applying this framework to other crops such as maize or wheat could enhance understanding of density-related physiological mechanisms and inform adaptive breeding strategies under climate and resource constraints.
