Prediction heavy metals accumulation risk in rice using machine learning and mapping pollution risk

Area of Science:

• Environmental Science
• Agricultural Science
• Computational Science

Background:

• Assessing metal bioaccumulation in crops is crucial for environmental risk evaluation.
• Machine learning offers potential for accurate prediction of heavy metal uptake in crops.

Purpose of the Study:

• To develop and apply machine learning models for predicting heavy metal (Cd, Hg, As, Pb) content in rice grain.
• To identify key factors influencing heavy metal bioaccumulation in rice.
• To visualize and map the spatial distribution of Cd and Hg risks in paddy soils and rice.

Main Methods:

• Collected 2123 soil-rice samples from Jiangsu province, China.
• Applied 10 machine learning algorithms (Extremely Randomized Tree, Random Forest) to predict metal content.
• Conducted feature importance analysis to identify influencing factors.
• Utilized 1867 additional soil samples for risk mapping.

Main Results:

• Extremely Randomized Tree model showed high performance for Cd (R²=0.824) and Hg (R²=0.626).
• Random Forest model performed best for As (R²=0.389) and Pb (R²=0.325).
• Soil-Cd and pH were critical for rice-Cd; temperature and soil organic carbon influenced rice-Hg.
• Spatial analysis revealed distinct hotspot distributions for soil-Hg and rice-Hg.

Conclusions:

• Machine learning models effectively predict heavy metal bioaccumulation in rice.
• Temperature is a significant, often overlooked, factor in mercury bioaccumulation risk.
• Findings support improved environmental risk assessment and agricultural management strategies.