Phytoremediation progress monitoring via metal-stress spectral indices
Metal-stressed vegetation on mine tailings shows distinct red-edge chlorophyll depression and SWIR reflectance shifts that differ from drought or nutrient stress. Sentinel-2, PRISMA and EMIT can track these signatures over time, providing auditable evidence of phytoremediation progress for regulators.
Sensors
- Sentinel-2 MSI (ESA): 10 m visible and NIR bands, 20 m red-edge bands (B5 704 nm, B6 740 nm, B7 783 nm) and SWIR bands (B11 1610 nm, B12 2190 nm). Five-day revisit at the equator with two satellites. Free archive from 2015. Red-edge bands are the primary tool for detecting chlorophyll depression under metal stress; SWIR bands respond to changes in leaf water content and cell structure.
- ASI PRISMA: Italian Space Agency hyperspectral imager, 30 m spatial resolution, 400–2500 nm range at approximately 10 nm spectral sampling, 30 km swath. Tasked on request; not a routine daily constellation. Allows full spectral curve fitting to separate metal-stress signatures from drought, senescence and nutrient deficiency with far greater confidence than multispectral data alone.
- NASA EMIT (ISS-mounted): Surface Mineral Dust Source Investigation instrument, 60 m spatial resolution, 380–2500 nm at approximately 7.4 nm sampling. Covers mid-latitudes opportunistically as the ISS orbit permits; not a tasked system. Publicly archived via NASA Earthdata. Useful for initial spectral characterisation of tailings mineralogy and establishing plant stress baselines over large waste footprints.
- AVIRIS-NG (NASA airborne): Airborne Visible/Infrared Imaging Spectrometer Next Generation, sub-5 m spatial resolution, 380–2510 nm at approximately 5 nm sampling. Campaign-based, not continuous. Sets the ground truth for spectral unmixing models that are then scaled to spaceborne data. Published studies have used AVIRIS data to map hyperaccumulator species on contaminated soils in California and elsewhere.
Why metal stress leaves a different spectral fingerprint
Heavy metals, particularly cadmium, zinc, lead and arsenic, interfere with chlorophyll synthesis and disrupt the photosynthetic apparatus in ways that drought stress does not fully replicate. The result is a characteristic depression of the red-edge reflectance slope, centred roughly between 700 and 740 nm, that is detectable even when plants appear visually green. Drought stress tends to shift the red-edge position without depressing its amplitude as sharply; nutrient deficiency produces a broader yellowing that affects the visible range more uniformly. This distinction matters enormously for phytoremediation monitoring because a site manager needs to know whether canopy decline is caused by metal uptake, insufficient irrigation or simple establishment failure.
SWIR reflectance, particularly around 1610 nm and 2190 nm, rises when leaf mesophyll structure degrades under metal toxicity, because cell wall integrity and water content both decline. Combining red-edge indices such as the Red-Edge Chlorophyll Index (CIre, using Sentinel-2 B7 and B5) with SWIR-based indices gives a two-dimensional diagnostic space that separates metal stress from competing explanations. Hyperspectral data from PRISMA or EMIT can resolve this further by fitting full spectral curves against published metal-stress libraries.
What Sentinel-2 can and cannot tell you
Sentinel-2's red-edge bands are genuinely useful here, not merely adequate. The CIre index and the MERIS Terrestrial Chlorophyll Index (MTCI), both computable from the 20 m red-edge bands, have been validated against field chlorophyll measurements in contaminated grassland and early-successional tailings communities. Five-day revisit means seasonal trajectories are well sampled, which matters because phytoremediation plots are typically assessed over growing seasons, not single dates.
The honest limits are worth stating plainly. At 20 m resolution, plots smaller than roughly 0.04 ha will be mixed with bare tailings or rock in most pixels, diluting the spectral signal. Sparse canopies below about 20–30 percent cover produce indices dominated by soil and substrate background, not by the plants themselves. Soil-adjusted variants of vegetation indices (SAVI, MSAVI) help, but they do not eliminate the problem. Sentinel-2 cannot distinguish hyperaccumulator species from stress-tolerant non-accumulators; that requires either hyperspectral data or field sampling. Cloud cover over tropical mine sites can reduce usable observations to fewer than ten per year in some regions, which is a real constraint on time-series density.
Where hyperspectral data earns its cost
PRISMA and EMIT data allow spectral unmixing at the sub-pixel level, separating the contributions of bare tailings, stressed vegetation and healthy vegetation within a single 30–60 m pixel. More importantly, the full 400–2500 nm curve allows fitting against laboratory spectra of known hyperaccumulator species and against published metal-stress spectral libraries. Specific absorption features near 2100–2200 nm are associated with clay mineralogy in exposed tailings, and separating these from vegetation SWIR features requires the spectral resolution that multispectral instruments simply cannot provide.
AVIRIS-NG campaigns, where they can be arranged, produce data at sub-5 m resolution that resolves individual plant clusters. This is the appropriate tool for establishing the spectral ground truth that underpins any spaceborne time-series model. The practical workflow is: one or two airborne campaigns at plot establishment and at a defined monitoring milestone, with Sentinel-2 time series filling the gaps between campaigns. EMIT data, freely available via NASA Earthdata, can serve as an intermediate check where PRISMA tasking is not yet scheduled.
Building a time series that regulators will accept
Regulatory acceptance of remote-sensing evidence depends on reproducibility and traceability. The analysis chain needs to be documented: atmospheric correction method (Sen2Cor for Sentinel-2, or a physics-based approach for PRISMA), index formulas with band centre wavelengths stated explicitly, cloud and shadow masking thresholds, and the spatial extent of each monitoring plot defined in a fixed coordinate system. Change between reporting periods should be expressed as a statistically bounded estimate, not a single number, because index variability between cloud-free observations within a single season is not negligible.
Phenological confounding is the most common analytical error in this type of monitoring. A plot that appears spectrally healthier in October than in July may simply be later in its growing season, not genuinely improving. Normalising against a reference phenology, either from an adjacent uncontaminated revegetation plot or from a multi-year Sentinel-2 baseline, is standard practice. Published work on mining rehabilitation in Australia and southern Africa has demonstrated that red-edge trajectory slopes over three to five growing seasons are more informative than single-date index values for distinguishing genuine canopy establishment from seasonal fluctuation.
Honest limits of the whole approach
Spectral indices are proxies. They do not measure metal concentration in plant tissue directly; that requires destructive sampling and laboratory analysis. What remote sensing provides is spatial coverage and temporal frequency that field sampling cannot match, and a defensible basis for prioritising where field samples should be taken. A site with consistently depressed CIre values across a plot over three seasons is a better candidate for targeted soil and tissue sampling than one with stable or improving indices.
Species composition matters and is largely invisible to current spaceborne sensors. A tailings plot colonised by metal-tolerant grasses may show improving spectral health while achieving little actual metal drawdown; a plot with genuine hyperaccumulators such as Noccaea caerulescens may show persistent stress signatures precisely because the plants are actively accumulating metals. Distinguishing these cases from orbit is not currently possible without supporting field data. Satellize's approach, as applied in structured monitoring programmes including the Tonga crop-estimation work, is to be explicit about what the spectral evidence can and cannot conclude, and to design field-sampling protocols that test the spectral hypotheses rather than replace them.
Typical figures
| Spatial resolution (multispectral) | 10 m (Sentinel-2 visible/NIR), 20 m (Sentinel-2 red-edge and SWIR) |
| Spatial resolution (hyperspectral) | 30 m (PRISMA), 60 m (EMIT), sub-5 m (AVIRIS-NG airborne) |
| Revisit frequency | 5 days (Sentinel-2 two-satellite constellation at equator); PRISMA and EMIT are tasked or opportunistic, not routine |
| Spectral range and sampling | Sentinel-2: discrete bands 443–2190 nm; PRISMA: 400–2500 nm at ~10 nm; EMIT: 380–2500 nm at ~7.4 nm; AVIRIS-NG: 380–2510 nm at ~5 nm |
| Key diagnostic bands | Red-edge: 704, 740, 783 nm (Sentinel-2 B5/B6/B7); SWIR: 1610, 2190 nm (B11/B12) |
| Minimum detectable plot size (multispectral) | Approximately 0.04 ha for pure-pixel analysis at 20 m; mixed-pixel methods extend this but with reduced index accuracy |
| Minimum canopy cover for reliable index retrieval | Approximately 20–30 percent fractional cover; below this, soil background dominates |
| Archive depth | Sentinel-2: from 2015 (Sentinel-2A); EMIT: from 2022; PRISMA: from 2019 |
| Latency (Sentinel-2) | Level-2A surface reflectance typically available within 3–5 days of acquisition via Copernicus Data Space |
| Delivery formats | GeoTIFF index rasters, vector plot polygons with per-period statistics, CSV time-series tables, PDF regulatory summary reports |
Analytics Satellize can run
| Red-edge chlorophyll index time series per plot | CIre and MTCI computed from atmospherically corrected Sentinel-2 Level-2A; phenology-normalised against reference baseline | Annual GIS layer set with per-plot index trajectories and season-on-season change statistics, suitable for regulatory submission |
| Metal-stress vs drought-stress classification | Two-dimensional index space (CIre vs SWIR-based index) with decision boundaries derived from published spectral libraries and site-specific field calibration | Per-plot stress-type classification map with confidence bounds, updated each growing season |
| Fractional vegetation cover mapping | Spectral mixture analysis using soil-adjusted vegetation indices (MSAVI2) and linear unmixing against bare tailings and green vegetation endmembers | Fractional cover raster at 20 m, time series CSV showing cover trajectory per monitoring plot |
| Hyperspectral stress curve fitting | Full-spectrum continuum removal and feature matching against USGS or published metal-stress spectral libraries using PRISMA or EMIT data | Spectral similarity score map identifying areas of strongest metal-stress signature, used to target field sampling |
| Multi-year canopy establishment trajectory | Harmonic regression on Sentinel-2 red-edge time series to separate phenological signal from inter-annual trend; Mann-Kendall trend test for monotonic improvement | Per-plot trend report with statistical significance, formatted for inclusion in closure or rehabilitation compliance documentation |
| Field-sampling priority ranking | Spectral anomaly detection identifying plots with persistently depressed or anomalously variable indices relative to site median | Ranked list of monitoring plots with recommended sampling priority and supporting spectral evidence, delivered as PDF and GIS layer |
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.