spectral-library 0.6.3

Spectral library build, preparation, and retrieval-based mapping toolkit.

spectral-library

spectral-library is a Python package and CLI for:

• preparing stable row-aligned mapping runtimes from SIAC-style hyperspectral exports
• mapping source-sensor reflectance to target sensors
• reconstructing VNIR, SWIR, or full 400-2500 nm spectra with retrieval-based spectral mapping

Project identity:

• distribution: spectral-library
• import package: spectral_library
• public CLI: spectral-library
• internal maintainer CLI: spectral-library-internal
• current version: 0.6.1
• license: MIT for repository software and repository-authored docs; referenced or redistributed third-party datasets, metadata, and derived artifacts may remain subject to upstream terms

Documentation:

• published site: https://marcyin.github.io/spectral_library/
• getting started: docs/mapping_quickstart.md
• theory: docs/theory.md
• official sensor examples: docs/official_sensor_examples.md
• CLI reference: docs/cli_reference.md
• Python API reference: docs/python_api_reference.md
• prepared runtime contract: docs/prepared_runtime_contract.md
• security and provenance: docs/security_provenance.md
• release process: docs/release_process.md

Install

Install the public package in Python 3.9 to 3.14:

python3 -m pip install .

Optional extras:

• docs site: python3 -m pip install ".[docs]"
• SciPy cKDTree KNN backend: python3 -m pip install ".[knn]"
• FAISS backend: python3 -m pip install ".[knn-faiss]"
• PyNNDescent backend: python3 -m pip install ".[knn-pynndescent]"
• ScaNN backend: python3 -m pip install ".[knn-scann]"
• all ANN extras: python3 -m pip install ".[knn-all]"
• internal build and example-regeneration tools: python3 -m pip install ".[internal-build]"

Quickstart

Download a pre-built runtime (77,125 spectra, MODIS/Sentinel-2/Landsat 8/9):

spectral-library download-prepared-library \
  --output-root build/mapping_runtime

Or build your own from a SIAC-style export:

spectral-library build-mapping-library \
  --siac-root build/siac_library \
  --srf-root path/to/srfs \
  --source-sensor SENSOR_A \
  --source-sensor SENSOR_B \
  --knn-index-backend faiss \
  --output-root build/mapping_runtime

Validate it:

spectral-library validate-prepared-library \
  --prepared-root build/mapping_runtime

Map one sample:

spectral-library map-reflectance \
  --source-sensor SENSOR_A \
  --target-sensor SENSOR_B \
  --input path/to/query.csv \
  --output-mode target_sensor \
  --neighbor-estimator simplex_mixture \
  --knn-backend faiss \
  --output path/to/mapped.csv

Map a batch and write diagnostics:

spectral-library map-reflectance-batch \
  --source-sensor SENSOR_A \
  --target-sensor SENSOR_B \
  --input path/to/query_batch.csv \
  --output-mode target_sensor \
  --neighbor-estimator simplex_mixture \
  --knn-backend faiss \
  --diagnostics-output path/to/diagnostics.json \
  --neighbor-review-output path/to/neighbor_review.csv \
  --output path/to/mapped_batch.csv

If you omit --prepared-root, the mapping CLI downloads the package-matched published runtime into the user cache on first use. Pass --prepared-root when you want to use a custom or locally built runtime instead.

Benchmark a prepared runtime:

python3 scripts/run_full_library_benchmarks.py \
  --prepared-root build/mapping_runtime \
  --source-sensor SENSOR_A \
  --source-sensor SENSOR_B \
  --neighbor-estimator simplex_mixture \
  --knn-backend numpy \
  --knn-backend faiss \
  --max-test-rows 200 \
  --output-root build/benchmark_outputs

For CSV layouts, SRF JSON shape, batch self-exclusion, diagnostics, confidence policy, and Python examples, start with docs/mapping_quickstart.md.

Repository Layout

The repository is now organized around four package areas:

• spectral_library.mapping public mapping runtime, prepared-runtime build, and retrieval engine
• spectral_library.distribution runtime download helpers used by download-prepared-library
• spectral_library.sources maintainer-oriented source manifests, fetchers, fetch batching, and catalog assembly
• spectral_library.normalization maintainer-oriented normalization, coverage filtering, quality plots, and SIAC package export

Compatibility wrappers keep older flat root modules like batch.py, build_db.py, normalize.py, and runtime_download.py importable while the implementation lives in the new package layout.

What The Package Ships

• Stable public Python API for prepare_mapping_library(...), build_mapping_library(...), build_mapping_runtime(...), coerce_sensor_input(...), validate_prepared_library(...), and SpectralMapper
• Public CLI for runtime build, validation, single-sample mapping, batch mapping, and benchmarking
• Prepared runtime format with manifest, checksums, sensor schema, and optional persisted ANN indexes
• Exact numpy KNN plus optional scipy_ckdtree, faiss, pynndescent, and scann backends
• Multiple retrieval estimators including mean, distance_weighted_mean, and simplex_mixture
• Rich diagnostics with neighbor identities, distances, weights, source-band fits, and confidence routing

Official Sensor Examples

The repository includes reproducible cross-sensor examples built from official:

• NASA MODIS response functions
• ESA Sentinel-2A response functions
• USGS Landsat 8 and Landsat 9 response functions

Those examples live in examples/official_mapping/README.md and are documented in docs/official_sensor_examples.md.

Important:

• the official example bundle expects the previously composed full SIAC library recorded in its manifest
• the docs example is a held-out evaluation against that full library, not a vendored miniature retrieval database
• Sentinel-2 uses B08 for the semantic nir band in the published example set

Production Readiness

The repository includes the production-release work needed to ship the package:

• package build and smoke-test workflows
• GitHub Pages docs deployment
• dependency review, pip-audit, and Python CodeQL
• weekly Dependabot update PRs
• pinned GitHub Action SHAs across workflows
• SBOM generation and provenance attestations in the release workflow
• scheduled and manual full-library benchmark workflows

See docs/security_provenance.md, SECURITY.md, and RELEASE.md.

Maintainers

Internal build and source-ingest material is intentionally separated from the public mapping package. If you are maintaining the full library build pipeline, start here:

• internal overview: docs/internal_overview.md
• internal build pipeline: docs/internal_build_pipeline.md
• release notes: docs/releases/0.6.1.md