Gowera et al. (2026) Spatial prediction and mapping of soil salinity using machine learning and remote sensing covariates

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

Identification

• Journal: Canadian Journal of Soil Science
• Year: 2026
• Date: 2026-03-25
• Authors: Grace Tariro Gowera, Preston Sorenson, Angela Bedard-Haughn, Steven J Shirtliffe
• DOI: 10.1139/cjss-2025-0092

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study evaluated remote sensing models for mapping soil salinity in irrigated agroecosystems with predominantly low electrical conductivity values, finding that Support Vector Machine models outperformed Random Forest using Landsat and LiDAR data.

Objective

• To evaluate the effectiveness of remote sensing-based models for mapping soil salinity in irrigated agroecosystems characterized by predominantly low electrical conductivity (0-2 dS/m) values, considering both irrigation-induced and naturally occurring salinity.

Study Configuration

• Spatial Scale: Riverhurst Irrigation District, Saskatchewan, Canada. Imagery resolutions: 30 meters (Landsat 8), 5 meters (LiDAR-derived digital elevation model).
• Temporal Scale: Not explicitly mentioned, but uses Landsat 8 imagery.

Methodology and Data

• Models used: Random Forest (RF), Support Vector Machine (SVM).
• Data sources:
  • Satellite imagery: Landsat 8 for vegetation and salinity indices (30 m resolution).
  • Topographic data: LiDAR-derived digital elevation model for geomorphometric variables (5 m resolution).
  • In-situ measurements: Soil electrical conductivity for three depth intervals (0–0.3 m, 0.3–0.6 m, and 0.6–0.9 m), with the majority of values in the 0-2 dS/m range.

Main Results

• The Support Vector Machine (SVM) model demonstrated superior performance compared to Random Forest (RF) in mapping soil salinity.
• SVM achieved a higher coefficient of determination (R²) of 0.77 and a lower root mean square error (RMSE) of 0.48 dS/m.
• RF achieved an R² of 0.66 and an RMSE of 0.51 dS/m.
• Model evaluations were conducted on an independent test dataset.

Contributions

• Provides insights into the effectiveness of remote sensing for soil salinity mapping in irrigated agroecosystems dominated by low electrical conductivity values, an area previously less understood.
• Compares the performance of Random Forest and Support Vector Machine models for this specific application, demonstrating SVM's superior capability.
• Utilizes a combination of Landsat 8 imagery and LiDAR-derived geomorphometric variables to enhance salinity mapping in complex environments.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Gowera2026Spatial,
  author = {Gowera, Grace Tariro and Sorenson, Preston and Bedard-Haughn, Angela and Shirtliffe, Steven J},
  title = {Spatial prediction and mapping of soil salinity using machine learning and remote sensing covariates},
  journal = {Canadian Journal of Soil Science},
  year = {2026},
  doi = {10.1139/cjss-2025-0092},
  url = {https://doi.org/10.1139/cjss-2025-0092}
}