Data Assimilation Research Team Events / Media

Particle filter (PF) is an ensemble data assimilation method that does not assume Gaussian errors for prior distribution. Applying the PF for high dimensional dynamical systems is generally difficult since the number of required particles or the ensemble size scales exponentially with the system size. Recently, local particle filters (LPFs), which uses localization as in the ensemble Kalman filter, have been proposed to apply the PF for high-dimensional dynamics efficiently. Among them, Penny and Miyoshi (2016) developed a Local Particle Filter (LPF) in a form as the ensemble transform matrix of the Local Ensemble Transform Kalman Filter (LETKF). The LETKF has been widely used for various geophysical systems including global and regional numerical weather prediction (NWP) models and Martian atmospheric models. Therefore, implementing consistently with an existing LETKF code is useful. German Weather Service (DWD) implemented the LPF with their operational global model ICON based on their operational LETKF code, and reported a stable performance in the operational setup (Potthast et al. 2019). Further, Walter and Potthast et al. (2020) extends the LPF to the Gaussian mixture (LPFGM) to improve the LPF. This study aims to develop an LPF consistently with an existing LETKF code. Here we use the LETKF code first developed by Miyoshi (2005) based on an intermediate AGCM known as the SPEEDY model. In this presentation, we would like to focus on the theory and code design of the LPF and its Gaussian mixture extension, with only minor modifications to the existing LETKF code. Our preliminary experiments revealed that the LPFGM potentially outperforms the LETKF in sparsely-observed regions. We also discuss remaining issues of the LPF and LPFGM that should be investigated further for improvements.