Uncertainty study on atmospheric dispersion simulations using meteorological ensembles with a Monte Carlo approach, applied to the Fukushima nuclear accident

Abstract

In emergency cases, when nuclear accidental releases take place, numerical models, developed by French Institute of Radiation Protection and Nuclear Safety (IRSN), are used to forecast the atmospheric dispersion of radionuclides. These models compute the quantity of radionuclides in the atmosphere, their deposited amount on the ground, and the subsequent gamma dose rate. Their results are used to make recommendations to protect the population in case of nuclear accident. However, the simulations are subject to considerable uncertainties. These uncertainties originate from different sources: input variables (weather forecasting, source term), physical parameters used in the models (turbulent diffusion, scavenging coefficient, deposition velocity, etc.) and model approximations (representativeness and numerical errors). This paper presents the propagation of input uncertainties through a Eulerian radionuclide transport model, ℓdX, applied to the Fukushima nuclear disaster. This uncertainty propagation involves perturbing the input variables and making numerous calls to the model. The perturbations should be broad enough to cover the possible range of variation of uncertain variables. Weather forecast ensembles are used to take into account meteorological uncertainties, and several source terms from the literature are included. The following step is to evaluate the spread of the outputs in order to draw insights about the subsequent uncertainties. In order to assess the quality of the ensemble of simulations, comparisons with radiological observations were carried out, using statistical indicators, both deterministic such as Root Mean Square Error (RMSE) or Figure of Merit in Space (FMS), and probabilistic indicators such as rank histograms, Brier score and Discrete Ranked Probability Scores (DRPS).