In conventional remote visualization a new image has to be transferred from the server to the client for each visualization parameter change. Network bandwidth and latency limitations result in low quality and reduced interactivity. In order to solve these problems, we propose to transmit superimages from the server to the client, where a superimage contains sufficient data for a quality reconstruction of thousands of frames at the client, without any additional data from the server. In the case of dynamic datasets the superimage has to encode not only sample color and position, but also sample trajectories. In this presentation I will introduce an efficient screen space compression method for dynamic datasets that provides fast encoding, trivial decoding, a good compression ratio, and a small, application controllable compression error.