Quantitative Soil Characterization for Biochar-Cd Adsorption: Machine Learning Prediction Models for Cd...

Area of Science:

• Environmental Science
• Soil Science
• Data Science
• Machine Learning

Background:

• Cadmium (Cd) soil pollution presents significant environmental and health risks.
• Biochar (BC) amendment is explored as a strategy for immobilizing Cd in soils.
• Accurate prediction of Cd concentrations is crucial for risk assessment and remediation.

Purpose of the Study:

• To assess the influence of biochar on cadmium adsorption in various soil types.
• To develop predictive models for Cd concentrations based on soil characteristics.
• To establish a framework for evaluating Cd risk in biochar-amended soils using machine learning.

Main Methods:

• Quantitative soil characterization and incubation of soil samples.
• Application of Long Short-Term Memory (LSTM), Bidirectional Gated Recurrent Unit (BiGRU), and 5-layer Convolutional Neural Networks (CNNs) for risk assessment.
• Statistical analysis of model performance using R-squared (R2) values.

Main Results:

• In control soils, the 5-layer CNN model achieved the highest accuracy (R2=0.91).
• In biochar-amended soils, the 5-layer CNN model demonstrated superior performance (R2=0.95), outperforming BiGRU (R2=0.93) and LSTM (R2=0.94).
• Soil properties like pH, clay, CEC, organic carbon, and EC were key factors in Cd immobilization.

Conclusions:

• Biochar amendment significantly enhances the prediction accuracy of Cd concentrations in soils.
• Convolutional Neural Networks (CNNs) are highly effective for predicting Cd transformation and risk in biochar-amended soils.
• The study supports the development of ecological soil remediation strategies for sustainable heavy metal management.