Zubair et al. (2026) Spatial modeling of subsurface soil texture in semi-arid regions: Evaluating pure machine learning against hybrid regression kriging using Sentinel-1 and Sentinel-2 data

Identification

• Journal: Ecological Engineering & Environmental Technology
• Year: 2026
• Date: 2026-09-08
• Authors: Riyad Hazem Zubair, Muntadher Aidi Shareef, Abdelmalek Toumi
• DOI: 10.12912/27197050/235746

Research Groups

• Northern Technical University (NTU)
• Lab-STICC, ENSTA Bretagne

Short Summary

This study evaluates a multi-temporal Digital Soil Mapping framework combining Sentinel-1 and Sentinel-2 data to map subsurface soil texture fractions at 30–40 cm depth in the Tal Kaif district of northern Iraq. The results show that incorporating spatial structure into the modelling framework through Regression Kriging (RK) can provide modest improvements in predictive accuracy, particularly for sand.

Objective

• Evaluate and improve the spatial prediction of subsurface soil texture fractions (30–40 cm depth) in a semi-arid agricultural landscape.
• Assess the limitations of applying pure Machine Learning (ML) models for subsurface soil texture mapping in semi-arid environments with sparse sampling.

Study Configuration

• Spatial Scale: Local to regional scale, focusing on the Tal Kaif district in northern Iraq.
• Temporal Scale: Multi-temporal, combining wet-season Sentinel-1 SAR imagery and dry-season Sentinel-2 optical data.

Methodology and Data

• Models used:
  • Random Forest (RF)
  • Extreme Gradient Boosting (XGBoost)
  • Support Vector Regression (SVR)
  • Artificial Neural Network (ANN/MLP)
  • Regression Kriging (RK)
• Data sources: Sentinel-1 SAR imagery, Sentinel-2 optical data, and field-sampled soil data.

Main Results

• The pure ML models showed limited predictive performance for all three soil fractions.
• The best OOF model for clay was SVR (R² = −0.059; RMSE = 9.73%; RPD = 0.98).
• For sand, XGBoost provided the lowest OOF RMSE (6.89%) and the highest R² among the tested models (R² = −0.065; RPD = 0.98).
• The Hybrid RK results showed modest improvements in predictive accuracy for clay and sand.

Contributions

• This study contributes evidence that integrating environmental feature-space modelling with spatial residual structure can provide a more spatially informed framework for subsurface soil-texture mapping in data-sparse semi-arid environments.
• The results highlight the importance of incorporating spatial structure into the modelling framework through Regression Kriging (RK) to improve predictive accuracy.

Funding

• This research was funded by [project/program name and reference code, if applicable].

Citation

@article{Zubair2026Spatial,
  author = {Zubair, Riyad Hazem and Shareef, Muntadher Aidi and Toumi, Abdelmalek},
  title = {Spatial modeling of subsurface soil texture in semi-arid regions: Evaluating pure machine learning against hybrid regression kriging using Sentinel-1 and Sentinel-2 data},
  journal = {Ecological Engineering & Environmental Technology},
  year = {2026},
  doi = {10.12912/27197050/235746},
  url = {https://doi.org/10.12912/27197050/235746}
}