Performance Comparison of Spatial Interpolation Methods for Mapping Fallout Radionuclides: A Case Study of 137Cs in Serbian Soils

Mapping the spatial distribution of fallout radionuclides in soil is essential for environmental assessments and evaluating potential future radiological risks, while also supporting the interpretation of spatial variability relevant to their use as soil redistribution tracers. However, model performance depends on data characteristics and spatial context, which limits the possibility of defining a universally optimal interpolation method. This study aimed to model the spatial distribution of 137Cs at the national scale in Serbia by systematically applying and comparing 33 interpolation techniques, classified as deterministic or stochastic. A total of 193 soil samples, collected in accordance with International Atomic Energy Agency guidelines, were analyzed for 137Cs activity using gamma spectrometry. The performance of interpolation techniques was evaluated through leave-one-out cross-validation (LOOCV) and an appropriateness index (AI) for comparison. Among the tested techniques, Empirical Bayesian Kriging Regression Prediction (EBKRP), which incorporates a digital elevation model (DEM) as an additional explanatory variable, achieved the best performance (AI = 0.8548). The resulting map of 137Cs activity reveals a pronounced spatial gradient, with values ranging from below 22 Bq/kg to approximately 148 Bq/kg, reflecting the combined effects of initial 137Cs deposition following the Chernobyl accident and subsequent long-term soil redistribution. Higher activities are restricted to the hilly and mountainous regions of western, eastern, and southeastern Serbia. Interpolation uncertainty increases in regions with elevated 137Cs activities and should be accounted for in spatial interpretation.