Radon potential mapping in Jangsu-gun, South Korea using probabilistic and deep learning algorithms

Area of Science:

• Environmental Science
• Geology
• Public Health

Background:

• Naturally occurring radon gas poses health risks through inhalation and ingestion.
• Identifying areas with high radon potential is essential for public safety and urban planning.

Purpose of the Study:

• To detect and map radon-prone areas using GIS-based probabilistic and machine learning techniques.
• To analyze key influencing factors on radon distribution.
• To compare the predictive performance of different machine learning models for radon potential mapping.

Main Methods:

• Employed a geographic information system (GIS) framework.
• Utilized frequency ratio (FR) and convolutional neural network (CNN) machine learning models.
• Analyzed ten influencing factors: elevation, slope, TWI, valley depth, fault density, lithology, and soil concentrations of Cu, CaO, Fe2O3, and Pb.
• Validated models using receiver operating characteristic (ROC) curve analysis.

Main Results:

• Approximately 40% of the study area was classified as having very high or high radon risk.
• The CNN model demonstrated superior performance (AUC ~0.84) compared to the FR method.
• Slope, lithology, and topographic wetness index (TWI) were identified as the most significant predictors of radon-affected areas.
• Both models exhibited high predictive power for radon potential mapping.

Conclusions:

• The study successfully identified and mapped high-risk radon zones using advanced analytical methods.
• Machine learning, particularly CNN, offers a robust approach for predicting radon potential.
• Findings can guide urban development and inform radon screening strategies to mitigate public health risks.