TopoSZ: Preserving Topology in Error-Bounded Lossy Compression

Area of Science:

• Scientific data compression
• Computational topology
• Data analysis

Background:

• Existing lossy compression methods prioritize pointwise error, often altering crucial topological features.
• Topological integrity is vital for accurate scientific insights from post hoc analysis of decompressed data.

Purpose of the Study:

• Introduce TopoSZ, an error-bounded lossy compression technique designed to preserve topological features in 2D and 3D scalar fields.
• Ensure the accurate representation of local extrema and level set relations in decompressed data.

Main Methods:

• Developed TopoSZ by integrating topological constraints derived from contour-tree-induced segmentation.
• Customized the error-controlled quantization strategy of the SZ compressor (version 1.4) to incorporate these topological constraints.
• Implemented user control over pointwise error and topological feature loss via global error bounds and persistence thresholds.

Main Results:

• TopoSZ successfully preserves topological features, including types and locations of local extrema and contour tree relations.
• Demonstrated control over both pointwise error and topological consistency during compression.
• Enabled topologically consistent scientific insights from decompressed data.

Conclusions:

• TopoSZ offers a novel approach to error-bounded lossy compression for scientific data, prioritizing topological feature preservation.
• This method enhances the reliability of topological analysis on compressed scientific datasets.
• Facilitates accurate and consistent scientific discovery from large-scale data.