Data assimilation is the task to combine model forecast with observations in order to reduce forecast uncertainties. Current data assimilation techniques include variational and ensemble-based methods. The focus of this talk in on ensemble-based method which can be viewed as linear transformations in the forecast ensembles. We call such methods linear ensemble transform filters (LETFs). Popular methods such as sequential Monte Carlo  and ensemble Kalman filters fit into this class. In order to further improve those methods we put LETFs into the framework of McKean models and optimal transportation. This perspective leads to the ensemble transform particle filter (ETPF). We finally demonstrate that localization of the analysis step allows one to apply ensemble Kalman filters as well at the ETPF to  spatially extended systems. Such infinite-dimensional problems cannot be tackled by standard sequential Monte Carlo methods/particle filters.