Development of a prediction model for ambient dose equivalent rate distribution based on ecological half-life profiles using LASSO regression and KURAMA data

The two-component model, comprising a fast-decay and a slow-decay component, has been widely used to approximate the decreasing trends of air dose rates in contaminated areas surrounding major nuclear accident sites. However, its adequacy is yet to be thoroughly validated. This study analyzed extensive car-borne survey data collected from 2011 to 2016 after the Fukushima Daiichi Nuclear Power Plant accident using the least absolute shrinkage and selection operator regression with a high-degree-of-freedom model. This analysis aimed to evaluate the adequacy of the two-component model and investigate the profiles of ecological half-lives. The results demonstrate that the two-component model can approximate the decreasing trend of air dose rates in the Fukushima area well in most cases. However, in ~20% of the cases, the one-component model provided a better fit. The fast-decay component in the two-component model exhibited a sharp ecological half-life peak below 1 y, with a frequency distribution peaking at 0.3-0.4 y. In contrast, the slow-decay component displayed a broader half-life peak in approximately half of the cases, with a frequency distribution spanning several years to over 50 y. The reduction speed of air dose rates was fastest in urban areas, followed by paddy fields, croplands, deciduous forests, and evergreen forests. The reduction speed decreased as the initial air dose rate increased, a trend explained by the weight assigned to the fast-decay component rather than the value of its ecological half-life. Future predictions of air dose rate distributions were made using a prediction model formula that incorporated the average ecological half-life profiles calculated for each land-use and initial air dose rate category. Prediction accuracy was verified through comparison with integrated map data, which merge air dose rate datasets obtained using different monitoring methods and represent the most currently reliable source. The predicted values tended to decay faster overall than the integrated map data, with an average deviation within 10% over the six-year period. This discrepancy arises because the car-borne survey data were collected on paved roads, where air dose rates decrease rapidly due to the quick washout of radiocesium. The differences between the prediction model values and integrated map values were larger for forests than for farmlands (paddy fields and croplands) and urban areas. This suggests that the reduction in air dose rates is slower in pure forests, where the car-borne surveys were rarely conducted.