Calibration and validation
Radiometric and geometric cal/val converts raw satellite counts into defensible physical measurements. Without it, change-detection fails, multi-source fusion breaks, and a government's investment in EO quietly degrades.
What a digital number actually means
Every Earth-observation sensor records a digital number: an integer representing the voltage produced when photons hit a detector. That number is not a physical quantity. It is a count. Converting it into top-of-atmosphere radiance, and then into surface reflectance or brightness temperature, requires a chain of coefficients determined before launch and updated continuously in orbit. Snap that chain and the data becomes a picture. Pictures are fine for human interpretation; they are useless for automated change detection or multi-temporal analysis.
The distinction matters most when a programme spans years. A sensor that drifts by two percent per year produces imagery that looks identical to the eye but will show a spurious 'greening' or 'browning' trend in any vegetation index. Governments investing in agricultural monitoring or forest-cover tracking are effectively buying a trend product. If the calibration is not maintained, the trend is an artefact.
Radiometric calibration: on-board and vicarious
Absolute radiometric calibration has two main routes. On-board calibration uses internal sources: solar diffuser panels (as on Sentinel-2 and Landsat-9), lamp assemblies, or shutter-based dark references. These track relative stability well but can themselves drift, so they must be cross-checked against external references. Cross-calibration against well-characterised heritage sensors, Landsat or Sentinel-2 for optical, is the most practical approach for a new national satellite: the reference sensor has years of validated data and the comparison requires only simultaneous or near-simultaneous overpasses of a stable target.
Vicarious calibration is the field-measurement route. A ground team deploys spectroradiometers at a high-altitude, low-aerosol site, such as Railroad Valley Playa in Nevada or the Saharan pseudo-invariant calibration sites used by CEOS, and measures surface reflectance at the moment of overpass. An atmospheric radiative-transfer model, typically 6SV or MODTRAN, converts that surface measurement to the top-of-atmosphere radiance the sensor should have recorded. The difference between prediction and measurement is the calibration error. Vicarious campaigns are expensive to run (field teams, instruments, logistics) and yield only occasional data points, so they complement rather than replace on-board monitoring. Uncertainty floors for well-executed vicarious campaigns are typically quoted at two to five percent in reflectance, depending on site stability and aerosol loading on the day.
Geometric accuracy is not just about pretty maps
Radiometric accuracy gets the attention, but geometric error is equally destructive to operational use. An image that is consistently displaced by 30 metres is annoying but manageable; an image whose displacement varies across the scene, or between acquisitions, makes pixel-level change detection impossible. Geometric cal/val characterises both absolute geolocation accuracy and relative consistency between scenes.
The standard method uses ground control points: features of known precise position (road intersections, airport markings, coastline inflection points) identified in the imagery and compared with reference coordinates from GNSS survey or a trusted reference dataset. Sentinel-2 achieves sub-pixel geolocation of around 12 metres (1-sigma) after systematic correction; Landsat-9 is specified at under 12 metres circular error 90 percent. A new national sensor should be validated against these benchmarks using the same GCP methodology. Orthorectification using a digital elevation model is a separate but related step: terrain-induced displacement in a 500-km swath optical sensor can reach hundreds of metres in mountainous terrain if uncorrected.
The honest limits of a cal/val programme
Cal/val is not a one-time activity. Sensors age. Detectors degrade at different rates across an array, producing striping. Optical coatings yellow. Solar diffusers accumulate contamination. A mission that funds a thorough pre-launch and commissioning campaign but then cuts the ongoing operations budget for cal/val is making a slow-motion error that will not be obvious until the data has already been used to make decisions.
Cloud is the blunt constraint for vicarious optical campaigns: you need a clear overpass over a stable site at the moment the field team is deployed. High-latitude programmes face short windows and variable aerosol. SAR sensors have their own geometric and radiometric cal/val requirements, including corner-reflector arrays of precisely known radar cross-section, and the processing chain for radiometric terrain correction is non-trivial. Thermal infrared sensors require blackbody references and are particularly sensitive to detector non-uniformity. No programme should assume that a single calibration site covers all spectral bands adequately: a site that is ideal for visible and near-infrared may be spectrally variable in the shortwave infrared.
Absolute accuracy also has a hard floor set by the atmospheric correction model, not the sensor. Even with perfect on-board calibration, surface reflectance retrievals carry a residual uncertainty from aerosol optical depth estimation. For most operational applications, two to three percent uncertainty in surface reflectance is acceptable. For hyperspectral mineral mapping or water-quality retrieval, it is not.
Institutional obligations and the CEOS framework
The Committee on Earth Observation Satellites coordinates a global cal/val framework through its Working Group on Calibration and Validation. Membership in CEOS is not mandatory for a national programme, but alignment with its protocols matters practically: data that cannot be compared with Landsat or Sentinel-2 will not be accepted into multi-source analytical chains used by international agencies, food-security monitors or climate services. A government that wants its satellite data to carry weight in international forums needs to demonstrate that the numbers are traceable.
Pseudo-invariant calibration sites recognised by CEOS include sites in the Sahara, the Atacama, Libya and the Arabian Peninsula. Access to some requires coordination with national authorities. A programme that builds its cal/val plan around a single site carries concentration risk: if the site is inaccessible or experiences an unusual atmospheric event on overpass day, the campaign is lost. Three to four sites across different continents, with a mix of vicarious and cross-calibration methods, is the minimum for a credible ongoing programme.
Engineering parameters
| Radiometric accuracy target (optical) | ±2–5% absolute reflectance (vicarious); ±1–2% relative stability (on-board monitoring) |
| Geolocation accuracy benchmark | <12 m CE90 (Landsat-9 / Sentinel-2 standard); new sensors validated against same GCP dataset |
| Vicarious calibration campaign frequency | Typically 2–4 field campaigns per year during commissioning; 1–2 per year operationally |
| On-board calibration update cadence | Monthly to quarterly coefficient updates typical for well-maintained optical missions |
| Corner-reflector size (SAR cal/val) | Inner leg 1–3 m depending on wavelength; trihedral or dihedral depending on polarisation requirement |
| Atmospheric correction residual uncertainty | 1–3% surface reflectance under low-aerosol conditions; higher in humid or dusty atmospheres |
| Spectral calibration stability (hyperspectral) | ±0.5–1 nm wavelength shift acceptable; beyond this, mineral-mapping products degrade materially |
| Cal/val site stability requirement | Temporal reflectance variation <1% per year; pseudo-invariant sites verified by CEOS meet this |
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 mission's cal/val plan.