Combining short-range dispersion simulations with fine-scale meteorological ensembles: probabilistic indicators and evaluation during a ⁸⁵Kr field campaign

Abstract

Abstract. Numerical models of atmospheric dispersion are used for predicting the health and environmental consequences of nuclear accidents, in order to anticipate the countermeasures necessary to protect the populations. However, the simulations of these models suffer from significant uncertainties, arising in particular from input data: weather conditions and source term. To characterize weather uncertainties, it is essential to combine a well-known source term data and meteorological ensembles to generate ensemble dispersion simulations which has the potential to produce different possible scenarios of radionuclides dispersion during emergency situations. In this study, the fine-scale operational weather ensemble AROME-EPS from Météo-France is coupled to the Gaussian puff model pX developed by French Institute for Radiation Protection and Nuclear Safety (IRSN). The source term data is provided by Orano La Hague reprocessing plant (RP) that regularly discharges 85Kr during the spent nuclear fuel reprocessing process. Then, to evaluate the dispersion results, a continuous measurement campaign of 85Kr air concentration was recently conducted by the Laboratory of Radioecology in Cherbourg (LRC) of IRSN, around RP in the North-Cotentin peninsula. This paper presents a probabilistic approach to study the meteorological uncertainties in dispersion simulations at local and medium distances (2–20 km). As first step, the quality of AROME-EPS forecasts is confirmed by comparison with observations from both Météo-France and IRSN. The following step is to assess the probabilistic performance of the dispersion ensemble simulation, as well as the sensitivity of dispersion results to the method used to calculate atmospheric stability fields and their associated dispersion Gaussian standard deviations. Two probabilistic scores are used: Relative Operating Characteristic (ROC) curves and Peirce Skill Score (PSS). The results show that the stability diagnostics of Pasquill provides better dispersion simulations. In addition, the ensemble dispersion performs better than deterministic one, and the optimum decision threshold (PSS maximum) is 3 members. These results highlight the added value of ensemble forecasts compared to a single deterministic one, and their potential interest in the decision process during crisis situations.