Efficient Streaming Lossy Compressor Design for Unsteady Cfd on Distributed Meshes

High-fidelity unsteady computational fluid dynamics (CFD) simulations generate extremely large datasets that are valuable for post-processing, visualization, and training data-driven models. However, limited disk space and filesystem bandwidth typically force practitioners to save only sparse temporal subsets, discarding valuable flow physics information. Lossy compression offers a solution by reducing dataset sizes while maintaining user-specified error bounds, enabling higher-frequency data storage. This work presents a temporal delta-frame compression method combined with spatial preprocessing to exploit spatio-temporal coherence in time-evolving flow fields. The approach compresses differences between consecutive snapshots rather than complete fields, with periodic keyframes to enable fault tolerance and efficient random access. Error compensation prevents accumulation over long sequences, and a distributed block-based implementation enables scalability on massively parallel systems. Evaluation on a Rotating Detonation Rocket Engine (RDRE) simulation with shock-dominated turbulent flow demonstrates compression ratios of around 50x for a 0.1\% relative error tolerance, achieving a 2-3x improvement over independent snapshot compression. For the same target error tolerance, the proposed method also demonstrates a 1.2-2.4x uplift in compression ratio over the widely-used ZFP compressor on delta frames. The method maintains exceptional encoding and decoding performance making it suitable for real-time compression during production CFD simulations.