QoI-Preserving Lossy Compression for Turbulent Flows and Combustion

High-fidelity simulations of turbulent flows and combustion generate massive datasets, creating significant challenges in I/O (∼20 GB/s), storage (∼100 GB per snapshot), and post-processing (∼1000 snapshots per run). In this work, we present a compression workflow that applies lossy compressors with and without quantity-of-interest (QoI) preservation modes to multi-terabyte DNS datasets of reactive flows. Compression ratios from O(10) to O(1000) are achieved depending on variable-specific error bounds (10⁻⁶–10⁻²). QoIs such as laminar flame speed, heat release rate, and species reaction rates derived from primary simulation fields are evaluated to quantify the impact of compression on flow physics. Relative L₂ errors in these QoIs remain within O(10^-1) to O(10^-3), indicating high fidelity in the compressed representations. This general approach enables scalable I/O, reduced storage, efficient restart capability, and improved in situ analysis workflows, while preserving physically meaningful quantities in large-scale reactive flow simulations.