Estimation of Potato Leaf Area Index and Aboveground Biomass Based on a New Texture Index Constructed from Unmanned Aerial Vehicles Multispectral Images

• Published in: Journal of soil science and plant nutrition Vol. 25; no. 3; pp. 7092 - 7107
• Main Authors: Li, Siqi, Xiang, Youzhen, Jin, Ming, Tang, Zijun, Sun, Tao, Liu, Xiaochi, Huang, Xiangyang, Li, Zhijun, Zhang, Fucang
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.09.2025; Springer Nature B.V
• Subjects: Accuracy; Agriculture; Biomass; Biomedical and Life Sciences; Cameras; Crop growth; Crop management; Crop yield; Crops; Data collection; Design of experiments; Ecology; Environment; Errors; Fertilizers; Image acquisition; Leaf area; Leaf area index; Least squares method; Leaves; Life Sciences; Machine learning; Original Paper; Phenotyping; Physiology; Plant Sciences; Potassium; Potatoes; Precipitation; Prediction models; Remote monitoring; Remote sensing; Soil Science & Conservation; Temperature; Texture; Unmanned aerial vehicles; Vegetation; Vegetation index; Potato; Texture index; Multispectral; Machine learning; Aboveground biomass; Leaf area index
• ISSN: 0718-9508; 0718-9516
• DOI: 10.1007/s42729-025-02582-x

Timely acquisition of the leaf Area Index (LAI) and Aboveground Biomass (AGB) is essential for accurately assessing crop growth and yield. This study focuses on the remote sensing monitoring of potato LAI and AGB using spectral parameters derived from multispectral imagery captured by unmanned aerial vehicles (UAVs). Potatoes in the tuber formation stage serve as the research subject, and machine learning techniques are applied to Reinforce the precision of LAI and AGB estimation. A novel three-dimensional texture indices (TTIs) was developed by incorporating crop dimensional information. Prediction models for LAI and AGB were constructed using Partial Least Squares Regression (PLSR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), integrating texture features (TFs), vegetation indices (VIs), and texture indices (TIs). The results indicated that TTIs improves estimation accuracy compared to using TFs, VIs, or TIs individually. The highest level of predictive accuracy was achieved by combining VIs, TFs, and TTIs, with the XGBoost model demonstrating superior performance. Under the optimal input combination, the coefficient of determination (R 2 ) for LAI and AGB reached 0.868 and 0.855, respectively. The root mean square errors (RMSE) were 0.135 and 0.322 kg hm⁻ 2 , while the mean relative errors (MRE) were 7.578% and 6.540%, respectively. This study offers innovative approaches for UAV-based multispectral monitoring of potato LAI and AGB. Furthermore, it offers valuable insights into the practical implementation of UAV-based high-throughput phenotyping technologies for precise crop management on a large scale.