Improving transfer functions to describe radiocesium wash-off fluxes for the Niida River by a Bayesian approach

This paper proposed methodological refinements of the generic transfer function approach to reconstruct radiocesium wash-off fluxes from contaminated catchments, by the integration of hydrological descriptors (passed volume of water, flow rate fluctuations and antecedent flow conditions). The approach was applied to the Niida River (Fukushima prefecture, Japan) for the period 03/2011-03/2015, for which daily flow rate (m³/s) and infrequent total radiocesium concentration (Bq/L) values were available from literature. Three models were defined, generic TF (Φ₀), flow-corrected time variant (Φ₁) and antecedent-flow corrected variant (Φ₂). Calibration of these models' parameters was performed with a Bayesian approach because it is particularly adapted to limited datasets and censored information, and it provides parameters distributions. The model selection showed strong evidence of model Φ₂ (indicated by marginal likelihood), which integrates current and recent hydrology in its formulation, and lower prediction errors (indicated by RMSE and ME). Models Φ₁ and Φ₂ better described wash-off dynamics compared to model Φ₀, due to the inclusion of one or several hydrological descriptors. From March 2011 to March 2015, model Φ₂ estimated ¹³⁷Cs export from Niida catchment between 0.32 and 0.67 TBq, with a median value of 0.49 TBq, which represents around 0.27% of the initial fallout and could represent a significant source-term to the Ocean compared to the direct release from Fukushima Dai-ichi Nuclear Power Plant (FDNPP). Moreover the remaining 99% of the initial radiocesium fallout within the catchment may constitute a persistent contamination source for wash-off. Although the proposed methodology brought improvements in the assessment of wash-off fluxes, it remains an empirical interpolation method with a limited predictive power, particularly for recent low activities. To improve predictions, modelling approaches require more observed data (particularly more activity values corresponding to more hydrological conditions), and the inclusion of more hydrological descriptors.