Digital mapping of soil erodibility factor in northwestern Iran using machine learning models

Area of Science:

• Soil Science
• Environmental Science
• Geospatial Analysis

Background:

• Soil erodibility factor (K-factor) mapping is vital for effective water erosion risk management at district scales.
• Understanding the spatial distribution of K-factor is essential for targeted conservation efforts.

Purpose of the Study:

• To predict and map low and high classes of the soil erodibility factor (K-factor) in Northwest Iran.
• To evaluate the performance of Random Forest (RF) and Artificial Neural Network (ANN) models for K-factor prediction.

Main Methods:

• Soil samples were collected at 64 points (1 km spacing) to determine particle size distribution and organic carbon (OC).
• Twenty-one terrain attributes were derived from a Digital Elevation Model (DEM).
• K-factor was modeled using Random Forest (RF) and Artificial Neural Network (ANN), followed by non-linear Multiple Logistic Regression (NMLR) for class prediction.

Main Results:

• The Random Forest (RF) model demonstrated superior performance over the Artificial Neural Network (ANN) model, with a high coefficient of determination (R² = 0.85) and accuracy (RMSE = 0.003).
• The study area was classified into low (60.4%) and high (39.6%) K-factor classes using NMLR, showing good correlation (R² Cox and Snell = 0.78, R² Nagelkerke = 0.65).

Conclusions:

• The Random Forest model is effective for predicting the spatial distribution of the K-factor.
• The generated map of low and high K-factor classes provides valuable information for farmers and managers to mitigate soil water erosion risks.