Lichen and cryptogamic crust cover mapping in Arctic and sub-Arctic tundra

Lichens and biological soil crusts underpin Arctic food webs and carbon cycling, yet their spectral signature is narrow and easily swamped by broadband sensors. Hyperspectral data from PRISMA and DESIS can isolate lichen-specific absorption near 1020 nm; Sentinel-2 composites then scale the result across landscapes.

Sensors

PRISMA (ASI): Italian Space Agency hyperspectral imager; 30 m spatial resolution, 239 spectral bands from 400 to 2500 nm at approximately 10 nm sampling. Resolves the lichen absorption feature near 1020 nm and the secondary feature around 1270 nm. Revisit is roughly 29 days at nadir, improving at high latitudes due to orbital geometry, but cloud and polar darkness constrain usable Arctic acquisitions to a narrow July–August window.
DESIS (DLR/Teledyne on ISS): VNIR-only hyperspectral sensor covering 400–1000 nm at 2.55 m spatial resolution and approximately 3.5 nm spectral sampling. The ISS orbit (51.6° inclination) limits coverage to roughly 52° N, which excludes the high Arctic but covers sub-Arctic Scandinavia, Alaska, and the Tibetan Plateau. Revisit is irregular and dependent on ISS attitude scheduling, typically 4–10 days at mid-latitudes.
Sentinel-2 MSI (ESA): 13-band multispectral imager at 10–20 m resolution; 5-day revisit with twin satellites. Lacks the spectral resolution to isolate the 1020 nm lichen feature directly, but bands B11 (1610 nm) and B12 (2190 nm) combined with red-edge bands B5–B7 support partial discrimination and provide the spatial and temporal density needed to scale hyperspectral training data across landscape extents. Summer composites are the backbone of cover-fraction mapping.
Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). Broadband SWIR bands at 1565 nm and 2200 nm provide a coarser but deep-archive (1982 for TM, 2013 for OLI) baseline for change detection. Useful for multi-decadal trend analysis where hyperspectral data do not exist.

Why broadband sensors miss the point

Reindeer lichen (Cladonia rangiferina and relatives) and biological soil crusts dominated by cyanobacteria and mosses occupy a spectral niche that broadband sensors blur into noise. Their reflectance in the visible is low and grey-green, easily confused with dry grass or frost-bleached sedge. The diagnostic feature is a subtle absorption trough near 1020 nm, attributed to secondary metabolites including usnic acid, and a secondary feature around 1270 nm. At Sentinel-2's 20 nm effective bandwidth in the SWIR, those features are invisible. PRISMA's 10 nm sampling resolves them clearly.

Biological soil crusts add further complexity. Wet cyanobacterial crusts darken significantly, shifting their reflectance toward bare soil. Dry crusts can resemble sand or gravel. Neither state has a clean multispectral index. This is not a failure of algorithm design; it is a physics constraint. Hyperspectral data do not make the problem easy, but they make it tractable.

The 1020 nm feature and how to use it

PRISMA acquisitions over tundra allow construction of a lichen absorption index centred on the 1020 nm trough, normalised against shoulder bands at roughly 970 nm and 1070 nm. Published work using airborne hyperspectral data (HyMap, AVIRIS) established this approach before spaceborne hyperspectral became available; PRISMA now provides a repeatable spaceborne equivalent at 30 m. The index separates lichen-dominated surfaces from vascular vegetation and bare ground with considerably less confusion than any broadband proxy.

DESIS covers only to 1000 nm, so it cannot directly observe the 1020 nm feature. It compensates with finer spatial resolution (2.55 m), which is useful for characterising small lichen patches and for generating high-resolution training data to sharpen PRISMA classifications. The two sensors are complementary rather than redundant. A practical workflow trains a spectral unmixing model on PRISMA, validates patch boundaries with DESIS where coverage overlaps, then applies the model to Sentinel-2 red-edge and SWIR bands across the full landscape.

Compositing strategy in a short, cloudy season

Usable acquisition windows in the Arctic are short. At 70° N, solar elevation stays above 20° for roughly ten weeks, and cloud cover over tundra in summer averages 60–80% on any given day. Wet snow persists into June at many sites and contaminates early-season composites with a reflectance signal that resembles dry lichen in some bands. The practical acquisition window for clean hyperspectral data is often late July to mid-August, a period of four to six weeks.

Sentinel-2's 5-day revisit makes it possible to build cloud-free composites within that window by selecting the clearest pixel per location across multiple passes. The standard approach uses a scene classification layer (SCL) to mask cloud, cloud shadow, and snow before compositing. For PRISMA, which revisits a given Arctic site roughly once per month, a single usable scene per season may be the realistic expectation. Multi-year stacking improves coverage but introduces phenological variability as a source of error.

Scaling from hyperspectral pixels to landscape maps

A lichen cover fraction map at landscape scale requires bridging the gap between PRISMA's 30 m hyperspectral pixels and the spatial heterogeneity of tundra, where lichen mats, sedge tussocks, and wet hollows can alternate at 1–5 m intervals. Spectral unmixing, rather than hard classification, is the appropriate model class. Each PRISMA pixel is decomposed into fractional endmember contributions using a linear mixture model, with endmembers defined from field spectra or from spectrally pure DESIS pixels.

The resulting per-pixel lichen fraction at 30 m is then used as a response variable in a regression against Sentinel-2 band composites, red-edge indices, and terrain derivatives. The regression model is applied across the full Sentinel-2 mosaic, producing a continuous cover-fraction surface at 10–20 m. Honest uncertainty quantification matters here: the model degrades at sites where the training hyperspectral data do not represent local lichen species assemblages, and accuracy in wet or shaded terrain is lower than in well-lit, dry-season acquisitions.

Landsat's 40-year archive allows the current cover map to be placed in a long-term context. Shrubification driven by warming has been documented across Siberia and Alaska using Landsat NDVI trends; lichen cover tends to decline where tall shrubs advance, because shading suppresses lichen growth. Combining the hyperspectral-derived cover estimate with a Landsat trend surface gives a more complete picture of where lichen communities are contracting.

What the maps cannot tell you

Cover fraction is not biomass. Lichen mats vary enormously in depth and density, and a pixel showing 70% lichen cover could represent a thin frost-disturbed crust or a decades-old deep mat. Ground-truthing with bulk density measurements remains necessary for any carbon-stock calculation. The satellite data define the spatial pattern; they do not replace field sampling for the vertical dimension.

Species-level discrimination is also beyond current spaceborne hyperspectral capability at tundra scale. PRISMA at 30 m cannot reliably separate Cladonia rangiferina from Cetraria islandica or cyanobacterial crusts from moss-dominated crusts in a mixed pixel. Airborne hyperspectral at 1–2 m resolution can approach genus-level separation in some studies, but that is not yet a scalable operational product from orbit. The maps produced from PRISMA and Sentinel-2 are best described as functional-group cover fractions, not species maps.

Satellize applies this workflow as part of its satellite analytics service, drawing on open Sentinel and Landsat archives and commercial PRISMA tasking where clients hold the appropriate licence. The Tonga crop-estimation programme demonstrated the same core approach of combining hyperspectral endmember training with multispectral scaling in a data-sparse environment.

Typical figures

Hyperspectral spatial resolution	PRISMA: 30 m; DESIS: 2.55 m
Hyperspectral spectral sampling	PRISMA: ~10 nm (400–2500 nm); DESIS: ~3.5 nm (400–1000 nm)
Multispectral spatial resolution	Sentinel-2: 10–20 m; Landsat 8/9: 30 m
Revisit (hyperspectral)	PRISMA: ~29 days nadir; DESIS: irregular, 4–10 days at ISS-accessible latitudes (≤52° N)
Revisit (multispectral)	Sentinel-2: 5 days (two satellites); Landsat 8+9 combined: ~8 days
Usable Arctic acquisition window	Approximately late July to mid-August; cloud and polar darkness constrain this to 4–6 weeks per year at high latitudes
Key diagnostic spectral feature	Lichen absorption trough near 1020 nm; secondary feature ~1270 nm (PRISMA-resolvable)
Minimum detectable lichen cover fraction	Approximately 15–20% sub-pixel fraction in a 30 m PRISMA pixel under clean-sky, dry-surface conditions; lower fractions are within noise
Archive depth	PRISMA: from 2019; Sentinel-2: from 2015; Landsat: from 1982 (TM) / 2013 (OLI)
Deliverable formats	GeoTIFF cover-fraction rasters, cloud-optimised GeoTIFF (COG), polygon shapefiles for classified zones, CSV change statistics

Analytics Satellize can run

Lichen cover fraction map	Linear spectral unmixing on PRISMA hyperspectral data using field-validated endmembers; 1020 nm absorption index as primary discriminator	GeoTIFF raster, 30 m resolution, per-pixel fraction 0–1 with uncertainty band
Landscape-scale cover surface	Regression of PRISMA-derived fractions against Sentinel-2 red-edge and SWIR composites; applied across full Sentinel-2 mosaic	Cloud-optimised GeoTIFF at 10–20 m, clipped to client area of interest
Seasonal cloud-free composite	Sentinel-2 pixel-based compositing using SCL cloud/snow masking over the July–August window; median or maximum-NDVI selection	Multi-band GeoTIFF composite with provenance metadata (date of contributing scene per pixel)
Multi-decadal lichen trend surface	Landsat OLI/TM SWIR and red-edge band trend analysis (Mann-Kendall or linear regression on annual composites) as a proxy for lichen-cover trajectory	Trend-magnitude and significance raster, annual summary CSV
Shrubification pressure layer	NDVI and NDWI trend analysis from Landsat archive to identify advancing tall-shrub zones, cross-referenced with lichen cover map	Classified change polygon layer (stable lichen / shrub advance / degraded crust), GIS shapefile
Acquisition feasibility and scene-quality report	Cloud-cover statistics and solar-elevation modelling for target latitude and season; PRISMA archive query for existing usable scenes	PDF report with recommended tasking windows and historical scene inventory

Sources

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.