Radiocaesium soil-to-plant transfer: a meta-analysis of key variables and data gaps on a global scale

Abstract

A harmonized, publicly accessible database of worldwide observations and experiments on radiocaesium transfer from soil to plants is lacking. Such a database is needed for evaluating and establishing transfer models, especially for regions with limited research but operational or planned nuclear reactors. Therefore, we systematically screened the literature for radiocaesium soil-to-plant transfer factors (CR, i.e., concentration ratios), extracted data that met the criteria for experimental soundness, relevance, and traceability, and compiled a harmonized database. The database included 7,182 CR data points and associated variables from 139 source documents. The CRs ranged from 0.000028 to 380 kg kg⁻¹, with the highest CR observed with soils from tropical climates and the lowest with soils from temperate climates. However, data from tropical (N = 411) and arid climates (N = 335) remained limited. Univariate and multivariate analyses revealed that CRs were most influenced by the specific study (methods and designs) in which the data were obtained, followed by soil properties and plant species-based categories. On a subset (N = 199) that contained all variables required for semi-mechanistic models, it was found that these models fitted the CR data rather well (R² = 0.42–0.50). Slightly better predictions with the same data were found with a random forest model (R² = 0.51) or a statistical mixed-effects model (R² = 0.58). More adequate machine learning models could not yet be created due to insufficient reliable data. The harmonized database in this study can be further completed and analyzed to support machine learning applications and improve impact assessments of food chain contamination following accidental radiocaesium deposition on agricultural land.