Liu et al. (2026) Leaf water content-guided multi-task learning for multi-depth soil moisture estimation using multisource UAV remote sensing data

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-06-29
• Authors: Quanshan Liu, Shunsheng Zheng, Ningbo Cui, Zhihui Wang, Daozhi Gong, L N Zhao, Fang Chen
• DOI: 10.1016/j.agwat.2026.110575

Research Groups

• State Key Laboratory of Hydraulics and Mountain River Engineering & College of Water Resource and Hydropower, Sichuan University, Chengdu, 610065, China
• Xiamen University of Technology, Xiamen, 361024, China
• Chinese Academy of Agricultural Sciences Institute of Environment and Sustainable Development in Agriculture, Beijing 100081, China

Short Summary

This study developed a multi-task learning (MTL) model that incorporates leaf water content (LWC) as an auxiliary physiological task to improve soil moisture retrieval across varying depths. The results showed that LWC was significantly correlated with soil moisture at 10, 20, and 40 cm, but not at 5 cm.

Objective

• To examine the relationships between LWC and soil moisture at different depths
• To compare the performance of Random Forest (RF), Single-task Learning (STL), and LWC-guided MTL models for multi-depth SM estimation

Study Configuration

• Spatial Scale: Orchard scale, with a total area of 6.07 ha
• Temporal Scale: Seven dates in 2022, with field measurements conducted synchronously with UAV image acquisition

Methodology and Data

• Models used:
  • Random Forest (RF)
  • Single-task Learning (STL) model
  • LWC-guided multi-task learning (MTL) model
• Data sources:
  • Multispectral imagery from a DJI Phantom 4 Multispectral
  • Thermal infrared imagery from a DJI Matrice 300 RTK equipped with a Zenmuse H20T sensor
  • Field measurements of soil moisture and LWC

Main Results

• The MTL model achieved the highest overall testing performance at subsurface depths, with R2 ranging from 0.432 to 0.695.
• The MTL model consistently outperformed RF and STL at 10, 20, and 40 cm depths.

Contributions

• This study provides a more physiologically informed and interpretable framework for depth-resolved soil moisture estimation in citrus orchards.
• The results demonstrate the potential of incorporating LWC as an auxiliary physiological task to improve SM retrieval across varying depths.

Funding

• Not specified

Citation

@article{Liu2026Leaf,
  author = {Liu, Quanshan and Zheng, Shunsheng and Cui, Ningbo and Wang, Zhihui and Gong, Daozhi and Zhao, L N and Chen, Fang},
  title = {Leaf water content-guided multi-task learning for multi-depth soil moisture estimation using multisource UAV remote sensing data},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110575},
  url = {https://doi.org/10.1016/j.agwat.2026.110575}
}