3rd model intercomparison projects of atmospheric dispersion model for 137Cs emitted from Fukushima Daiichi Nuclear Power Plant, and application of MIPs' results for usage in an emergency

<p>The 3rd model intercomparison project (MIP) of atmospheric dispersion model targeting on <sup>137</sup>Cs emitted from the Fukushima Daiichi Nuclear Power Plant (FDNPP) in March 2011 was conducted (Sato et al. 2020). Nine models participated in the 3rd MIP. All participated models used the identical source term of Katata et al. (2015) and the identical meteorological data (Sekiyama and Kajino, 2020) as in the previous MIP (i.e., 2nd MIP Sato et al. 2018), but finer horizontal grid resolution (1 km) than that of 2nd MIP (3 km) was used for understanding the behavior of atmospheric <sup>137</sup>Cs measured in the vicinity of FDNPP. Results of the models elucidated that, as in the 2nd MIP, most of the observed high atmospheric <sup>137</sup>Cs concentrations (plumes) were reasonably well simulated by the models, and the good performance of some models cancelled a bad performance of some models when used as an ensemble, which highlights the advantage of the multimodel ensemble. The analyses also indicated that the use of the finer grid resolution (1 km) improved the meteorological field in the vicinity of FNDPP. As a consequence, the atmospheric <sup>137</sup>Cs measured near FDNPP was more reasonably reproduced in 3rd MIP than 2nd MIP.</p><p>As well as the evaluation of the performance of the model, we examined the usefulness of the results of atmospheric dispersion simulation in an emergency base on the results of 2nd and 3rd MIPs. For the analyses we defined the worst situation as that plume is observed but the model does not simulate it. The analyses reported that the worst situation happened in only 3% of the total calculation period by using the multimodel ensemble, even if the absolute value of the simulated <sup>137</sup>Cs in each model was different in the range of factor 3-6. The analyses also indicated that from six to eight models are required for making most of the advantages of the multimodel ensemble.</p>