This study focuses on the non-Gaussian statistics based on a 10 240-member ensemble data assimilation. The 10 240 members can resolve the detailed structures of the probability density functions (PDFs) and indicates that the non-Gaussian PDF is caused by multimodality and outliers. The results show that the outliers appear randomly and that large multimodality corresponds well with large analysis error, mainly in the tropical regions where highly nonlinear convective processes appear frequently.