Turn Sentinel scenes into observations your software can trust
A Copernicus integration discovers the right acquisitions, applies explicit optical or SAR processing, measures usable coverage, and returns parcel- or asset-level results with enough provenance to reproduce them. The deliverable is not a folder of imagery—it is a stable data contract.
There is no single Copernicus API
CDSE exposes complementary interfaces. The official API overview makes the separation explicit, while the STAC documentation notes that STAC complements rather than replaces OData. Pick by operation, collection coverage, latency, output, and scale.
STAC 1.1
Discover available collections and items through a standard geospatial model. Use the current /v1 endpoint, interrogate collection queryables, follow returned links, and test CQL2, fields, sort, and pagination behavior.
Interoperable metadata search
OData
Search the wider product catalogue, expand attributes and locations, list product nodes, and download authenticated products by UUID. Treat catalogue and download hosts as separate dependencies.
Detailed metadata and products
S3-compatible access
Read EOData objects through account-specific access keys when parallel access and object-level control matter. Keep endpoint, credentials, concurrency, checksums, and retry policy configurable.
Bulk and parallel reads
Sentinel Hub
Use Process for pixels or rasters, Statistical for aggregates, and async or batch variants for larger workloads. Version evalscripts and every processing option that can change a result.
Server-side raster processing
openEO
Express collection loading, filtering, reduction, and export as a portable process graph. Confirm the chosen backend, collection IDs, supported processes, credits, and synchronous versus batch limits.
Reusable analysis graphs
Your observation API
Present one domain contract to claims, farm, logistics, or monitoring systems. Provider item IDs, processing versions, quality fields, and lineage remain visible without leaking provider choreography into every client.
Stable downstream integration
Migrate semantics, not just URLs
CDSE withdrew OpenSearch/Resto on 2 March 2026. Its decommissioning notice points integrations to STAC or OData. Sentinel Hub also introduced service-first paths such as /process/v1 and /catalog/v1 in March 2026; both formats currently work, but the official update says legacy paths will be deprecated later. Keep endpoints in configuration and run contract fixtures against real responses.
Build a replayable observation pipeline
The same parcel and interval should lead to an explainable set of candidate scenes and a reproducible result. Preserve the source identity, selection rule, processing configuration, quality mask, algorithm, and release—not only the final number.
Register the area and question
Validate polygon topology, CRS, tenant ownership, asset or parcel version, observation interval, permitted sensors, business purpose, and expected latency before querying CDSE.
Discover candidate acquisitions
Search by collection, intersection and time. Page by provider links or tokens, deduplicate by collection and item ID, and capture publication or processing metadata separately from sensing time.
Select with explicit rules
Apply optical cloud limits, orbit and polarization requirements, timeliness, coverage, baseline, and product availability. Record every exclusion instead of silently taking the first scene.
Process deterministically
Pin bands, units, CRS, pixel size, resampling, mosaicking order, cloud mask, SAR coefficient, orthorectification, terrain correction, temporal composite, and code or evalscript version.
Aggregate with quality
Intersect the processed raster with the authoritative geometry and return usable-pixel fraction, coverage, observation count, distribution statistics, flags, and null reasons beside each feature.
Publish and reconcile
Write an idempotent observation event, cache immutable derived assets, expose provenance to clients, monitor provider changes, and replay affected outputs when a source or algorithm changes.
Current public STAC search
Illustrative discovery requestPOST https://stac.dataspace.copernicus.eu/v1/search
Content-Type: application/json
{
"collections": ["sentinel-2-l2a"],
"intersects": { "type": "Polygon", "coordinates": [/* parcel */] },
"datetime": "2026-08-01T00:00:00Z/2026-08-31T23:59:59Z",
"filter": {
"op": "<=",
"args": [{ "property": "eo:cloud_cover" }, 40]
},
"limit": 100
}Collection-level cloud cover only narrows candidates. The pipeline must still measure valid pixels over the requested geometry. At this review, live CDSE collection responses report STAC 1.1.0 for both sentinel-2-l2a and sentinel-1-grd.
Optical and radar need different contracts
A generic “satellite score” hides the conditions that make a time series comparable. Expose sensor-specific acquisition and processing fields all the way to the derived observation.
Sentinel-2 Level-2A optical
Level-2A supplies atmospherically corrected surface reflectance and a Scene Classification Layer. The official S2L2A reference documents bands, units, resolution variants, mosaicking, and SCL values. Use those inputs rather than treating RGB previews as analysis data.
- Selection
- Sensing time, tile, baseline, cloud metadata, coverage
- Processing
- Bands and units, mask, resampling, CRS, composite
- Quality
- Valid-pixel, cloud, shadow, snow, no-data fractions
- Example feature
- NDVI = (B08 − B04) / (B08 + B04), with null rules
Sentinel-1 GRD radar
GRD makes cloud-independent monitoring possible, but raw detected amplitude is not a ready business signal. The official S1GRD processing reference describes calibration, backscatter coefficients, optional speckle filtering, orthorectification, and radiometric terrain correction.
- Selection
- Mode, VV/VH or HH/HV, orbit direction, relative orbit
- Processing
- Coefficient, linear or dB, DEM, RTC, orthorectification
- Quality
- Coverage, border/no-data, terrain shadow, incidence context
- Example feature
- Calibrated VV/VH change against a comparable reference
Fusion happens after comparable preprocessing
Align optical and radar observations by geometry, temporal window, crop or asset context, and declared quality. Never forward-fill a missing optical metric with radar as though they measure the same thing. A fused model needs training and evaluation across regions, seasons, crop types, sensors, acquisition regimes, and known edge cases.
Quality is part of the result
A value without its observable area, temporal basis, source, and processing version is unsafe to automate. Return a structured null or review state when the observation does not support the requested inference.
| Risk | Evidence | Contract behavior |
|---|---|---|
| Parcel obscured | Valid, cloud, shadow, snow, and no-data fractions over the parcel | Return null or low-coverage status; do not publish an index as complete |
| Partial intersection | Requested versus observed area and raster footprint | Expose coverage ratio and the geometry version used |
| Incomparable SAR | Mode, polarization, orbit direction, relative orbit, coefficient, terrain options | Partition the series or reject the comparison |
| Provider throttling | HTTP status, rate-limit headers, attempts, processing units, queue age | Retry with jitter, defer, or serve a declared cached result |
| Source reprocessing | Item ID, checksum, product baseline, publication time, algorithm version | Mark affected observations and replay by policy |
| Weak business inference | Feature distribution, reference window, model version, validation scope | Route to review; keep sensor evidence distinct from the business decision |
Authentication and quota behavior
Sentinel Hub uses OAuth 2.0 client credentials. The authentication guide says to reuse a token through its validity rather than fetch one per request. The rate-limit guide distinguishes request and processing-unit limits and documents HTTP 429 handling. Keep secrets out of browser clients and expose queue depth, cache hit rate, 429s, and provider latency.
A downstream observation, not a mystery score
{
"parcelId": "field-cz-1842@7",
"interval": "2026-08-01/2026-08-31",
"status": "review",
"metrics": { "ndviMedian": 0.64, "validPixelFraction": 0.78 },
"source": {
"collection": "sentinel-2-l2a",
"itemId": "S2B_MSIL2A_...",
"sensingTime": "2026-08-10T10:10:19Z"
},
"provenance": {
"selectionPolicy": "s2-monthly@4",
"mask": "scl-agri@3",
"algorithm": "parcel-observation@12"
},
"flags": ["PARTIAL_CLOUD"]
}Prove one observation journey first
Start with a bounded geography, sensor pair, historic interval, latency target, and downstream decision. A pilot should prove that the same input can be discovered, processed, explained, and replayed before expanding coverage.
Input contract
Authoritative AOI and version, observation question, time window, sensor policy, output units, latency, and retention.
Provider adapter
Current STAC/OData discovery, OAuth and S3 credentials, response normalization, pagination, retries, caching, and endpoint configuration.
Processing specification
Optical and SAR parameter sets, masks, spatial and temporal aggregation, deterministic containers or evalscripts, and provenance fields.
Validation corpus
Known parcels or assets across clear, cloudy, partial, terrain, missing-data, reprocessed, and provider-failure cases with expected results.
Operational controls
Idempotency, work queue, concurrency, token cache, quota telemetry, dead-letter handling, replay, drift alerts, and cost per observation.
Consumer API
Versioned schema, quality and null semantics, provenance, derived-asset links, review state, webhook or batch delivery, and service objectives.
Open data still needs a licence ledger
CDSE states that Copernicus Sentinel data is available on a free, full, and open basis under the Sentinel Data Legal Notice. The terms distinguish that data from other portal content and datasets. Record the licence URL, acquisition year, source notice, modification status, provider, and any collection-specific restrictions with each derived asset.
Sentinel source
Use the prescribed Copernicus Sentinel data notice when distributing or communicating data.
Modified output
Use the modified-data wording where Sentinel data has been adapted or combined.
Other collections
Check Contributing Mission, DEM, service-information, and third-party terms separately.
Copernicus API integration FAQ
Which Copernicus API should an application use?
Use the interface that matches the workload. STAC 1.1 is the interoperable discovery layer for its available collections. OData supports detailed product search, node listing, and authenticated product download. S3 is suited to parallel object access. Sentinel Hub Process and Statistical APIs return processed pixels or aggregates without requiring every raw product locally. Async, batch, or openEO paths are better for larger jobs. A production connector can use more than one while exposing one stable business contract.
Is the old Copernicus OpenSearch or Resto catalogue still available?
No. CDSE withdrew the OpenSearch Catalogue API on 2 March 2026 and directed integrations to STAC or OData. The separate legacy STAC endpoint under catalogue.dataspace.copernicus.eu was deprecated from 17 November 2025; the current STAC base is https://stac.dataspace.copernicus.eu/v1/. Migrations should replace endpoint strings and re-test collection identifiers, filters, fields, sorting, and pagination semantics.
Can global scene cloud cover tell whether a farm parcel is clear?
No. Scene-level cloud cover is useful for coarse catalogue filtering, but clouds and shadows may still cover the actual parcel. For Sentinel-2 Level-2A, calculate parcel-level usable-pixel coverage from the Scene Classification Layer or an approved cloud mask and return that quality measure with every metric. Keep the masking algorithm and version in provenance.
Does Sentinel-1 SAR solve every cloudy-image problem?
No. Radar can observe through cloud and does not depend on daylight, but useful comparison still requires consistent acquisition mode, polarization, orbit direction, relative orbit, calibration, geometry, and terrain treatment. Speckle and terrain effects remain. Optical and radar features are complementary signals, not interchangeable measurements.
Can NDVI or radar backscatter prove crop damage for an insurance claim?
Not by itself. A vegetation index or backscatter change is an observation derived from a specific sensor, acquisition, mask, algorithm, and reference period. It may support triage or review, but attribution can be confounded by crop stage, management, soil moisture, geometry, clouds, terrain, and data gaps. Consequential decisions need a validated model, other evidence, documented thresholds, and a human-review path.
Should the integration download and retain every SAFE product?
Not necessarily. Retain the minimum inputs needed for repeatability, contractual evidence, latency, and cost. Some systems store item identifiers, checksums, processing parameters, quality outputs, and derived assets while re-reading source data when needed; others require an immutable local archive. Make that an explicit retention and replay decision rather than an accidental by-product of the first prototype.
Is Copernicus Sentinel data free for commercial products?
Copernicus Sentinel data is provided on a free, full, and open basis under the Sentinel Data Legal Notice, including adaptation and combination for lawful uses. Public distribution or communication carries source-notice requirements, and modified data needs the specified modified-data notice. Do not assume that every Copernicus Contributing Mission dataset, DEM, third-party layer, or item elsewhere on the portal has identical rights; record the collection-specific licence and attribution text.
How should OAuth tokens and CDSE quotas be handled?
Keep client secrets server-side, cache each access token until shortly before its expiry, and never request a new token for every API call. Model request and processing-unit limits separately, read rate-limit response headers, honor HTTP 429 with bounded exponential backoff and jitter, cache deterministic outputs, and queue workloads. Quotas are account-specific and can change, so configuration and observability are safer than hard-coded limits.
Primary sources and status
Reviewed 4 September 2026. CDSE documentation and operational notices define the current interfaces, endpoints, collection behavior, authentication, rate limiting, and migrations; these are provider specifications and may change. The CDSE terms and linked Sentinel Data Legal Notice govern use and attribution within their scope. Yarify's pipeline, quality contract, storage choices, validation gates, and pilot criteria are engineering recommendations, not guarantees of agronomic, insurance, or legal outcomes.
- Copernicus Data Space Ecosystem — API overviewOfficial map of catalogue, download, streamlined-access, and processing interfaces available to application teams.
- CDSE STAC product catalogueOfficial STAC 1.1 documentation, current endpoint, collection identifiers, queryables, CQL2 filters, pagination, and supported extensions.
- CDSE OData catalogue and product downloadOfficial OData query, product metadata, node listing, authentication, and download documentation.
- CDSE access to Earth-observation data through S3Official guidance for scalable object access through the eodata S3-compatible endpoint and account-specific access keys.
- Sentinel Hub API authenticationOfficial OAuth 2.0 client-credentials flow, token endpoint, expiry handling, and instruction to reuse tokens instead of requesting one per call.
- Sentinel Hub API rate limitingOfficial behavior for request and processing-unit limits, HTTP 429 responses, response headers, caching, backoff, and account-specific policies.
- CDSE Sentinel Hub API path-structure updateOfficial March 2026 notice introducing service-first paths such as /process/v1 and /catalog/v1 alongside legacy paths, with future legacy deprecation planned.
- CDSE OpenSearch Catalogue API decommissioningOfficial notice confirming the Resto/OpenSearch catalogue endpoint was withdrawn on 2 March 2026 and integrations should use STAC or OData.
- Sentinel-2 Level-2A data in Sentinel HubOfficial band, unit, scene-classification, resolution, filtering, mosaicking, and processing documentation for surface-reflectance products.
- Sentinel-1 GRD processing in Sentinel HubOfficial SAR processing chain and options for calibration, backscatter coefficients, speckle filtering, orthorectification, and radiometric terrain correction.
- Copernicus Data Space Ecosystem terms and conditionsOfficial terms distinguishing free, full, and open Sentinel data from other portal content and linking the governing Sentinel Data Legal Notice.