Estimation of 90Sr and 137Cs activity concentrations in Chornobyl wood: significance of factors and classical vs. machine learning methods

doi:10.1016/j.jenvrad.2025.107839

Estimation of 90Sr and 137Cs activity concentrations in Chornobyl wood: significance of factors and classical vs. machine learning methods

2025-10-01

0

PRE

AI

DOI：10.1016/j.jenvrad.2025.107839

Original

Original request for help

Save

Abstract

En 中文

The Chornobyl forests have undergone significant ecological transformation since the 1986 accident. Despite this, harvested wood from the region continues to be used for economic purposes. This study aimed to assess the importance of forest site characteristics, as well as the predictive capabilities of classical and machine learning (ML) methods, for estimating the content of 90Sr and 137Cs in wood. A total of 205 circular sample plots were created across contrasting forest stands, where trunk wood and mineral soil samples were collected and analyzed. Aggregated transfer factors for the studied radionuclides showed wide variability. The highest 90Sr transfer was observed in pine and birch that were growing on dry, nutrient-poor soils. The average stand diameter was significantly correlated with 90Sr content in wood, while soil fertility showed a consistent inverse relationship with the translocation of both 90Sr and 137Cs. Extreme gradient boosting (XGBoost) models were trained to predict radionuclide concentrations based on site characteristics. Shapley additive explanations (SHAP) analysis of the parameterized ML algorithms revealed that soil contamination density, stand diameter, tree age, and dose rate were the most important predictors, whereas tree species had limited explanatory power. XGBoost outperformed classical Tag-based approaches for 137Cs and achieved comparable performance for 90Sr in predicting radionuclide activity concentrations in trunk wood. Obtained results demonstrate the value of incorporating forest features into ML algorithms for improving the assessment of radionuclide contamination in forest biomass and provide new insights into the relative importance of site characteristics in determining 90Sr and 137Cs accumulation in trees.

Keywords:

Chornobyl forest

Radionuclides

Aggregated transfer factor

Activity concentration

XGBoost

Estimation of 90Sr and 137Cs activity concentrations in Chornobyl wood: significance of factors and classical vs. machine learning methods

Abstract

Journal

Organization