Hyperspectral pigment and weathering mapping at open-air rock-art sites
Airborne and spaceborne hyperspectral imagery can distinguish mineral pigments, biological crusts and salt efflorescence on rock-art surfaces without physical contact, turning spectral physics into a condition-monitoring tool for sites too fragile or remote to survey on foot.
Sensors
- ASI PRISMA: Italian Space Agency pushbroom hyperspectral imager on a dedicated satellite. Delivers 239 contiguous bands from 400 to 2500 nm at approximately 30 m ground sampling distance, with a 30 km swath and a revisit of roughly 29 days at the equator (better at higher latitudes). The VNIR and SWIR arms together cover the diagnostic absorption features of iron oxides, sulphates, carbonates and biological pigments. Archive open from 2019.
- DESIS (DLR / Teledyne FLIR on ISS): VNIR-only hyperspectral sensor mounted on the International Space Station. Covers 400 to 1000 nm at 2.55 m ground sampling distance and roughly 30 nm spectral sampling across 235 bands. The fine spatial resolution is useful for isolating individual painted panels at sites with complex topography, though the SWIR range is absent, limiting salt and carbonate discrimination.
- AVIRIS-NG (NASA JPL, airborne): Airborne Visible/Infrared Imaging Spectrometer Next Generation. Covers 380 to 2510 nm in 425 bands at approximately 5 nm spectral sampling; spatial resolution adjustable from under 1 m to several metres depending on flight altitude. The published method standard for rock-art spectral studies. Not a satellite system, but its published spectral libraries and unmixing workflows define the analytical baseline against which spaceborne results are validated.
- EnMAP: German hyperspectral satellite launched April 2022. Covers 420 to 2450 nm across 228 bands at 30 m spatial resolution and a 30 km swath, with a revisit of 27 days globally. Designed for land-surface mineralogy and vegetation biochemistry, making it directly applicable to pigment and weathering mapping. Data available through ESA and DLR portals.
- Sentinel-2 MSI (ESA): Thirteen multispectral bands from 443 to 2190 nm at 10 to 60 m resolution, five-day revisit with two satellites. Not hyperspectral, so it cannot resolve narrow absorption features. Useful as a high-frequency change-detection layer between hyperspectral acquisitions, flagging large-scale surface changes worth investigating spectrally.
What the spectrum actually sees on a painted surface
Rock-art pigments are mostly mineral: ochres and haematite (iron oxides absorbing strongly below 550 nm), white kaolin or gypsum, manganese dioxide for blacks. Each has a characteristic reflectance curve with absorption features at wavelengths specific enough to distinguish one compound from another, provided the spectrometer has narrow enough bands. AVIRIS-NG at 5 nm sampling can separate haematite from goethite; PRISMA at roughly 10 nm sampling can do it reliably for dominant pigments but struggles with thin or mixed layers. DESIS at 2.55 m resolution can isolate individual painted panels, but its VNIR-only range misses the SWIR features that identify sulphates and carbonates.
The complication is that the rock substrate, biological crusts and atmospheric dust all contribute to the measured signal. Spectral unmixing, treating each pixel as a linear or nonlinear mixture of end-member spectra, is the standard method for separating pigment from substrate. End-member libraries derived from AVIRIS-NG field campaigns at sites including those in the American Southwest and southern Africa provide the reference spectra. Spaceborne sensors then apply the same unmixing logic, accepting a coarser spatial and spectral resolution in exchange for repeat coverage.
Salt, crust and biological colonisation: the three accelerants
Weathering at open-air sites is rarely a single process. Salt efflorescence, driven by groundwater capillary rise or aerosol deposition, produces sulphate and halite crusts whose SWIR absorption features near 1750 nm and 2200 nm are clearly detectable in PRISMA and EnMAP imagery. Biological crusts, lichens and cyanobacteria, have chlorophyll absorption near 680 nm and a red-edge inflection between 700 and 740 nm that is distinct from bare mineral surfaces. Mapping these separately matters because they have different causes and different management responses.
Visitor pressure raises local humidity and CO2, accelerating both biological colonisation and carbonate dissolution. Microclimate shifts from nearby vegetation clearance or construction alter moisture cycling. A single hyperspectral acquisition cannot attribute cause, but a time series can rank sites by rate of change. The honest limit is that PRISMA's 29-day revisit and 30 m resolution means that small panels, under roughly 10 by 10 metres, are spectrally mixed with their surroundings. For those, airborne AVIRIS-NG or DESIS tasking is the practical answer.
Change detection without touching the surface
The appeal of spaceborne hyperspectral monitoring is not a single snapshot. It is the ability to compare acquisitions taken months or years apart and quantify shifts in the spectral signature of a specific surface. If the haematite absorption feature at 860 nm weakens between two PRISMA scenes, that is consistent with surface bleaching or crust overgrowth. If the sulphate feature at 2200 nm strengthens, salt accumulation is progressing. These are not diagnoses; they are flags that direct conservators to investigate specific panels.
Atmospheric correction is the largest source of error. Converting raw radiance to surface reflectance requires accurate aerosol and water-vapour retrieval, and small errors propagate into spurious spectral change. The standard approach uses MODIS or Sentinel-5P TROPOMI atmospheric products as ancillary inputs, combined with empirical line correction where field spectrometer readings are available. Sites in arid regions with low aerosol variability, which describes most major rock-art concentrations in the Sahara, Australian interior and American Southwest, are better candidates for reliable multi-temporal analysis than humid tropical sites.
Where the method reaches its limits
Spatial resolution is the binding constraint for most rock-art applications. The painted surface at a typical site occupies tens of square metres, not hectares. At PRISMA's 30 m pixel, a two-metre-wide painted panel contributes perhaps two percent of the pixel's signal. Spectral unmixing can extract it if the pigment is spectrally distinctive and the panel is not shadowed, but the uncertainty is high. DESIS at 2.55 m is more useful for panel-level work, though its VNIR-only range limits the weathering products it can identify.
Cloud cover is less of a problem at arid rock-art sites than elsewhere, but it is not zero. A single cloudy acquisition in a sparse time series can create a gap of several months in the monitoring record. Topographic shadowing is a persistent issue: many painted surfaces face away from the sun for part of the day, and shadowed pixels cannot be corrected to reliable surface reflectance. Scheduling acquisitions for the solar geometry that illuminates the target face is possible with commercial tasking but adds cost and planning lead time.
Building a monitoring programme from available data
A practical workflow starts with the free archive. PRISMA data for most sites with known rock art can be searched through the ASI PRISMA portal; EnMAP data through the DLR EOWEB portal. A baseline spectral map of pigment distribution, biological crust extent and salt efflorescence is generated from the earliest available cloud-free scene. Subsequent acquisitions are co-registered to the baseline and differenced band by band, with change flagged at a threshold calibrated to the atmospheric correction uncertainty.
Where the free archive has insufficient temporal density or the spatial resolution is inadequate for specific panels, commercial airborne tasking fills the gap. The published AVIRIS-NG workflow, documented in peer-reviewed literature and NASA technical reports, provides the analytical template. Satellize can run the spectral unmixing and change-detection pipeline on PRISMA and EnMAP acquisitions, producing georeferenced GIS layers of pigment and weathering-product distribution alongside a condition-change report timed to the site manager's review cycle. The Tonga crop-estimation programme demonstrated that applying established spectral methods to open-constellation data produces operationally useful outputs; the same logic applies here, with different end-members and a different client.
The output is not a substitute for close-range photogrammetry or portable X-ray fluorescence analysis. It is a triage tool: a way to rank panels by urgency and direct limited conservation budgets to the surfaces changing fastest.
Typical figures
| Spatial resolution (spaceborne) | PRISMA and EnMAP: ~30 m GSD. DESIS: 2.55 m GSD. Sentinel-2 (change detection layer): 10–60 m depending on band. |
| Spectral range | PRISMA: 400–2500 nm (VNIR + SWIR). EnMAP: 420–2450 nm. DESIS: 400–1000 nm (VNIR only). AVIRIS-NG airborne: 380–2510 nm. |
| Number of bands | PRISMA: 239 contiguous bands. EnMAP: 228 bands. DESIS: 235 bands. AVIRIS-NG: 425 bands at ~5 nm sampling. |
| Revisit interval | PRISMA: ~29 days at equator. EnMAP: ~27 days. DESIS: irregular (ISS orbital precession), typically 3–5 days at mid-latitudes but not on demand. Sentinel-2: 5 days (two-satellite). |
| Minimum detectable panel size (spaceborne) | Spectrally resolvable by unmixing at ~10×10 m for PRISMA/EnMAP if pigment is spectrally distinct; ~3×3 m for DESIS VNIR. |
| Archive depth | PRISMA: from 2019. EnMAP: from April 2022. DESIS: from 2018. Sentinel-2: from 2015. AVIRIS-NG: campaign-specific, published from ~2013. |
| Atmospheric correction requirement | Full radiative-transfer correction (e.g. ATCOR, 6SV) required before spectral unmixing. Residual error in surface reflectance typically ±0.01–0.02 reflectance units under clear-sky arid conditions. |
| Delivery formats | Georeferenced GeoTIFF reflectance cubes, ENVI .hdr format, GIS-ready mineral/pigment abundance layers (GeoTIFF or shapefile), PDF condition-change report. |
| Data access | PRISMA: free via ASI portal (registration required). EnMAP: free via DLR EOWEB. DESIS: via USGS EarthExplorer. Sentinel-2: Copernicus Data Space. AVIRIS-NG: NASA Earthdata. |
Analytics Satellize can run
| Baseline pigment distribution map | Spectral unmixing (linear or multiple end-member spectral mixture analysis, MESMA) applied to atmospherically corrected PRISMA or EnMAP reflectance cube, using published mineral and pigment spectral libraries | Georeferenced GeoTIFF layer showing fractional abundance of haematite, goethite, kaolin, gypsum and manganese end-members per pixel, with uncertainty estimates |
| Biological crust extent map | Chlorophyll index (ratio of near-infrared to red reflectance) combined with red-edge inflection detection across DESIS or PRISMA VNIR bands | GIS polygon layer of crust extent with area statistics, updated per new acquisition |
| Salt efflorescence distribution map | SWIR absorption feature mapping at 1750 nm and 2200 nm in PRISMA or EnMAP data, thresholded against bare-rock end-member spectra | GeoTIFF layer of sulphate and halite abundance, included in quarterly condition report |
| Multi-temporal weathering-change detection | Band-by-band differencing of co-registered atmospherically corrected scenes, with change significance tested against atmospheric-correction uncertainty floor; flagging of pixels exceeding threshold change in diagnostic absorption features | Change-magnitude GeoTIFF and ranked list of panels showing accelerated spectral change, delivered as PDF site-condition report |
| Acquisition scheduling recommendation | Solar geometry modelling for target cliff aspect and latitude to identify acquisition windows when the painted face is fully illuminated and shadow fraction is minimised | Tasking brief specifying optimal overpass dates and sun-elevation constraints for PRISMA or commercial airborne operator |
| Triage priority ranking | Composite scoring of rate of spectral change, biological crust cover fraction and salt abundance, normalised across panels within a site | Ranked panel inventory table (CSV and PDF) for conservation budget allocation |
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.