When assimilating remotely sensed soil moisture or soil temperature in a land surface vegetation model, i think that using a coupled land-atmosphere model is better as then atmospheric component will take care of the land-atmosphere feedback mechanisms for the future simulation