[Use of Machine Learning Methods to Identify Soil Parent Materials in a High-cadmium Geological Background Area]

Area of Science:

• Geochemistry
• Environmental Science
• Soil Science

Context:

• Karstic soils in South China exhibit high cadmium (Cd) content but low mobility.
• Understanding soil parent material is vital for assessing Cd geochemical behavior and ecological risk.
• Tropical climates in South China hinder accurate parent material identification due to limited rock outcrops.

Purpose:

• To identify soil parameters influencing lithology distribution and soil Cd activity.
• To predict soil parent materials using machine learning models based on soil characteristics.
• To assess the ecological risk of soil Cd in high geological background areas.

Summary:

• Over 5,000 soil samples were analyzed, revealing that underlying bedrock controls soil properties and Cd distribution.
• Sequential extraction and correlation analyses identified Fe/Mn oxides, total organic carbon (TOC), CaO, and pH as key factors for Cd content and mobility.
• Artificial neural network (ANN), random forest (RF), and support vector machine (SVM) models were employed for parent material prediction.
• The RF model demonstrated superior accuracy (higher Kappa coefficients and overall accuracies) compared to ANN and SVM, indicating its potential for large-scale lithology mapping.

Impact:

• This research offers a novel approach for mapping lithology distribution in areas with challenging geological identification.
• Provides a new method for assessing soil Cd ecological risk in high geological background regions.
• Highlights the effectiveness of machine learning, specifically RF, in predicting soil parent materials from extensive soil datasets.