Lazaar et al. (2025) Rapid Estimation of Soil Electrical Conductivity (ECe) in Arid Regions Using Pedotransfer Functions, FTIR Spectroscopy and Machine Learning

Identification

• Journal: Earth Systems and Environment
• Year: 2025
• Date: 2025-10-17
• Authors: Ayoub Lazaar, Tarik El Moatassem, Laila Tajeddine, Laila Ait Mansour, Fassil Kebede
• DOI: 10.1007/s41748-025-00800-1

Research Groups

• Center of Excellence for Soil and Fertilizer Research in Africa (CESFRA), College for Sustainable Agriculture and Environmental Sciences (SAES), Mohammed VI Polytechnic University (UM6P), Ben Guerir, Morocco.

Short Summary

This study develops pedotransfer functions and an innovative FTIR spectroscopy approach coupled with machine learning to rapidly and accurately estimate soil saturated paste extract electrical conductivity (ECe) from diluted soil-to-water extracts. It demonstrates that soil-type-specific conversion factors and integrated soil properties (CEC, CaCO₃) significantly enhance ECe prediction, offering a rapid, cost-effective method for large-scale salinity monitoring.

Objective

• To develop empirical equations (pedotransfer functions) adapted to specific soil types, as well as combined models, to predict ECe using different soil-to-water ratio methods (EC 1:1, EC 1:2.5, EC 1:5).
• To evaluate the performance of Mid-infrared (MIR or FTIRS) spectroscopy coupled with Partial Least Squares Regression (PLSR) in predicting EC values derived from the saturated paste extract (ECe) and the soil-to-water ratio methods.
• To explore the potential of the Random Forest machine learning algorithm to predict ECe from EC 1:5 and to investigate how the integration of two key soil properties (CaCO₃ and CEC) influences the performance of the predictive models.

Study Configuration

• Spatial Scale: Irrigated perimeter of Triffa plain, eastern region of Morocco, near Berkane city (geographical center: 34°59’10.63"N, 2°20’13.05"W). Total area of approximately 140 km². 59 soil samples were collected from 22 profiles across five soil types (Mollisols, Ultisols, Aridisols, Histosols, Entisols).
• Temporal Scale: Soil sampling was conducted in February 2024.

Methodology and Data

• Models used:
  • Pedotransfer functions (simple linear regression)
  • Partial Least Squares Regression (PLSR)
  • Random Forest (RF) machine learning algorithm
  • Two-way ANOVA
• Data sources:
  • Soil samples: 59 soil samples collected from 22 soil profiles at three depths (0–20 cm, 20–50 cm, and 50–100 cm) in the Triffa plain, Morocco.
  • Electrical Conductivity (EC) measurements: EC of saturated paste extract (ECe), and EC from 1:1, 1:2.5, and 1:5 soil-to-water ratios, measured with a conductivity meter.
  • Soil properties: Calcium carbonate (CaCO₃) content (French standard NF EN ISO 10693 (2014)) and Cation Exchange Capacity (CEC) (NF X31-130 (1999) standard).
  • FTIR Spectroscopy: Mid-Infrared (MIR) spectra (2500–25000 nm or 4000–600 cm⁻¹) acquired using a Bruker Tensor II HTS-XT spectrometer with a spectral resolution of 4 cm⁻¹, averaging 60 scans per sample. Spectral preprocessing included Savitzky-Golay smoothing.

Main Results

• Pedotransfer Functions: A significant linear correlation was found between ECe and diluted extract EC values, with R² > 0.89 across all extracts. Conversion factors (CF) for ECe = CF × EC soil-to-water ratios varied significantly among soil types (e.g., Mollisols: 1.87 (1:1), 5.30 (1:2.5), 9.52 (1:5); Ultisols: 1.77 (1:1), 6.03 (1:2.5), 8.66 (1:5)). A combined soil model showed strong correlations with R² = 0.93 (1:1), R² = 0.89 (1:2.5), and R² = 0.95 (1:5).
• FTIR Spectroscopy: FTIR-developed PLSR models demonstrated high predictive accuracy for all soil-to-water extracts (R² = 0.86–0.91, RMSE = 0.41–3.69 dS/m) and for ECe (R² = 0.87, RMSE = 3.69 dS/m). Distinct spectral features at 1970–2550 cm⁻¹ and 2867–3086 cm⁻¹ were identified as the most sensitive regions for EC prediction.
• Machine Learning (Random Forest): Random Forest (RF) models accurately predicted ECe from EC1:5 (R² = 0.92, RMSE = 2.05 dS/m). Prediction performance was enhanced when including CEC and CaCO₃ content (R² = 0.95, RMSE = 1.51 dS/m for the combined model). EC1:5 was the most influential parameter (93.80%), followed by CEC (3.80%) and CaCO₃ (2.40%).
• Salinity Gradient: Salinity varied significantly across soil types and depths. Histosols and Mollisols exhibited the highest ECe values in surface horizons (e.g., Histosols mean ECe = 27.70 dS/m at 0-20 cm). Salinity generally decreased with depth, though an increase with depth was observed in Aridisols and some Mollisols/Ultisols profiles.
• Mapping: Predicted EC maps showed strong alignment with measured values across all depths, indicating FTIR as an effective tool for estimating soil salinity, particularly in the northwestern zone of the Triffa plain.

Contributions

• Developed novel soil-type-specific pedotransfer functions for ECe estimation from diluted soil-to-water extracts, addressing a critical gap in regional salinity assessment for the Triffa plain and other arid/semi-arid African regions.
• Demonstrated the high effectiveness of integrating FTIR spectroscopy with machine learning (PLSR and Random Forest) for rapid, accurate, and cost-effective prediction of ECe, significantly reducing reliance on time-consuming saturated paste analysis.
• Identified specific mid-infrared spectral ranges (1970–2550 cm⁻¹ and 2867–3086 cm⁻¹) as highly sensitive for EC prediction, linking them to key carbonate and bicarbonate compounds.
• Quantified the significant improvement in ECe prediction accuracy by incorporating additional soil properties like Cation Exchange Capacity (CEC) and Calcium Carbonate (CaCO₃) into machine learning models.
• Provided a practical and robust solution for large-scale precision agriculture and soil salinity monitoring, supporting sustainable soil management and food security.

Funding

• Université Mohammed VI Polytechnique

Citation

@article{Lazaar2025Rapid,
  author = {Lazaar, Ayoub and Moatassem, Tarik El and Tajeddine, Laila and Mansour, Laila Ait and Kebede, Fassil},
  title = {Rapid Estimation of Soil Electrical Conductivity (ECe) in Arid Regions Using Pedotransfer Functions, FTIR Spectroscopy and Machine Learning},
  journal = {Earth Systems and Environment},
  year = {2025},
  doi = {10.1007/s41748-025-00800-1},
  url = {https://doi.org/10.1007/s41748-025-00800-1}
}