Ouzemou et al. (2026) Integrating post-rainfall multispectral satellite-derived features and multi-source datasets to enhance soil salinity mapping accuracy

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2026
• Date: 2026-01-01
• Authors: Jamal-Eddine Ouzemou, Ahmed LAAMRANI, Ali El Battay, Joann K Whalen, Abdelghani CHEHBOUNI
• DOI: 10.1016/j.rsase.2026.101896

Research Groups

• Center for Remote Sensing Applications (CRSA), Mohammed VI Polytechnic University (UM6P), Ben Guerir, Morocco
• College of Agriculture and Environmental Sciences (CAES), Mohammed VI Polytechnic University (UM6P), Ben Guerir, Morocco
• Department of Geography, Environment & Geomatics, University of Guelph, Guelph, ON, Canada
• Center of Sustainable Soil Sciences (C3S), Mohammed VI Polytechnic University (UM6P), Ben Guerir, Morocco
• Department of Natural Resource Sciences, McGill University, Ste-Anne-de-Bellevue, QC, Canada

Short Summary

This study integrates Sentinel-2, Landsat-9, and PlanetScope imagery with field-measured electrical conductivity and machine learning to enhance soil salinity mapping accuracy in Morocco's Sehb El Masjoune area. It found that combining Sentinel-2 with a Gradient Boosting Regressor model and novel post-rainfall proxies (Depression Proxy and Soil Clusters Proxy) achieved the highest accuracy (R² = 0.85), highlighting the critical role of micro-topography and soil properties in salinity distribution.

Objective

• To enhance soil salinity mapping accuracy in the Sehb El Masjoune area, Morocco, by integrating post-rainfall multispectral satellite-derived features and multi-source datasets.
• To test three hypotheses: (i) micro-topographic depressions retaining moisture enhance localized salt accumulation; (ii) variations in soil texture and composition affect salt retention capacity; and (iii) the integration of multi-source Earth Observation data with novel proxies (Depression Proxy and Soil Clusters Proxy) yields more accurate and interpretable spatial patterns of salinity than conventional approaches.
• To produce a robust dry-season soil salinity map and explore intra-annual salinity dynamics qualitatively.

Study Configuration

• Spatial Scale:
  • Study area: Sehb El Masjoune region, Bahira plain, central Morocco (approximately 532 km²).
  • Satellite imagery resolutions: PlanetScope (3 m), Sentinel-2 (10 m), Landsat-9 (30 m).
  • Digital Elevation Model (DEM) resolution: 30 m.
  • Field plots: 1 m × 1 m quadrats.
• Temporal Scale:
  • Field sampling: Summer 2022 (dry-season conditions).
  • Satellite imagery acquisition:
    • Landsat-9: September 07, 2022 (Dry season).
    • Sentinel-2: June 05, 2022 (Wet season), October 07, 2022 (Dry season).
    • PlanetScope: December 29, 2021 (Wet season), October 07, 2022 (Dry season), and time-series from December 23, 2021, to January 03, 2022 (for proxy development).
  • Temporal monitoring: Qualitative comparison of predicted salinity patterns between May 06, 2022 (wet season) and July (dry season).

Methodology and Data

• Models used: Random Forest (RF), Gradient Boosting Regressor (GBR).
• Data sources:
  • Satellite imagery: Sentinel-2 MSI, Landsat-9 OLI-2, PlanetScope (surface reflectance products).
  • Field data: 121 topsoil samples (0–5 cm depth) with georeferenced Electrical Conductivity (ECe) measurements (ranging from 0.08 to 227 dS/m).
  • Derived features:
    • Spectral indices (e.g., Brightness Index, Normalized Difference Salinity Index, Salinity Index, ASTER-Salinity Index, Normalized Difference Water Index, Normalized Difference Vegetation Index, Soil-Adjusted Vegetation Index).
    • Principal Component Analysis (PCA) on spectral bands.
    • Topographic variables from a 30 m DEM: Elevation, Slope, Topographic Wetness Index (TWI), Flow Accumulation, Aspect, Convergence Index, Hillshade, Longitudinal Curvature, Plan Curvature, Profile Curvature, Valley Depth, Tangential Curvature.
    • Novel post-rainfall proxies derived from PlanetScope imagery: Depression Proxy (DP) and Soil Clusters Proxy (SCP).

Main Results

• The sequential integration of the Soil Clusters Proxy (SCP) and Depression Proxy (DP) systematically improved model accuracy and stability compared to baseline approaches.
• The Sentinel-2 dataset combined with the Gradient Boosting Regressor (GBR) model and the Depression Proxy (DP) (Approach-3) yielded the highest accuracy on an independent test set (R² = 0.85, Root Mean Square Error (RMSE) = 28.16 dS/m, Mean Absolute Error (MAE) = 17.72 dS/m).
• For Landsat-9, the Random Forest (RF) model in Approach-2 achieved peak robustness (R² = 0.82, RMSE = 29.14 dS/m, MAE = 17.92 dS/m).
• For PlanetScope, the GBR model in Approach-3 performed best (R² = 0.74, RMSE = 28.48 dS/m, MAE = 16.79 dS/m).
• Within the critical agronomic salinity range (0–64 dS/m), the Sentinel-2 GBR (Approach-3) model achieved RMSE = 17.39 dS/m and MAE = 11.66 dS/m.
• Categorical evaluation of the Sentinel-2 GBR (Approach-3) test set showed 56% exact class accuracy and 80% within ±1 salinity class.
• Feature importance analysis indicated that terrain-controlled variables, particularly the Depression Proxy (DP) for Sentinel-2 and PlanetScope, and Valley Depth for Landsat-9, were the strongest predictors. Other topographic attributes (Slope, Hillshade, Elevation, Aspect) and spectral indices (SI-2, PVI, NDWI1, CRSI, GDVI) were also consistently important. The Soil Clusters Proxy (SCP) provided meaningful pedological information.
• The generated dry-season salinity map revealed that 70% of the study area was affected by varying degrees of salinization, with pronounced salt concentrations in the central basin and southern lowlands, consistent with hydrological and topographic controls.
• Qualitative temporal monitoring suggested lower predicted salinity in the wet season (May) compared to the dry season (July), with a clear intensification of salt accumulation in the central flatlands during the dry season due to evaporative concentration.

Contributions

• Introduces two novel dynamic post-rainfall proxies, Depression Proxy (DP) and Soil Clusters Proxy (SCP), which provide process-informed insights into moisture retention and soil heterogeneity, enhancing the detection of salt accumulation zones, particularly in Moroccan salinity assessments.
• Establishes a modular framework that integrates these hydrologically informed indicators with remote sensing and topographic variables, improving the accuracy and interpretability of salinity mapping.
• Demonstrates that combining multi-source Earth Observation data with these proxies significantly improves soil salinity mapping accuracy and interpretability in heterogeneous dryland environments.
• Provides the first high-accuracy salinity map for the Sehb El Masjoune region, Morocco, with physically consistent spatial patterns.

Funding

• SELMAS project (multidisciplinary)
• OCP group foundation (APRA program)
• Mohammed VI Polytechnic University (UM6P)

Citation

@article{Ouzemou2026Integrating,
  author = {Ouzemou, Jamal-Eddine and LAAMRANI, Ahmed and Battay, Ali El and Whalen, Joann K and CHEHBOUNI, Abdelghani},
  title = {Integrating post-rainfall multispectral satellite-derived features and multi-source datasets to enhance soil salinity mapping accuracy},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2026},
  doi = {10.1016/j.rsase.2026.101896},
  url = {https://doi.org/10.1016/j.rsase.2026.101896}
}