The pipeline is the product
A satellite returns raw digital numbers. Before those numbers mean anything, they must pass through radiometric correction, geometric registration, atmospheric compensation and, usually, several domain-specific algorithms. Each of those steps is a software dependency. Taken together they form the exploitation pipeline, and in most national programmes it is the pipeline, not the sensor, that determines how quickly intelligence reaches a decision-maker.
The distinction between a catalogue, a processing pipeline and an analyst tool matters because each layer has different sovereignty implications. A catalogue indexes scenes and controls access; a pipeline converts level-0 data to analysis-ready products; analyst tools let trained staff query, visualise and annotate results. Governments that treat these three as a single procurement often discover, too late, that they have signed away audit rights to the layer that matters most.
Open-source foundations are not free, but they are auditable
ESA's SNAP (Sentinel Application Platform) and the GDAL/OGR library underpin a large fraction of the world's operational satellite processing. Both are open-source, both are maintained by communities with no single commercial owner, and both are auditable line by line. SNAP handles Sentinel-1 SAR processing, Sentinel-2 optical correction and several third-party formats out of the box. GDAL translates between essentially every geospatial raster and vector format in operational use. A national programme built on these foundations can be inspected, forked and modified by the government's own engineers without licence negotiation.
The honest cost of open-source is engineering time, not licence fees. SNAP's batch processing performance is adequate for moderate data volumes but requires careful tuning for high-throughput pipelines. Containerised deployments using Docker or Kubernetes, combined with a workflow orchestrator such as Apache Airflow, are the standard approach for scaling open-source chains to operational tempo. None of this is exotic, but it does require engineers who know what they are doing. Governments that lack that capacity in-house need a delivery partner who will train national staff rather than simply operate the system on their behalf.
Proprietary stacks: capability at the cost of transparency
Commercial exploitation platforms, including ENVI, Hexagon's ERDAS IMAGINE and several cloud-native offerings, offer mature graphical interfaces, well-documented algorithms and vendor support. For a programme that needs to be operational quickly with limited in-house expertise, that is a real advantage. The trade-off is structural: algorithms are not auditable, pricing is set by the vendor, and export-control clauses in licence agreements can restrict which nationals may operate the software or which data may be processed through it.
Cloud-hosted proprietary platforms introduce a further complication. If processing occurs on infrastructure outside the country, even data that was collected by a sovereign satellite may be subject to the legal jurisdiction of the cloud provider's home state during the processing step. For intelligence-adjacent applications, that is not a theoretical concern. It is a condition that should appear explicitly in any legal review before a platform is selected.
Where the honest limits sit
No exploitation software eliminates the underlying limits of the sensor. A 10-metre optical image processed through the most sophisticated atmospheric correction chain is still a 10-metre image. Change-detection algorithms applied to SAR data will flag geometric changes reliably but cannot distinguish a new building from a large vehicle without additional context. Machine-learning classifiers trained on one region frequently underperform in another because of spectral differences in soil, vegetation and built material. Accuracy figures quoted in academic literature are almost always derived from the training region; operational accuracy in a new geography is lower, sometimes substantially.
Temporal gaps compound the problem. Even with a constellation providing daily revisit, cloud cover over tropical regions can produce multi-week data gaps for optical sensors. An exploitation pipeline that does not explicitly handle missing-data periods will produce misleading outputs, particularly in time-series products such as crop-condition indices or flood-extent tracking. Any honest specification for an exploitation system must state how the pipeline behaves when inputs are absent, not only when they are ideal.
Avoiding lock-in on data you already own
The most common form of vendor lock-in in satellite exploitation is not the one governments expect. It is not the licence fee; it is the proprietary data format. When a processing chain writes outputs to a vendor-specific binary format, migrating to a different platform requires re-processing the entire archive. For a programme with several years of accumulated data, that is a significant cost. Specifying open output formats, specifically GeoTIFF with embedded metadata for rasters and GeoPackage or GeoJSON for vectors, from the outset costs nothing and eliminates the problem entirely.
Interoperability standards from the Open Geospatial Consortium, particularly the OGC API suite and the STAC (SpatioTemporal Asset Catalog) specification, have become the practical baseline for catalogue interoperability. A catalogue built to STAC can be queried by any compliant client, including QGIS, Jupyter notebooks and purpose-built government dashboards. Specifying STAC compliance in a procurement is a concrete, testable requirement, not an aspiration. Satellize applied this principle in the Kingdom of Tonga crop-estimation programme, where the analytics outputs were structured to remain queryable by national staff using open tools after the engagement concluded.
What a procurement specification should actually contain
A well-written exploitation software specification names the processing levels required (ESA's Level-1 to Level-3 taxonomy is the standard reference), the output formats, the performance benchmarks in scenes-per-hour at a defined data volume, and the conditions under which source code or algorithm documentation must be disclosed. It states which staff roles will operate each layer and what training is required to reach operational competency. It identifies which components may be hosted outside the country and which must not.
It also names a test. Before acceptance, a defined dataset should be processed through the full pipeline and the outputs compared against an independent reference. That reference might be field measurements, a published validation dataset, or cross-comparison with a trusted external product. Without a named acceptance test, a pipeline is delivered on the vendor's terms, not the government's.
Engineering parameters
| Processing level taxonomy | ESA L0 (raw) through L3 (analysis-ready product); L2A atmospheric correction is the typical operational baseline for optical data |
| Open-source core libraries | GDAL/OGR (raster/vector I/O), SNAP (Sentinel SAR and optical), ORFEO Toolbox (high-resolution optical), QGIS (desktop visualisation) |
| Catalogue interoperability standard | STAC 1.0 (SpatioTemporal Asset Catalog); OGC API – Features for vector query |
| Raster output format baseline | Cloud-Optimised GeoTIFF (COG) with embedded EPSG projection and acquisition metadata |
| Typical pipeline throughput (open-source, tuned) | 50–200 Sentinel-2 scenes per hour on a 16-core server; scales horizontally with containerised deployment |
| Classifier accuracy (operational caveat) | Published land-cover classifiers report 85–95% overall accuracy on training regions; expect 10–20 percentage point degradation in novel geographies without local retraining |
| Minimum analyst training to operational competency | Typically 3–6 months for SNAP/QGIS workflows; 6–12 months for pipeline administration and algorithm adaptation |
| Licence model risk | Proprietary platforms carry export-control and jurisdiction clauses; open-source under Apache 2.0 or GPL carries no use restriction by nationality or geography |
| Archive migration cost (proprietary format) | Full re-processing required on format change; open COG/GeoPackage outputs eliminate this cost at no additional processing overhead |
One contract, one accountable engineer
Commissioned as one programme, not a stack of contracts: spacecraft, launch, ground segment, mission control, training and handover are priced together. Source-access terms and audit rights are agreed in writing before signature. Review your pipeline's sovereignty terms.