Prediction of caesium dynamics in soil using Mid-Infrared Spectroscopy (MIRS)

The release of radiocaesium radionuclides (RCs) has affected food safety after the accidents at Chernobyl and Fukushima Daiichi nuclear power plants. ¹³⁷Cs, in particular, is of major concern in terms of agriculture due to its relatively long half-life (30.2 years for ¹³⁷Cs), its strong fixation by the soil, and easy absorption by plants. Therefore, several countermeasures have been undertaken. For instance, topsoil removal and potassium fertilizer successfully reduced the level of RCs contamination in agricultural products. However, the relation between the transfer factor and exchangeable potassium (K_ex) differs depending on the soil, meaning that there are several parameters including K_ex that influences caesium (Cs) uptake by plants. The reason remains unclear, but a previous studies suggested that exchangeable ¹³⁷Cs could be a crucial factor in explaining the variation. Also, some factors such as solid/liquid distribution coefficient (Kd) of Cs, or the ratio of exchangeable ¹³⁷Cs versus total ¹³⁷Cs (¹³⁷Cs_ex / ¹³⁷Cs_total) in the soil could be involved in the determination of the risk of ¹³⁷Cs uptake by plants. Furthermore, a rapid risk assessment is needed while these parameters can take a huge amount of time to be determined. Hence, Mid-Infrared Spectroscopy (MIRS), being faster, more cost-effective, and non-destructive, may be utilized for the determination of these parameters. However, the prediction of these parameters using MIRS has yet to be assessed. In this study, we aimed to assess whether MIRS can predict Cs-related parameters in the soil such as Kd, ¹³⁷Cs_ex / ¹³⁷Cs_total, K_ex and other parameters that may be of influence on the behaviour of RCs in the soil, such as CEC, pH, and soil organic carbon.

In total about 1700 soil samples were collected from agricultural fields in the Fukushima Prefecture in Japan. The soil samples were air-dried and ground. The MIRS data were obtained using a Thermo Scientific Nicolet iS20 spectrometer. Using Partial Least Squares Regression as a baseline, the spectra data and the wet chemistry data including Kd, ¹³⁷Cs_ex / ¹³⁷Cs_total, and K_ex, among other soil parameters, were used for modelling and prediction. Even though until present (at submission of this abstract) only 176 samples have been measured, we found that balancing the range of values between training and validation sets enabled Partial Least Regression Estimation methods to provide a relatively high R² valid score for the prediction of each wet chemistry data, especially soil organic carbon, CEC, Mg_ex and Ca_ex, ranging between 0.73 and 0.88. Using less than 200 samples, however, the validation scores of Cs -related parameters were less than 0.5. Further MIRS data are expected for up to about 1600 soil samples, for ¹³⁷Cs_ex / ¹³⁷Cs_total in the soil. Additional processing and modelling techniques will be tested aiming at further improving the validation scores, and results will be shown at EGU.