Katipoğlu et al. (2025) Prediction of soil moisture via feature selection, model optimization, and climate data integration

Identification

• Journal: Physics and Chemistry of the Earth Parts A/B/C
• Year: 2025
• Date: 2025-12-17
• Authors: Okan Mert Katipoğlu, Veysi Kartal, Muhammed Ernur Akıner, Zeynep Özge Terzioğlu, Zeyneb Kılıç
• DOI: 10.1016/j.pce.2025.104250

Research Groups

• Department of Civil Engineering, Erzincan Binali Yıldırım University, Türkiye.
• Civil Engineering Department, Siirt University, Türkiye.
• Department of Environmental Protection Technologies, Akdeniz University, Türkiye.
• Civil Engineering Department, Adıyaman University, Türkiye.

Short Summary

This study evaluates the performance of four machine learning algorithms in predicting soil moisture across the Konya Closed Basin, Türkiye, using long-term climate data from 1950 to 2022. The results identify Deep Neural Networks (DNN) as the most accurate model for estimating soil moisture dynamics in the region.

Objective

• To develop and compare the effectiveness of different artificial intelligence models (DNN, AdaBoost, ESN, and LGBM) for the long-term estimation of soil moisture using climate-based input variables and feature selection.

Study Configuration

• Spatial Scale: Regional; focused on six stations within the Konya Closed Basin, Türkiye.
• Temporal Scale: Long-term historical analysis covering the period from 1950 to 2022.

Methodology and Data

• Models used: Deep Neural Networks (DNN), Adaptive Boosting (AdaBoost), Echo State Networks (ESN), and Light Gradient Boosting Machine (LGBM).
• Data sources: Reanalysis climate data and calculated potential evapotranspiration.
• Input Variables: Precipitation (PR), solar radiation (RSDS), wind speed (WS), mean temperature (Tave), maximum temperature (Tmax), minimum temperature (Tmin), and potential evapotranspiration (ET0) calculated via the Thornthwaite equation.

Main Results

• The DNN model demonstrated superior performance, achieving $R^2$ values ranging from 0.87 to 0.97 across the study sites.
• AdaBoost and ESN were identified as the second most successful models, showing reliable predictive capabilities.
• The LGBM model exhibited the lowest accuracy and the highest deviations compared to the other three algorithms.
• Feature selection successfully identified the most critical climatic parameters, enhancing the efficiency of the soil moisture estimation process.

Contributions

• Provides a comprehensive comparative assessment of modern machine learning techniques for soil moisture prediction in a semi-arid basin.
• Demonstrates the utility of integrating long-term reanalysis climate data with feature selection to improve hydrological modeling.
• Establishes a scalable framework for soil moisture estimation that can support water resource management and agricultural planning in data-scarce or complex climatic regions.

Funding

• Not specified in the provided text.

Citation

@article{Katipoğlu2025Prediction,
  author = {Katipoğlu, Okan Mert and Kartal, Veysi and Akıner, Muhammed Ernur and Terzioğlu, Zeynep Özge and Kılıç, Zeyneb},
  title = {Prediction of soil moisture via feature selection, model optimization, and climate data integration},
  journal = {Physics and Chemistry of the Earth Parts A/B/C},
  year = {2025},
  doi = {10.1016/j.pce.2025.104250},
  url = {https://doi.org/10.1016/j.pce.2025.104250}
}

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