Ali (2025) Machine learning approaches for soil moisture prediction: enhancing agricultural water management with integrated data

Identification

• Journal: Modeling Earth Systems and Environment
• Year: 2025
• Date: 2025-11-05
• Authors: Aram Ali
• DOI: 10.1007/s40808-025-02607-5

Research Groups

• Centre for Sustainable Agricultural Systems, University of Southern Queensland, Australia.
• Soil and Water Science Department, Salahaddin University-Erbil, Iraq.
• School of Science, Edith Cowan University, Australia.

Short Summary

This study evaluates the effectiveness of nine machine learning algorithms for predicting soil moisture at two depths in New South Wales, Australia, using integrated climate, soil, and vegetation data. The results demonstrate that ensemble models, particularly Random Forest and XGBoost, significantly outperform traditional linear models, providing a robust framework for precision irrigation management.

Objective

• To develop and optimize high-precision soil moisture prediction models by comparing various machine learning algorithms and identifying the most influential environmental and soil-based predictors.

Study Configuration

• Spatial Scale: Regional scale covering main cropping areas in New South Wales, Australia (Humid Subtropical Climate), utilizing the OzNet Hydrological Monitoring Network.
• Temporal Scale: Long-term analysis spanning 20 years (2000–2020), with a 70/30 split for training (14 years) and validation (6 years).

Methodology and Data

• Models used: Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), Gaussian Process Regression (GPR), Partial Least Squares (PLS), Multiple Linear Regression (MLR), Decision Tree (DT), Cubist Regression, and K-Nearest Neighbour (KNN).
• Data sources:
  • In situ observations: Soil moisture data from OzNet sensors at depths of 0–0.3 m and 0–1.0 m.
  • Meteorological data: SILO-long paddock (daily temperature, vapor pressure, solar radiation, relative humidity, rainfall, and calculated evapotranspiration).
  • Satellite data: Terra MODIS 8-day Leaf Area Index (LAI) (MOD15A2H).
  • Soil properties: Clay, silt, and sand proportions (%), bulk density (g cm⁻³), and soil organic carbon (SOC, %).

Main Results

• Model Performance: Random Forest (RF) and XGBoost consistently provided the highest accuracy. At a depth of 0–0.3 m, RF achieved an NSE of 0.81 and $R^2$ of 0.84. At 0–1.0 m, RF achieved an NSE of 0.80 and $R^2$ of 0.81.
• Ensemble Advantage: Ensemble machine learning models improved the Nash–Sutcliffe Efficiency (NSE) by up to 25% compared to traditional models like MLR and PLS.
• Data Robustness: RF maintained high performance (NSE > 0.75) even when restricted to climate-only input data, suggesting reliability in data-scarce environments.
• Predictor Importance: Feature importance analysis identified Evapotranspiration (ET), Leaf Area Index (LAI), and Soil Organic Carbon (SOC) as the most critical variables for accurate moisture prediction.
• Depth Sensitivity: Prediction accuracy was generally higher at shallower depths (0–0.3 m) due to the higher sensitivity of these layers to immediate climate and vegetation dynamics.

Contributions

• Provides a comprehensive comparative analysis of nine ML algorithms for soil moisture, establishing RF and XGBoost as superior tools for agricultural water management.
• Quantifies the significant predictive value of integrating vegetation indices (LAI) and soil organic matter into moisture models.
• Demonstrates a scalable, non-intrusive approach for operational soil moisture monitoring that can be adapted for precision agriculture in various climatic zones.

Funding

• University of Southern Queensland.
• Queensland Department of Primary Industries.
• Broadacre Cropping Initiative (BACI).

Citation

@article{Ali2025Machine,
  author = {Ali, Aram},
  title = {Machine learning approaches for soil moisture prediction: enhancing agricultural water management with integrated data},
  journal = {Modeling Earth Systems and Environment},
  year = {2025},
  doi = {10.1007/s40808-025-02607-5},
  url = {https://doi.org/10.1007/s40808-025-02607-5}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)