Target data sources
The prototype targets heterogeneity across modalities and scales, not arbitrary generic tables. Real integration requires code that retrieves and processes actual provider data. Current fixtures do not meet that definition. No provider records are redistributed in this repository.
| Source | Contribution / modality | Spatial / temporal scale | Access boundary | Current status and rights considerations |
|---|---|---|---|---|
| NEON | Field, organismal, sensors, airborne; vegetation/environment | Organism/plot/site/landscape; product-specific intervals | Future product metadata/API adapter | Synthetic field fixture plus live public product metadata; token CSV adapter implemented, authenticated acquisition/normalization pending; preserve CC BY 4.0 and product citation |
| LTER | Independently designed ecological observations | Plot/ecosystem; long-term series | Future dataset-specific catalog/repository adapter | Planned; schemas and licenses differ by dataset |
| NASA ORNL DAAC | Biomass, carbon, gridded Earth observations | Site/regional/global; product-specific | Future product-specific access adapter | Planned; preserve product/version, spatial support, citation and rights |
| AmeriFlux | Carbon/water/energy flux and meteorology | Tower/ecosystem; subdaily to long-term | Future network/product access adapter | Planned; cross-network definitions and licensing need verification |
| TRY | Organismal/species plant traits | Organism/species; observation-specific time | Future permitted-subset adapter | Planned; taxonomic identity, trait method, units and reuse permission require review |
| CartograPlant | Spatial botanical observations | Organism/geographic; record-specific time | Future contributing-resource adapter | Planned; preserve contributing source rights; no blanket reuse assumption |
| Landsat | Satellite imagery | Pixel/landscape/continental; sensor acquisitions | Future STAC discovery and windowed raster access | Planned; sensor, processing level, cloud masks and asset rights remain explicit |
| Sentinel-2 | Satellite imagery | Pixel/landscape/continental; sensor acquisitions | Earth Search STAC metadata and bounded COG windows; direct stream-to-encoder path | Live native STAC discovery and actual NIR/SCL pixel-window reads verified; DN/scale/offset/sensor/support retained; fixtures remain for CI |
Except for the verified Sentinel and documented NEON boundaries described in the live workflow, these are target access mechanisms rather than verified provider integrations. Before using real data, record the precise source URI, format, access method, version, license, derivative/redistribution permissions, and citation. Unclear rights require human review. Do not bulk-download archives into git. Use the metadata consensus and streaming path for new Sentinel harmonization: documented scale/offset is applied in the reviewed executor, with no raw or harmonized dataset files. Historical snapshots remain explicit legacy examples.
Vertical slices and scientific boundaries
- NEON + Landsat or Sentinel: demonstrate actual field-to-raster spatial and temporal support alignment. Start with a few sites and a short available period; example sites/time belong in configuration, never in core code.
- NEON + AmeriFlux or TRY: test cross-network quantity definitions or cross-scale biological context. Do not join traits to sites by species name alone.
- NEON + ORNL DAAC: test site-to-regional context and spatial support, beyond a nearest-neighbor join.
Food and water tracks may select shared vegetation, traits, environmental, flux, or satellite resources for different questions. Source identity remains separate from scientific use. The third track's scope stays configurable. None of these future scientific use cases is currently encoded as a validated domain model.
Fixture policy
The six committed test records are project-authored synthetic normalized contracts, dedicated to the public domain under CC0-1.0 as stated in fixture metadata. They do not claim to reproduce native NEON or Sentinel schemas. Original identifiers, methods, platform/sensor, support, and source-specific rights survive ingestion. Additional native-schema-compatible fixtures for LTER, ORNL DAAC, AmeriFlux, TRY, CartograPlant, and Landsat remain planned; do not create empty files or imply coverage.
Semantic evidence to inspect by source
These are metadata review targets, not claims that every product exposes the same fields or has a verified ontology mapping. Confirm the actual product/version before access. Only the documented Sentinel/NEON boundaries above are implemented.
| Source / selection status | Evidence to seek and preserve |
|---|---|
| NEON — PROTOTYPE SOURCE | Product variable definitions, units, methods, taxon identifiers, sensor/site metadata, quality flags and sampling intervals |
| LTER — BENCHMARK CANDIDATE | Dataset-specific definitions, methods, EML or other metadata, taxonomic treatments, sampling designs and licenses |
| ORNL DAAC — BENCHMARK CANDIDATE | Product scientific definitions, processing/model documentation, grid/support, version and citation |
| AmeriFlux — BENCHMARK CANDIDATE | Flux/meteorological variable definitions, units, processing and quality conventions, tower footprint and intervals |
| TRY — BENCHMARK CANDIDATE | Trait definitions, taxa, methods, observation context, units and access/reuse terms |
| CartograPlant — UNDER EVALUATION | Contributing source identity, botanical/taxon evidence, locations, collection methods and rights |
| Landsat — FUTURE SOURCE | Sensor/platform, processing level, bands, scale/offset, masks, native pixel support and acquisition |
| Sentinel-2 — PROTOTYPE SOURCE | Preserved native STAC/B08/SCL metadata, calibration coefficients, platform/sensor, baseline, native CRS/pixels and rights evidence |
Ontology annotations, where supplied, remain source assertions rather than approved project mappings. Other candidates include permitted community inventories, trait databases, biodiversity occurrences and flux/satellite pairs. The dataset benchmark framework describes how these could be assessed; it has not selected a production source or flagship.
Scientific validation update — 22 October 2026
The new scientific validation findings retain real graph measurements, canonical requirement diagnostics, scoped dataset/pair assessments and live model results. Earlier unmeasured/planned statements describe the prior architecture snapshot: depth is now measured for three local graphs, while exact scientific mappings, independent ecological validity and the weaker-model hypothesis remain unvalidated. The real Sentinel and NEON–Sentinel cases still require explicit expert decisions; no scientific execution is approved.