Jia et al. (2025) Non-Invasive Inversion and Characteristic Analysis of Soil Moisture in 0–300 cm Agricultural Soil Layers

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Agriculture
• Year: 2025
• Date: 2025-10-15
• Authors: Shujie Jia, Yaoyu Li, Bing Cao, Yingchun Cheng, Abdul Sattar Mashori, Zhongrui Bai, Maxwell Cui, Zhimin Zhang, Linqiang Deng, Wuping Zhang
• DOI: 10.3390/agriculture15202143

Research Groups

Not explicitly stated in the provided text.

Short Summary

This study systematically benchmarks eight regression algorithms to non-invasively infer deep soil moisture (20–300 cm) using surface soil moisture and meteorological variables. It finds that non-linear models, particularly Multi-Layer Perceptron (MLP), consistently outperform linear models for deep layers, and proposes a depth-adaptive modeling strategy for practical application.

Objective

• To systematically benchmark various regression algorithms for non-invasively inferring deep soil moisture (20–300 cm) using easily accessible surface soil moisture (0–20 cm) and meteorological variables.

Study Configuration

• Spatial Scale: Site-specific (a typical agricultural site in Wenxi, Shanxi, China)
• Temporal Scale: Three years (2020–2022)

Methodology and Data

• Models used: Linear regression, Lasso, Ridge, Elastic Net, Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), Random Forest (RF), Gradient Boosting Trees (GBDT).
• Data sources: In-situ observations of 0–20 cm surface soil moisture and ten meteorological variables.

Main Results

• Non-linear ensemble models significantly outperform linear baselines across all depths.
• Ridge Regression achieved the highest accuracy for 0–20 cm soil moisture prediction.
• Support Vector Regression (SVR) performed best for the 20–40 cm depth.
• Multi-Layer Perceptron (MLP) consistently showed optimal performance for deep layers from 60 cm to 300 cm (R² = 0.895–0.978, KGE = 0.826–0.985).
• Ensemble models like Random Forest (RF) and Gradient Boosting Trees (GBDT) exhibited strong fitting but limited generalization under temporal validation.
• Surface soil moisture is the dominant predictor across all depths, showing a clear attenuation trend and a critical transition zone between 160 cm and 200 cm.
• Precipitation and humidity primarily drive soil moisture in shallow to mid-layers (20–140 cm), while temperature variables gain relative importance in deeper profiles (200–300 cm).
• A depth-adaptive modeling strategy, assigning the best-performing model to each soil layer, is proposed for practical deep soil moisture prediction.

Contributions

• Systematic benchmarking of eight diverse regression algorithms for non-invasive deep soil moisture profiling.
• Identification of optimal model types for different soil depth ranges (Ridge for 0–20 cm, SVR for 20–40 cm, MLP for 60–300 cm).
• Comprehensive interpretability analysis using SHAP, PDP, and ALE to elucidate the dominant predictors and the mechanisms of surface-to-deep information transmission, including a critical transition zone.
• Proposal of a practical depth-adaptive modeling strategy to enhance the accuracy and utility of deep soil moisture prediction for precision irrigation and water resource management.

Funding

Not explicitly stated in the provided text.

Citation

@article{Jia2025NonInvasive,
  author = {Jia, Shujie and Li, Yaoyu and Cao, Bing and Cheng, Yingchun and Mashori, Abdul Sattar and Bai, Zhongrui and Cui, Maxwell and Zhang, Zhimin and Deng, Linqiang and Zhang, Wuping},
  title = {Non-Invasive Inversion and Characteristic Analysis of Soil Moisture in 0–300 cm Agricultural Soil Layers},
  journal = {Agriculture},
  year = {2025},
  doi = {10.3390/agriculture15202143},
  url = {https://doi.org/10.3390/agriculture15202143}
}