Sensors
- PRISMA (ASI): Italian Space Agency hyperspectral mission. 30 m spatial resolution, 400–2500 nm range, ~10 nm spectral sampling, 30 km swath. Revisit roughly 29 days at nadir but programmable tasking can improve this. The key nitrogen absorption windows at 1510 nm and 2050 nm fall cleanly within its SWIR detector.
- DESIS (DLR / Teledyne): Mounted on the ISS, so revisit is irregular and orbit-dependent rather than fixed. 30 m resolution, 400–1000 nm range only. Covers the chlorophyll red-edge and some protein-linked features but cannot reach the 1510 nm or 2050 nm nitrogen bands. Useful for chlorophyll-proxy nitrogen estimation, not full PROSPECT inversion.
- EnMAP (DLR): German hyperspectral satellite launched April 2022. 30 m resolution, 420–2450 nm, ~6.5 nm sampling in VNIR and ~10 nm in SWIR. Covers both primary nitrogen absorption windows. 30 km swath, 27-day revisit at nadir. Signal-to-noise ratio is high enough to support radiative-transfer inversion with reasonable uncertainty.
- AVIRIS-NG (NASA): Airborne instrument, not a satellite. Spatial resolution typically 3–8 m depending on flight altitude, 380–2510 nm, ~5 nm sampling. The reference standard for canopy nitrogen retrieval in published validation studies. Used to build and validate the look-up tables that spaceborne missions inherit. Not operationally scalable to continental areas.
- CHIME (ESA, forthcoming): Copernicus Hyperspectral Imaging Mission for the Environment, planned for the early 2030s. Target: 20 m resolution, 400–2500 nm, revisit of around five days through a constellation of two units. Designed explicitly for vegetation biochemistry retrieval at continental scale. Specifications are still subject to change in the mission design phase.
Why nitrogen shows up in reflected light
Nitrogen is bound into chlorophyll, Rubisco and other leaf proteins. Those proteins absorb shortwave infrared radiation at characteristic wavelengths: the N-H bond stretch produces a broad feature near 1510 nm, and a secondary combination band appears around 2050–2100 nm. At the leaf scale, the PROSPECT radiative-transfer model treats leaf nitrogen content (expressed as leaf mass per area, LMA, and nitrogen concentration per unit dry mass) as a parameter that shapes reflectance and transmittance across these windows. The physics is well established; the difficulty is scaling it.
Canopy reflectance is not leaf reflectance. Between the leaf and the satellite sensor sit layers of canopy structure, shadow, understorey, soil background and atmosphere. The SCOPE model (Soil Canopy Observation of Photosynthesis and Energy fluxes) couples PROSPECT leaf optics with a canopy-level radiative-transfer scheme and a sun-sensor geometry model. Inverting PROSPECT-SCOPE against measured hyperspectral reflectance is the standard retrieval pathway, but it is computationally demanding and the inversion is ill-posed: different combinations of leaf nitrogen, LMA, water content and canopy structure can produce near-identical top-of-canopy spectra.
What the retrieval actually delivers, and where it breaks
In single-species or low-diversity stands, PROSPECT-SCOPE inversion against EnMAP or PRISMA data can retrieve canopy nitrogen content with root-mean-square errors in the range of 0.2–0.4 g N m⁻² in published validation studies using AVIRIS-NG and field data. That is a useful signal for forest health screening. In mixed-species canopies, the picture is less clean.
Species-specific LMA varies by a factor of three or more across common temperate and tropical tree species. Because LMA and nitrogen concentration are partially covariate in the PROSPECT parameterisation, the inversion cannot cleanly separate them without species-composition priors. A canopy dominated by thick-leaved oaks will produce a different spectral signature from one dominated by thin-leaved birches, even at the same canopy nitrogen content per unit ground area. Without a species map as an input layer, retrieved nitrogen values carry an additional uncertainty that is difficult to bound tightly.
Cloud cover is a hard constraint. PRISMA and EnMAP are passive optical instruments; a single cloud-free acquisition over a target area in humid tropical or temperate maritime climates may require weeks of tasking attempts. Atmospheric correction also introduces uncertainty in the 1510 nm and 2050 nm windows, where water vapour absorption is strong. The ATCOR and L2A processors used for PRISMA and EnMAP reduce this, but residual errors propagate into retrieved nitrogen values.
The canopy-to-leaf scaling problem in practice
Even a perfect leaf-level nitrogen retrieval needs to be expressed as a canopy-level quantity before it is useful for forest health or carbon-cycle applications. The two common currencies are nitrogen per unit leaf area (g N m⁻² leaf) and nitrogen per unit ground area (g N m⁻² ground). Converting between them requires leaf area index (LAI). LAI is itself retrieved from the same hyperspectral data or from ancillary multispectral imagery, and it carries its own uncertainty, typically ±0.5–1.0 LAI units in closed canopies.
Uncertainty propagation through this chain is non-trivial. A 15% error in LAI combined with a 20% error in leaf nitrogen concentration compounds to a canopy nitrogen uncertainty that can exceed 30% in absolute terms. For carbon-cycle modelling, where canopy nitrogen is used to parameterise maximum carboxylation capacity (Vcmax) via the well-documented Vcmax-nitrogen relationship, this uncertainty translates directly into uncertainty in gross primary production estimates. Users should request uncertainty maps alongside point-estimate outputs, not just mean retrieved values.
Forest health screening and carbon modelling: where the signal earns its keep
Nitrogen deficiency in forest canopies is an early indicator of soil acidification, drought stress and certain pest pressures. Broadband vegetation indices cannot distinguish nitrogen stress from water stress or chlorophyll dilution; hyperspectral retrieval, even with its uncertainties, provides a more specific signal. Repeat acquisitions over the same stand, spaced weeks to months apart, reveal directional change in canopy nitrogen that is diagnostically useful even when absolute accuracy is limited.
In carbon-cycle modelling, the link between canopy nitrogen and photosynthetic capacity is one of the stronger empirical relationships in plant physiology. Global vegetation models increasingly use spatially explicit Vcmax maps derived from remote-sensing nitrogen retrievals to reduce uncertainty in terrestrial carbon flux estimates. PRISMA and EnMAP data have been used in published research to generate these inputs at 30 m resolution, which is finer than anything available from broadband sensors. CHIME, when operational, is expected to make this feasible at continental scale with meaningful revisit frequency.
Satellize applies PROSPECT-SCOPE inversion pipelines to EnMAP and PRISMA archives for clients requiring stand-level nutrient diagnostics, building on the same radiative-transfer framework used in its Tonga crop-estimation programme to handle the leaf-to-canopy scaling step.
Honest limits and the path to better retrievals
No current spaceborne hyperspectral mission delivers daily revisit. For monitoring applications that need to track nitrogen dynamics through a growing season, the 27–29 day nadir revisit of EnMAP and PRISMA is often the binding constraint, not retrieval algorithm quality. Tasking both sensors over the same area on offset schedules can compress the effective revisit to roughly two weeks in cloud-free conditions, but this requires coordinated acquisition planning.
Machine-learning emulators of PROSPECT-SCOPE, trained on synthetic spectra from the full model, reduce inversion time by orders of magnitude and are now standard in operational pipelines. They do not reduce the fundamental ill-posedness of the inversion; they simply make it faster to explore the solution space. Regularisation using species composition maps, LiDAR-derived canopy structure or prior distributions from field campaigns materially improves retrieval precision. Where those ancillary data exist, they should be used. Where they do not, the uncertainty bounds on retrieved nitrogen should be reported honestly rather than suppressed.
Typical figures
| Spatial resolution (spaceborne) | 30 m (PRISMA, EnMAP, DESIS); 20 m planned (CHIME) |
| Spectral range covering N-absorption windows | 420–2450 nm (EnMAP); 400–2500 nm (PRISMA); DESIS limited to 400–1000 nm |
| Spectral sampling interval | ~6.5–10 nm (EnMAP, PRISMA); ~2.5 nm (DESIS VNIR only) |
| Revisit at nadir | 27 days (EnMAP); ~29 days (PRISMA programmable); irregular (DESIS/ISS); ~5 days planned (CHIME, two-unit constellation) |
| Swath width | 30 km (PRISMA, EnMAP); 30 km (DESIS) |
| Key nitrogen absorption bands | ~1510 nm (N-H stretch), ~2050–2100 nm (combination band); red-edge 700–730 nm (chlorophyll proxy) |
| Typical retrieval RMSE (single-species stands) | 0.2–0.4 g N m⁻² canopy (published AVIRIS-NG validation studies; spaceborne results vary) |
| Cloud sensitivity | Complete data loss under cloud; atmospheric water vapour introduces residual error in SWIR windows even after correction |
| Archive depth | EnMAP: from April 2022; PRISMA: from 2019; DESIS: from 2018 |
| Delivery formats | GeoTIFF nitrogen and uncertainty rasters; netCDF time-series stacks; vector stand-summary statistics |
Analytics Satellize can run
| Canopy nitrogen concentration map | PROSPECT-SCOPE radiative-transfer inversion (look-up table or machine-learning emulator) applied to atmospherically corrected EnMAP or PRISMA L2 reflectance | 30 m GeoTIFF of retrieved leaf nitrogen (g N g⁻¹ dry mass) with per-pixel posterior uncertainty layer |
| Canopy nitrogen content per unit ground area | Leaf nitrogen concentration scaled by LAI retrieved from the same hyperspectral scene or fused Sentinel-2 multispectral data | 30 m GeoTIFF of canopy nitrogen content (g N m⁻² ground) with propagated LAI uncertainty included |
| Vcmax proxy map for carbon-cycle model input | Empirical Vcmax-nitrogen relationship (published Evans/Walker parameterisations) applied to retrieved canopy nitrogen | Gridded Vcmax estimate (µmol CO₂ m⁻² s⁻¹) delivered as GeoTIFF or netCDF, formatted for input to JULES, CLM or similar land-surface models |
| Seasonal nitrogen change detection | Multi-date PROSPECT-SCOPE inversion across available cloud-free acquisitions; change vector analysis on retrieved parameters | Time-series chart per stand polygon showing nitrogen trajectory across the growing season, flagged where change exceeds retrieval uncertainty |
| Forest health nitrogen anomaly report | Z-score normalisation of retrieved canopy nitrogen against a multi-year baseline for the same stand and phenological stage | PDF stand-level report with anomaly maps, uncertainty summary and recommended follow-up acquisition dates |
| Mixed-species uncertainty assessment | Monte Carlo propagation of LMA prior uncertainty through PROSPECT inversion, conditioned on available species composition data or, where absent, on published LMA distributions for the regional flora | Per-pixel uncertainty classification layer (low / medium / high confidence) accompanying all nitrogen retrieval outputs |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.