The study demonstrates that combining LiDAR, multispectral, and thermal infrared imaging with ensemble learning significantly enhances prediction accuracy, supporting smarter crop management.
Aboveground biomass is a vital agroecological metric that reflects crop growth patterns, photosynthetic efficiency, and yield potential. Reliable AGB estimation supports nutrient management, pest control, and productivity forecasting. Traditional destructive sampling methods, while accurate, are time-consuming, costly, and unsuitable for large-scale monitoring. Satellite remote sensing offers broad coverage but is limited by atmospheric interference and coarse resolution. UAV platforms, equipped with advanced sensors, now provide high-resolution, flexible, and frequent monitoring, making them ideal for precision agriculture. Yet, questions remain about the adaptability of different prediction methods under varied environmental conditions and growth stages. Addressing these gaps is essential for achieving efficient, scalable, and affordable biomass forecasting.
A study (DOI: 10.1016/j.plaphe.2025.100068) published in Plant Phenomics on 11 June 2025 by Wei Su’s & Zhen Chen’s team, China Agricultural University & Chinese Academy of Agricultural Sciences, can streamline data collection, reduce costs, and improve the timeliness of agricultural decisions.
The study evaluated two complementary approaches for predicting corn aboveground biomass (AGB) across multiple growth stages and agronomic treatments: (i) machine learning (ML) models driven by multi-sensor data fusion and (ii) the Vegetation Index Weighted Canopy Volume Model (CVMVI). Using UAV-mounted LiDAR, multispectral (MS), and thermal infrared (TIR) sensors, together with field-measured AGB and leaf area index (LAI), the team trained and compared StackingDNN, Random Forest (RF), CatBoost, linear regression (LR), and deep neural networks (DNN). Model performance depended on the input modality: with MS features alone, StackingDNN reached R² = 0.75 (MAE = 1.93 t/ha; RMSE = 2.75 t/ha), while RF performed best on LiDAR features (R² = 0.78; MAE = 1.85 t/ha; RMSE = 2.58 t/ha). Ensemble learning consistently surpassed single algorithms, and full multi-sensor fusion (MS+LiDAR+TIR+LAI) with StackingDNN delivered the highest accuracy (R² = 0.86; MAE = 1.54 t/ha; RMSE = 2.06 t/ha). Feature importance analyses (SHAP) converged on LAI, canopy height metrics (e.g., Z_max, canopy height model), and TIR-derived variables as the dominant predictors; DNN prioritized LAI, whereas RF emphasized height. By contrast, CVMVI was highly effective early in the season (R² ≈ 0.78) but lost fidelity as biomass and canopy complexity increased, indicating a cost-efficient option for early monitoring and a clear handoff to ML fusion in mid-to-late stages. Treatment analyses showed biomass responded strongly to nitrogen during tasseling (TS) and before tasseling (BTS), while irrigation effects were significant only early and waned with adequate rainfall later. Spatial diagnostics using Moran’s I revealed that multi-sensor fusion reduced spatial dependence, strengthening model adaptability across fields; cross-validation under varying irrigation and fertilization regimes confirmed robustness. Collectively, these results demonstrate that stage-aware, ensemble-based fusion improves AGB accuracy, interpretability, and spatial stability, while CVMVI offers a streamlined early-stage alternative that can lower data and computation burdens before transitioning to more powerful ML fusion as canopies develop.
This research provides a scalable framework for stage-specific AGB prediction, supporting precision agriculture, phenotyping, and environmental monitoring. Farmers can apply CVMVI for cost-effective, rapid assessments in early growth, while adopting ML-based fusion models for mid-to-late growth stages where higher accuracy is critical. The ability to detect treatment effects on biomass also aids in optimizing fertilizer and irrigation management, potentially reducing costs and environmental impacts. More broadly, accurate AGB forecasting enhances regional yield predictions, improves food security planning, and supports climate-smart agricultural policies.
###
References
DOI
10.1016/j.plaphe.2025.100068
Original URL
https://doi.org/10.1016/j.plaphe.2025.100068
Funding information
This study is funded by the National Natural Science Foundation of China under the project (No. 42471402) and Beijing Natural Science Foundation (L251053).
About Plant Phenomics
Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.