Hyperspectral imagers
Hyperspectral imagers split reflected light into 100-400+ narrow bands, exposing mineral composition, plant chemistry and trace gases invisible to conventional cameras. The physics is powerful; the data volumes and SNR demands are unforgiving.
What the spectrometer actually measures
A hyperspectral imager is a pushbroom spectrometer. As the satellite moves forward, a narrow slit collects one spatial line of the scene at a time. A diffraction grating or prism then disperses that line across a two-dimensional detector array: one axis is spatial (across-track pixels), the other is spectral (wavelength). The result is a data cube: two spatial dimensions and one spectral dimension, sampled in tens to hundreds of contiguous, narrow bands typically 5–10 nm wide.
That spectral resolution is what separates hyperspectral instruments from multispectral ones. Chlorophyll absorption features near 680 nm and the red-edge inflection near 700–740 nm, the hydroxyl overtone bands that fingerprint clay minerals around 2200 nm, methane absorption near 2300 nm: these are features that require bands narrower than 20 nm to resolve cleanly. Multispectral sensors, with their broad bands, average across these features and lose the diagnostic signal entirely.
The instruments that set the reference standard
ESA's PRISMA satellite, launched in 2019, carries a combined panchromatic and hyperspectral instrument covering 400–2500 nm in 239 bands at roughly 30 m ground sampling distance. Germany's EnMAP, launched in 2022, covers the same VNIR-SWIR range in 228 bands at 30 m GSD with a signal-to-noise ratio specified above 400:1 in the VNIR. Both are purpose-built government science missions and represent the upper end of what a dedicated small-to-medium satellite bus can carry.
The commercial tier has moved quickly. Planet's Tanager-1, launched in 2023 as part of the Carbon Mapper programme, targets the SWIR specifically for point-source methane and CO₂ detection, with a GSD around 30 m and sensitivity designed to detect facility-level emissions above roughly 100 kg/hour. DESIS, a DLR instrument hosted on the ISS, demonstrated that a compact hyperspectral payload (VNIR, 235 bands, 30 m GSD) can operate effectively from a non-dedicated platform, though the ISS inclination limits coverage to latitudes below about 52°.
Smaller form factors exist. Organisations including Satellogic have flown hyperspectral instruments on sub-100 kg platforms, typically trading spectral range or SNR for mass and cost. The trade is real: a 6U cubesat hyperspectral payload may offer 25 bands in the VNIR at SNR below 100:1, which is adequate for broad vegetation indices but not for mineral mapping or gas detection.
Applications: where narrow bands earn their cost
Mineral exploration is one of the strongest use cases. Hydroxyl, carbonate and sulphate minerals each have diagnostic absorption features in the SWIR (1900–2500 nm) that can be mapped from orbit with instruments like EnMAP or PRISMA. Published work using AVIRIS airborne data and PRISMA spaceborne data has demonstrated discrimination of kaolinite, illite and alunite at deposit scale, which is meaningful for directing ground survey effort.
Vegetation chemistry goes beyond the normalised difference vegetation index that a multispectral sensor provides. Leaf nitrogen concentration, chlorophyll content and canopy water status can each be estimated from continuum-removed spectral features, provided the instrument has sufficient SNR and spectral resolution in the red-edge and NIR. This is directly relevant to precision agriculture programmes and to national crop-yield estimation work of the kind Satellize has conducted for the Kingdom of Tonga.
Greenhouse gas detection from orbit is the newest commercial frontier. Methane has strong absorption features near 2300 nm; CO₂ near 1600 nm and 2000 nm. Detecting a single industrial facility requires a GSD of 20–30 m, SNR above 200:1 in the relevant SWIR bands, and careful atmospheric correction. It is demanding, but the Tanager and GHGSat programmes have shown it is achievable with purpose-designed payloads.
Data volume, SNR and the trade-offs that govern everything
A hyperspectral imager generates data at a rate that punishes inattention during mission design. EnMAP produces roughly 4.6 Gbit per minute of imaging at full resolution. A 30-band cubesat payload at lower GSD may produce far less, but the ratio of useful signal to raw volume is still poor compared with a panchromatic camera. On-board processing, spectral binning and lossy compression are all used to manage downlink budgets, and each introduces a degradation that must be characterised before the data is used for quantitative applications.
SNR is the central engineering tension. Narrow spectral bands collect fewer photons per pixel per integration time than broad bands do. To recover SNR, designers can widen the entrance slit (degrading spatial resolution), increase aperture (adding mass and cost), slow the satellite (not possible in LEO) or bin spectral channels (reducing the spectral resolution that justified the instrument in the first place). There is no free solution. EnMAP's 30 cm aperture and careful detector cooling are what allow it to achieve SNR above 400:1; that aperture and thermal system account for a substantial fraction of the satellite's mass and complexity.
Honest limits: what hyperspectral cannot do
Cloud cover is an absolute blocker. Unlike SAR, a hyperspectral imager in the VNIR-SWIR receives no useful signal through cloud. In humid tropical regions, cloud-free revisits at 30 m GSD may be available only a handful of times per year from a single satellite. Constellation approaches or data fusion with SAR are the only mitigations, and neither is trivial.
Atmospheric correction is not optional and not simple. Water vapour, aerosols and surface-pressure variations all imprint on the measured spectrum. Without per-pixel atmospheric correction, surface reflectance retrievals carry errors large enough to invalidate mineral or gas detection. Established methods (ATCOR, 6S, ISOFIT for mineral applications) exist, but they require ancillary atmospheric data and introduce their own uncertainty, typically 2–5% in reflectance, which matters when the diagnostic feature you are chasing is a 10% absorption dip.
Spatial resolution is a genuine constraint. At 30 m GSD, a single pixel over mixed terrain contains contributions from soil, vegetation and shadow simultaneously. Spectral unmixing algorithms can partially separate these contributions, but sub-pixel accuracy depends on assumptions about endmember spectra that are rarely perfectly known. For urban mapping or infrastructure monitoring, 30 m is often too coarse; for regional geology or agricultural monitoring, it is generally adequate. Instruments offering finer GSD, such as airborne AVIRIS-NG at sub-5 m, achieve this through low altitude and large aperture, neither of which is easily replicated in a cost-constrained satellite programme.
Selecting a hyperspectral payload for a national programme
The first question is spectral range. VNIR alone (400–1000 nm) covers vegetation and water applications at lower instrument complexity and mass. Adding SWIR (1000–2500 nm) opens mineral mapping and gas detection but roughly doubles the payload complexity, requires detector cooling to suppress thermal noise, and increases mass significantly. Most government buyers who need both ranges should expect a dedicated small satellite rather than a secondary payload.
The second question is whether a sovereign instrument is necessary or whether data purchase from EnMAP, PRISMA or commercial operators satisfies the requirement. Sovereign ownership makes sense when the application is operationally sensitive, when national coverage frequency is insufficient from shared assets, or when the programme is building domestic technical capacity. For many first-generation national programmes, a data-purchase agreement with a processing pipeline built domestically is a faster and lower-risk path to operational capability than a bespoke instrument. That is a genuine trade to evaluate before committing to a build.
Engineering parameters
| Spectral range (typical) | VNIR: 400–1000 nm; SWIR extension to 2500 nm for mineral/gas applications |
| Number of bands | 100–400 contiguous bands; science missions (EnMAP, PRISMA) use 228–239 |
| Spectral sampling interval | 5–12 nm typical; finer sampling increases data volume without proportional SNR gain |
| Ground sampling distance (spaceborne) | 20–30 m for dedicated small-to-medium satellites; 100–500 m for cubesat-class payloads |
| Signal-to-noise ratio | 200–500:1 (VNIR, science class); below 100:1 for compact cubesat payloads in narrow bands |
| Raw data rate | 1–10 Gbit/min depending on swath, GSD and band count; on-board compression typically 2–4× lossless |
| Payload mass (instrument only) | 15–80 kg for dedicated science/commercial instruments; sub-5 kg for cubesat hyperspectral modules |
| Power consumption | 40–150 W (instrument + cooling) for VNIR-SWIR class; 5–20 W for cubesat-class VNIR-only |
| Detector cooling requirement | SWIR detectors (InGaAs, HgCdTe) typically cooled to 200–220 K; adds mass and complexity |
| Swath width | 30–60 km typical at 30 m GSD from 500–650 km orbit; wider swath trades against spatial resolution |
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. Request a hyperspectral mission scoping call.