Comparison of Different Machine Learning Methods for Predicting Cation Exchange Capacity Using Environmental and...

Area of Science:

• Soil Science
• Environmental Science
• Geospatial Analysis

Background:

• Cation exchange capacity (CEC) is a critical soil property influencing nutrient availability and contaminant transport.
• Predicting CEC is essential for effective land management and agricultural practices.
• Integrating topographic and remote sensing data offers a promising approach for large-scale CEC estimation.

Purpose of the Study:

• To evaluate the predictive capability of topographic features and remote sensing data for soil CEC.
• To compare the performance of various machine learning models (Random Forest, k-NN, Cubist, SVM) in predicting CEC.
• To identify the most influential environmental variables for CEC prediction in western Iran.

Main Methods:

• Collected 97 surface soil samples (0-20 cm) for laboratory analysis of soil properties and CEC.
• Utilized topographic data, remote sensing indices (NDMI, salinity index), and environmental variables (geology, geomorphology).
• Employed machine learning models including Random Forest (RF), k-NN, Cubist, and SVM with 10-fold cross-validation.

Main Results:

• The Random Forest (RF) model demonstrated superior performance in the training dataset (R² = 0.86, RMSE = 2.76).
• The Cubist model showed better performance in the validation dataset (R² = 0.49, RMSE = 4.51).
• Higher CEC was associated with early Quaternary deposits rich in smectite and vermiculite, while lower CEC was found in mountainous, coarse-textured soils.

Conclusions:

• Topographic attributes (valley depth, elevation, slope, TRI) and remote sensing data (ferric oxides, NDMI, salinity index) are key predictors of soil CEC.
• Machine learning models, particularly RF, effectively predict CEC by integrating diverse environmental data.
• The study provides a valuable framework for mapping and understanding soil CEC variability using geospatial techniques.