Uncertainties in short term prediction of atmospheric dispersion of radionuclides. A case study of a hypothetical accident in a nuclear floating power plant off the West coast of Norway.

Short-term predictions for dispersion of radionuclides in the atmosphere following releases from nuclear incidents are associated with uncertainties originating from meteorology, source term and parameterization. Characterization of these uncertainties is of key importance for preparedness, decision making during an accident and for the further uncertainty propagation in the subsequent modelling of human and ecosystem exposures. Increased traffic of nuclear-propulsion vessels in Norwegian territorial waters gives rise to growing concern of a potential nuclear accident along the coast of Norway. In the present study, we have quantified and inter-compared the uncertainties associated with the model outputs for a hypothetical loss of coolant accident with an ensuing fire in a nuclear vessel situated along the Norwegian coastline, applying two different atmospheric dispersion models: the SNAP Lagrangian particle model (SNAP-Severe Nuclear Accident Program) and the DIPCOT Lagrangian puff model (DIPCOT - Dispersion over Complex Terrain). The case highlights a situation with atmospheric transport from the offshore area to the coast of Western Norway, combined with large wet deposition in inland mountainous terrain, i.e. a common weather situation in this region. The meteorological data include an Ensemble Prediction System with nine ensemble members in addition to a deterministic base run. Five different 7 h emission scenarios with the same total released activity were considered. Hourly wind data at 10 m above ground for a 24 h period, showed that 36% of the wind direction and 41% of the wind speed data were outside the spread of the meteorological ensemble. About 55% and 13% of the measured values fell outside the ensemble for hourly 2 m above ground temperatures and 3 hourly accumulated precipitation, respectively, indicating that the ensemble did not cover all uncertainties in the meteorological fields. The maps of accumulated concentrations and depositions were qualitatively similar for the two models, but SNAP predicted higher accumulated concentration levels compared to DIPCOT for quite large areas, while DIPCOT yielded larger total depositions in the same areas. Furthermore, the direction, speed of movement and spatial extension of the radioactive plume from the accident varied considerably from one model to the other. The spread in the dispersion of the radionuclides ranged from a factor of about 1-3 in the source area to a factor of about 2-5 further away. The spreads due to meteorology and emission scenarios were of similar magnitude. Considering the ratio of the 50th percentiles of the two models, the spread varied by a factor of about 1-9, indicating that uncertainties arising from the formulation of the dispersion model could be as important or even larger than those associated with meteorology and emissions. Thus, it is recommended to include the uncertainty originating from the choice of the dispersion model into the overall uncertainty of short-term prediction of the dispersion of radionuclides and to exploit this further by generating an ensemble of several dispersion models.