Abstract: Compressing sequences of complex time-varying surfaces as generated by medical instrumentations or complex physical simulations can be extremely challenging: repeated topology changes during the surface evolution render most of the previous techniques for compression of time-varying surfaces inefficient or impractical. In order to provide a viable solution, we propose a new approach based upon an existing isosurface compression technique designed for static surfaces. We exploit temporal coherence of the data by adopting the paradigm of block-based motion prediction developed in video coding and extending it using local surface registration. The resulting prediction errors across frames are treated as a static isosurface and encoded progressively using an adaptive octree-based scheme. We also exploit local spatiotemporal patterns through context-based arithmetic coding. Fine-grain geometric residuals are encoded separately with user-specified precision. The other design choices made to handle large datasets are detailed.