Data assimilation for non-linear systems with a hybrid non-linear particle filter

The proposed methodology utilizes a novel data assimilation technique based on hybrid non-linear particle filters, a framework that has demonstrated efficacy in non-linear and non-Gaussian scenarios within the domain of Earth sciences. In numerical sensitivity experiments conducted on a non-linear dynamical system (Lorenz 63), the new method prevents filter divergence using only 10 particles for both dense and sparse observation networks. A comparison of the newly developed hybrid non-linear method with the local ensemble transform Kalman filter (LETKF) reveals the merits of the former in data assimilation applications analogous to geophysical data. Specifically, the newly developed filter exhibits significant advantages over the LETKF, particularly when the observation network consists of densely spaced measurements that are non-linearly related to the model state, akin to remote sensing data frequently employed in atmospheric analyses.