Skip to content

Explanation

Background material that helps you understand why Rasteret works the way it does.

  • Architecture: component overview and data flow
  • Design Decisions: why Parquet, why a custom reader, why COGs stay COGs, and why these choices enable an open community
  • Correctness Contract: the user-visible guarantees Rasteret makes (and the contract contributors must preserve)
  • Schema Contract: column guarantees for contributors building ingest drivers
  • Ecosystem Comparison: how Rasteret relates to TorchGeo, xarray, rasterio, and similar tools
  • Benchmarks: apples-to-apples performance comparison
  • Compatibility Matrix: supported Python, dependency, and format versions
  • TorchGeo Benchmark: interactive notebook - Rasteret vs TorchGeo native, side-by-side

Every design choice here (Parquet indexes, open catalog descriptors, decoupled storage backends) is made to keep Rasteret interoperable and community-driven. The catalog format is intentionally spec-aligned and evolving toward a portable standard shareable across tools, not just Rasteret.

These are deliberate trade-offs, not accidental ones. If you have ideas, see something that could work differently, or want to discuss the reasoning, open a Discussion. The design benefits from more perspectives.