Zhu et al. (2026) Enhanced leaf area index estimation in the Drip-Irrigated kiwifruit orchard based on optimized multi-source UAV-based indices

Identification

• Journal: Computers and Electronics in Agriculture
• Year: 2026
• Date: 2026-04-03
• Authors: Shidan Zhu, NingBo Cui, Li Guo, Jiujiang Wu, Zhihui Wang, Shouzheng Jiang, Liwen Xing, Zongjun Wu, Fei Chen, Quanshan Liu
• DOI: 10.1016/j.compag.2026.111695

Research Groups

• College of Water Resource and Hydropower, Sichuan University, Chengdu, China
• Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu, China

Short Summary

This study developed optimized multi-source UAV-based spectral, spectral-texture, and thermal infrared-texture indices combined with machine learning models to enhance Leaf Area Index (LAI) estimation in drip-irrigated kiwifruit orchards across different growth stages. The optimized spectral-texture index (VI_MT) coupled with Random Forest Regression (RFR) significantly improved LAI prediction accuracy, especially for stage-specific estimations.

Objective

• To develop and evaluate optimized multi-source UAV-based indices (spectral, spectral-texture, and thermal infrared-texture) combined with machine learning regression models for accurate and timely Leaf Area Index (LAI) estimation in kiwifruit orchards across different growth stages.

Study Configuration

• Spatial Scale: Drip-irrigated kiwifruit orchard; centimetre-level imagery resolution.
• Temporal Scale: Across different kiwifruit growth stages; full growth cycle estimation.

Methodology and Data

• Models used: Support Vector Regression (SVR), Random Forest Regression (RFR), and four Boosting-based algorithms.
• Data sources: UAV-acquired multispectral data, thermal infrared data, and derived texture features.

Main Results

• Optimized indices (VIMS, VIMT, VITT) strengthened correlation with LAI across growth stages, with the spectral-texture index (VIMT) performing best.
• Random Forest Regression (RFR) showed the highest prediction accuracy among the machine learning models, followed by Boosting-based algorithms, while SVR performed worst.
• Combining the optimized spectral-texture index (VI_MT) with RFR or Boosting models significantly improved kiwifruit LAI estimation accuracy.
• Stage-specific LAI estimation outperformed full growth cycle estimation, with mean R² values of 0.916 ± 0.036 and 0.893 ± 0.083, and mean Mean Absolute Error (MAE) values of 0.039 ± 0.008 and 0.043 ± 0.017, respectively.

Contributions

• Development of novel optimized multi-source UAV-based spectral-texture and thermal infrared-texture indices for LAI estimation in perennial fruit trees.
• Comprehensive evaluation of various machine learning regression models for LAI estimation using these optimized indices.
• Demonstration of improved kiwifruit LAI estimation accuracy by coupling optimized VI_MT with RFR or Boosting models, providing new insights for dynamic orchard canopy monitoring.
• Highlighted the superior performance of stage-specific LAI estimation compared to full growth cycle estimation.

Funding

• Not available in the provided paper text.

Citation

@article{Zhu2026Enhanced,
  author = {Zhu, Shidan and Cui, NingBo and Guo, Li and Wu, Jiujiang and Wang, Zhihui and Jiang, Shouzheng and Xing, Liwen and Wu, Zongjun and Chen, Fei and Liu, Quanshan},
  title = {Enhanced leaf area index estimation in the Drip-Irrigated kiwifruit orchard based on optimized multi-source UAV-based indices},
  journal = {Computers and Electronics in Agriculture},
  year = {2026},
  doi = {10.1016/j.compag.2026.111695},
  url = {https://doi.org/10.1016/j.compag.2026.111695}
}