DOE CSGF (Krell) - MCS - Rovera, Giovanni - 9.1.26.

Error-bounded lossy compression has become an important technique for reducing the storage and I/O costs of large-scale scientific data while providing strict control over data distortion. Existing compressors typically exploit spatial or temporal correlations through predefined prediction models, which may not effectively capture the complex underlying dynamics of scientific datasets. In this work, we plan to investigate the use of Sparse Identification of Nonlinear Dynamics (SINDy) to improve error-bounded lossy compression. We wukk evaluate the approach on representative scientific datasets and compare it with state-of-the-art error-bounded lossy compressors. Our results will demonstrate the potential of incorporating data-driven SINDy data predictor into scientific data compression and provide insights into when SINDy-based prediction can improve compression ratios and downstream data quality.