spectrAccess

spectrAccess is a Python client that finds and downloads spectral and atmospheric reference data from RadCalNet, GSICS, AERONET, CAMS, Landsat, Sentinel-2 and NASA EMIT, and parses each into a pandas DataFrame. GSICS, RadCalNet and AERONET measurements can also be returned in one shared schema where every value carries an uncertainty record and its source, URL and retrieval time (see Canonical schema below). Sentinel-2, EMIT and Landsat return scene metadata and leave pixels to your tools. CAMS returns a table of the downloaded files.

It is written for remote sensing and Earth observation scientists, calibration engineers and satellite data providers who spend too much time on portal logins, file formats and unit conventions instead of the comparison itself.

spectrAccess fetches data on your behalf, using your own free account where a source needs one. It never re-serves or redistributes source data; each connector's DATA_TERMS.md lists that provider's terms and citation requirements.

Install

pip install spectraccess

For local development:

pip install -e ".[test]"

Large-archive adapters keep their maintained provider clients optional. For Sentinel-2 discovery and download through CDSETool:

pip install "spectraccess[cdse]"

For CAMS EAC4 access through ECMWF's maintained CDS API client:

pip install "spectraccess[cams]"

For EMIT L1B/L2A discovery and download through NASA earthaccess:

pip install "spectraccess[emit]"

The EMIT extra requires Python 3.12 or newer (the requirement of the reviewed earthaccess==0.18.0 client). The base package and other extras continue to support the Python versions declared in the package metadata.

For Landsat Collection-2 discovery and download through EODAG/USGS:

pip install "spectraccess[landsat]"

Quickstart

from spectraccess.connectors.gsics import GSICSConnector

connector = GSICSConnector()
datasets = connector.discover()
target = datasets[0]
raw = connector.fetch(target)
table = connector.parse(raw)
print(table.head())

Credentials

Run spectraccess login <provider> once for personal access, or hand over a credential source in code. See credentials and migration to 0.2. Credential environment variables and provider-client login files are no longer read.

Connectors

Sensor links in the table go to EO-Atlas, SpectraWorks' open catalogue of Earth observation satellites, sensors and data products, for background on each instrument.

NOAA/NESDIS GSICS products are also mirrored on the EUMETSAT collaboration server's master THREDDS catalog (nesdisProducts.xml), so some NESDIS product families may already be reachable via the EUMETSAT connector default even while the canonical NOAA STAR host is down.

Landsat (USGS), RadCalNet, and CAMS via ADS have fixture tests, but they have not yet been exercised against the live services in CI: they need account credentials, and the weekly live checks skip them until those are configured. The free JASMIN CAMS mirror has no files after 2025-10-03, so recent CAMS dates need an ADS account and token.

Canonical schema

Alongside each connector's native parse() output, connectors can additionally emit a shared, versioned, long/tidy canonical schema (spectraccess.core.schema, currently SCHEMA_VERSION = "1.0") so downstream tools can consume any source through one stable contract: one row per quantity value plus its uncertainty record. GSICS, RadCalNet and AERONET expose this via to_canonical(native_frame, ...) and the connector convenience method parse_canonical(raw, ...); the Sentinel-2 CDSE, EMIT and Landsat connectors emit canonical scene metadata via target_to_canonical(target). CAMS returns its native frame only.

Uncertainty is a record, not a bare number: unc_value may be null, but unc_status never is. unc_status is one of:

• provided: the source itself supplied the uncertainty.
• derived: computed from other quantities by a downstream tool (no connector emits this yet).
• prior: an assumed/prior uncertainty, not measured for this row.
• unknown: no uncertainty value is available (unc_value is null).

RadCalNet .output files carry an absolute, dimensionless uncertainty value for each wavelength and observation. The native frame preserves it as toa_reflectance_unc together with toa_reflectance_unc_status, toa_reflectance_unc_provider, and toa_reflectance_unc_k. Positive source values are provided; negative values are climatological magnitudes and are therefore prior; fill or absent values are unknown. RadCalNet R2 does not state a coverage factor, so toa_reflectance_unc_k and canonical unc_k stay null. spectrAccess never substitutes a fixed percentage or assumes k=1. .input files hold the surface reflectance the TOA values were propagated from, so they parse to surface_reflectance and its matching _unc columns (canonical quantity="surface_reflectance").

The Sentinel-2 CDSE connector canonicalizes the provider's scene cloud-cover metadata as quantity="scene_cloud_cover". CDSE does not publish a numerical uncertainty for that field, so it is honestly labelled unc_status="unknown" with null unc_value, unc_k, and unc_provider. Product identity, footprint, processing version, catalogue/download URLs, and the exact provider metadata remain attached as provenance. spectrAccess does not parse SAFE pixels or reimplement CDSE transport; those stay with CDSETool and downstream consumers.

The CAMS connector returns a CAMSResult that names base_dir (the directory containing the date subtree) and date_dir (the YYYY_MM_DD subtree) separately. This is an intentional typed contract: consumers such as SIAC that append the date must use base_dir, while format converters can work inside date_dir. requested_source, resolved_source, dataset URL, retrieval time, cache status, and exact local assets are retained as native provenance. The connector retrieves source assets only; atmospheric-correction and model-specific format conversion remain downstream responsibilities. Constructor arguments override environment variables: CAMS_SOURCE, CAMS_FORECAST_CYCLE, CAMS_FORECAST_LEAD_HOURS, SPECTRACCESS_CAMS_CACHE_DIR (default ~/.cache/spectraccess/cams), and SPECTRACCESS_CAMS_FALLBACK_URL (a second mirror with the JASMIN layout).

The EMIT connector likewise keeps the multi-gigabyte science cubes opaque. Its canonical output covers only source-provided scene metadata (cloud cover and solar angles), each labelled unc_status="unknown" because CMR does not publish an uncertainty for those metadata values. Exact collection/native IDs, footprint, orbit/scene, asset URLs, byte sizes, and SHA-512 checksums remain in the target provenance. Cube/GLT interpretation and scientific admission are deliberately downstream concerns; connector availability is not a claim-grade endorsement.

The Landsat connector applies the same boundary to Collection-2 L1TP products: EODAG owns USGS search, authentication, retries, and download transport; spectrAccess preserves the provider product ID, display ID, collection number, tier (T1, T2, or RT), WRS-2 path/row, footprint, provider metadata, and a stable cache identifier. Its canonical row describes only provider scene cloud cover, with unknown uncertainty; archive pixels remain a downstream concern.

Call spectraccess.core.schema.validate(df) to check a frame against the schema; it raises SchemaError naming every violation found. Extra, connector-specific columns are always allowed and pass through validation untouched.

Maintainer: SpectraWorks B.V. Built by SpectraWorks, makers of RefCal, the cross-sensor calibration layer. Also from SpectraWorks: EO-Atlas, a catalogue of Earth observation satellites, sensors and data products.

spectrAccess code is licensed under Apache-2.0. Source data remains governed by each external portal's own data terms; see each connector's DATA_TERMS.md.