Fruit quality formation is regulated by the balance between assimilate supply from leaves and demand from developing fruits. Excessive fruit load intensifies competition for assimilates, reducing fruit size and quality, whereas insufficient load improves individual fruit attributes at the cost of overall yield. The leaf-to-fruit ratio (LFR) method is biologically robust, but its application requires accurate whole-canopy leaf number estimation, which remains technically challenging in mature pear trees because of dense foliage and complex crown architecture. Traditional leaf counting and point cloud-based leaf segmentation are labor-intensive and sensitive to scanning quality. Given these challenges, there is a need to develop a simpler, more efficient proxy for estimating total leaf number under field conditions.
Researchers from Nanjing Agricultural University, the University of Tokyo, and Xinjiang Agricultural University published (DOI: 10.1016/j.plaphe.2026.100225) the study in Plant Phenomics on 2026. The team integrated terrestrial laser scanning (TLS)-derived three-dimensional (3D) point cloud analysis with machine learning to estimate leaf number and determine reasonable fruit load in pear trees. The study was conducted across four growing seasons (2022–2025) using three pear cultivars: 'Cuiguan' (CG), 'Cuiyu' (CY), and 'Xialu' (XL).
The team defined 44 pear tree 3D traits and extracted them using a standalone software tool called FTPCT, which provides a graphical user interface for standardized trait acquisition. After correlation analysis, multicollinearity diagnosis, and variance inflation factor (VIF) screening, five key traits were retained: East-West crown diameter, South-North crown diameter, tree height, tree crown volume, and tree crown surface area. Five machine learning models were optimized using Bayesian optimization (BO): partial least squares regression (PLSR), elastic net regression (ENR), XGBoost regression (XGBR), random forest regression (RFR), and support vector machine regression (SVR). The optimized random forest regression model (RFR-BO) achieved the best balance of accuracy and stability, with an R² of 0.85, root mean square error (RMSE) of 239.74, and mean absolute error (MAE) of 149.26 on the test set. SHAP (Shapley Additive explanations) analysis identified tree crown volume as the dominant contributor to leaf number estimation, with contribution rates of 79.87%, 95.55%, and 60.95% for XGBR-BO, RFR-BO, and SVR-BO models, respectively. The approach avoids explicit leaf-level reconstruction and relies on a small number of canopy-scale traits.
The authors said the key advance is that leaf number can be reliably estimated from canopy structure alone, without ultra-high-resolution point clouds or exhaustive leaf segmentation. They said tree crown volume integrates the cumulative effects of shoot growth, branching, and leaf expansion, making it a robust predictor of whole-tree leaf abundance. They added that the framework substantially reduces data requirements and computational cost, enabling rapid, non-destructive estimation at the whole-tree scale, which is critical for timely thinning decisions within the narrow operational window. They also noted that orchard management practices and tree age may affect crown volume, and future work should test the model across more diverse orchard systems.
Field validation in 2024 and 2025 compared conventional thinning (T1) with model-guided thinning (T2). Yield did not differ significantly between treatments, but T2 trees produced fruit with significantly greater weight, horizontal diameter, and longitudinal diameter, along with lower acid content. The framework offers an objective, scalable solution for data-driven fruit load regulation, supporting consistent production of high-quality pears. Its reliance on canopy-scale traits makes it well suited for large orchards and time-sensitive thinning operations. Future integration with unmanned aerial vehicle (UAV)-based LiDAR or multi-view photogrammetry could further improve scalability and cost-effectiveness.
###
References
DOI
10.1016/j.plaphe.2026.100225
Original Source URL
https://doi.org/10.1016/j.plaphe.2026.100225
Funding information
This work was supported by the Xinjiang Uygur Autonomous Region Major Scientific and Technological Project (2024A02006), Basic Research Program of Jiangsu Province(BK20251547), China Postdoctoral Science Foundation (2025M773747), and China Agriculture Research System-Pear (CARS-28).
About Plant Phenomics
Plant Phenomics publishes breakthrough research that integrates genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics and computer science to advance plant phenotyping—from the cellular level to entire populations. The journal tackles key scientific challenges in phenomics by showcasing cutting‑edge technologies for data acquisition, management, interpretation and modelling. Its scope spans practical applications across all plant sciences, aiming to translate novel phenotyping tools into real‑world solutions for breeding, crop management and fundamental plant biology. By fostering interdisciplinary innovation, Plant Phenomics serves as a hub for researchers developing high‑throughput, accurate and scalable methods to decipher complex plant traits, ultimately accelerating progress in agriculture and plant science.