Legacy lead contamination risk mapping via spectral soil proxies
Hyperspectral sensors identify bare-soil mineral assemblages that co-occur with lead contamination at smelter fallout zones, mine tailings and demolition sites, producing spatial risk priors for childhood blood-lead exposure. Satellite data cannot directly detect lead or see through vegetation.
Sensors
- EnMAP: German hyperspectral satellite with 230 spectral bands across 420–2450 nm at 30 m spatial resolution and roughly 27-day revisit at nadir. Spectral sampling of ~6.5 nm in VNIR and ~10 nm in SWIR allows discrimination of iron-oxide polymorphs (goethite, haematite, jarosite) and carbonate minerals that are diagnostic of acid-mine-drainage and smelter-fallout soils.
- DESIS (ISS): DLR/Teledyne hyperspectral imager on the International Space Station, covering 400–1000 nm at ~2.55 nm spectral sampling and 30 m ground sampling. ISS orbital inclination (51.6°) gives irregular revisit, typically weeks to months for any given site, and SWIR coverage is absent, limiting carbonate mapping but preserving iron-oxide and clay-mineral discrimination in the VNIR.
- Landsat 8/9 OLI: Six reflective bands plus a panchromatic band at 30 m (15 m pan), 16-day revisit per satellite (8-day combined). Multispectral resolution cannot distinguish individual minerals, but band ratios (e.g. OLI bands 4/2 for iron oxide, 6/7 for clay/carbonate) provide a coarser but globally consistent screening layer useful for prioritising hyperspectral tasking.
- AVIRIS-NG (airborne reference): NASA's Airborne Visible/Infrared Imaging Spectrometer Next Generation covers 380–2510 nm at ~5 nm spectral sampling with spatial resolution adjustable from ~0.3 m to ~4 m depending on flight altitude. Used as the ground-truth benchmark for mineral mapping at contaminated sites; its published datasets from Superfund sites and mine tailings provide the spectral libraries against which spaceborne results are validated.
What a smelter leaves in the soil, and why it shows up spectrally
Lead does not have a diagnostic absorption feature in the solar-reflectance spectrum. You cannot point a hyperspectral sensor at contaminated ground and read out a lead concentration. What you can do is read the mineral company it keeps. Smelter fallout and acid-mine-drainage alter soil mineralogy in predictable ways: iron is oxidised to goethite and haematite, sulphides weather to jarosite, and carbonate phases accumulate as neutralisation products. These minerals have strong, well-characterised absorption features in the 400–2500 nm range that hyperspectral sensors resolve clearly.
The practical logic is one of co-occurrence rather than direct detection. Peer-reviewed studies of Superfund-listed smelter sites in the United States and mine tailings in Europe have consistently found that zones with elevated soil lead also show elevated iron-oxide and jarosite spectral signatures, because both result from the same industrial process and weathering chemistry. Satellite spectral mapping therefore produces a spatial prior: areas where the mineralogical fingerprint is present are candidates for ground-truth sampling, not confirmed contamination. The distinction matters enormously for any public-health application.
How spectral mixture analysis turns pixels into risk zones
A single 30 m EnMAP pixel over a legacy industrial site typically contains a mixture of contaminated bare soil, uncontaminated soil, sparse vegetation and possibly gravel or concrete. Spectral mixture analysis (SMA) decomposes each pixel into fractional contributions from a set of pure spectral endmembers. The endmembers are drawn from laboratory or field spectral libraries, ideally validated with AVIRIS-NG data from analogous sites. The output is a set of abundance maps: what fraction of each pixel is goethite, jarosite, carbonate, green vegetation, and so on.
EnMAP's 230-band coverage across VNIR and SWIR is well suited to this. Jarosite has a diagnostic absorption near 900 nm and again near 2270 nm. Goethite absorbs strongly around 900 nm. Carbonate minerals show a doublet near 2300–2350 nm. DESIS contributes the VNIR portion of this picture but misses the SWIR carbonate features entirely, making it a partial tool. Landsat OLI band ratios can flag broad iron-oxide anomalies at continental scale but cannot separate jarosite from goethite or distinguish carbonate from clay, so they are most useful for screening large areas before committing hyperspectral tasking budget.
Constrained linear SMA assumes endmembers sum to one and that abundances are non-negative. Where soils are spectrally complex or endmembers are poorly chosen, mixture residuals are high and results degrade. Multiple-endmember SMA (MESMA) allows the endmember set to vary pixel by pixel, improving accuracy in heterogeneous urban-fringe environments typical of legacy smelter sites.
The vegetation problem is not minor
Grass, weeds and tree canopy suppress the soil signal almost completely. Even 30–40% vegetation cover in a pixel can obscure the iron-oxide signature that would otherwise flag contamination. This is a serious operational limit. Many legacy smelter sites in temperate climates are now partially revegetated, either naturally or through remediation planting. The spectral proxy works best on bare or sparsely vegetated ground: active tailings ponds, recently disturbed demolition sites, arid-climate mine waste.
Seasonal bare-soil windows help. Acquiring imagery in late winter or early spring, before vegetation flush, maximises bare-soil fraction. Multi-date compositing across several years of EnMAP or Landsat imagery can identify pixels that are seasonally bare and spectrally anomalous. But this requires an archive of cloud-free acquisitions, which is not guaranteed for any specific site, and the 27-day EnMAP revisit means cloud probability accumulates between passes. There is no workaround that fully solves the vegetation-occlusion problem from orbit.
Connecting spectral maps to childhood blood-lead risk
The pathway from spectral anomaly to public-health risk estimate runs through several steps, each adding uncertainty. First, the spectral abundance map is converted to a binary or graded soil-risk layer by thresholding iron-oxide and jarosite abundance against values calibrated from sites with known lead concentrations. Second, that risk layer is intersected with residential land use, population density and, where available, census data on children under six. Third, the resulting spatial prior is used to prioritise soil sampling campaigns, which produce the actual concentration measurements needed for regulatory action.
Satellite data cannot replace soil sampling. It can make sampling dramatically more efficient by concentrating field effort on high-probability zones. In large post-industrial regions with hundreds of potential sites, that prioritisation function has real value. The output is a ranked list of candidate sites, not a contamination certificate. Any downstream health communication must be clear about that distinction.
Honest limits of the method
Subsurface lead is invisible to any passive optical sensor. If contamination is buried under clean fill, sealed under tarmac or covered by vegetation, the spectral proxy produces no signal. Urban demolition sites with legacy leaded paint are particularly problematic: the contamination is often mixed into rubble and covered within weeks. The spectral window for detection may be very short.
False positives occur. Natural iron-rich soils, laterites and some anthropogenic fills produce iron-oxide signatures without lead contamination. Validation against known-clean reference sites in the same geological setting is essential to set a sensible detection threshold. Reported detection limits in the peer-reviewed literature for spectral mapping of mine-tailings mineralogy typically require bare-soil fractions above roughly 60–70% in a pixel; below that, results become unreliable.
EnMAP's archive is still young. The satellite launched in April 2022, so multi-year time series are limited. DESIS has operated since 2018 but with irregular tasking. For historical baseline work, AVIRIS-NG archival data from NASA campaigns over US Superfund sites provides the longest validated record, though coverage is geographically patchy and not global.
Satellize runs spectral mixture workflows on EnMAP and Landsat OLI data for government environmental agencies, drawing on the same open-archive methods used in its Tonga crop-estimation programme. The analytic outputs feed into GIS layers that field teams can use directly to plan sampling campaigns.
What a useful deliverable actually looks like
A credible spectral risk map for a legacy smelter catchment should include: a per-pixel iron-oxide and jarosite abundance layer with associated uncertainty; a bare-soil fraction mask that flags where the spectral proxy is and is not reliable; a ranked list of candidate sampling zones with the spectral evidence behind each ranking; and a clear statement of the seasonal acquisition window used and cloud-cover statistics for the imagery.
It should not include lead-concentration estimates, health risk scores expressed as probabilities of exceedance, or any claim that the map identifies contaminated land. Those conclusions require soil chemistry. What the map provides is a spatially explicit argument for where to look first. For a cash-constrained environmental agency managing hundreds of square kilometres of post-industrial land, that argument can be worth a great deal.
Typical figures
| Spatial resolution (hyperspectral) | 30 m (EnMAP, DESIS); ~0.3–4 m (AVIRIS-NG airborne reference) |
| Spatial resolution (multispectral screening) | 30 m (Landsat 8/9 OLI); 15 m panchromatic |
| Spectral range | EnMAP: 420–2450 nm (230 bands); DESIS: 400–1000 nm (~235 bands); Landsat OLI: 6 reflective bands, 443–2200 nm |
| Revisit (EnMAP) | ~27 days at nadir; varies with off-nadir tasking |
| Revisit (Landsat 8 + 9 combined) | ~8 days at equator |
| Minimum detectable bare-soil fraction | ~60–70% bare soil per pixel required for reliable mineral discrimination (published literature range) |
| Key diagnostic spectral features | Jarosite: ~900 nm, ~2270 nm; Goethite: ~900 nm; Carbonate: ~2300–2350 nm doublet |
| Archive depth | EnMAP from April 2022; DESIS from 2018; Landsat from 1972 (OLI from 2013) |
| Cloud limitation | Passive optical only; cloud-free acquisition required; multi-date compositing recommended |
| Deliverable formats | GeoTIFF abundance maps, GeoPackage risk-zone polygons, CSV ranked candidate-site lists |
Analytics Satellize can run
| Iron-oxide and jarosite abundance map | Constrained linear spectral mixture analysis (SMA) or multiple-endmember SMA (MESMA) on EnMAP L2A reflectance data, using published spectral libraries (USGS spectral library) | Per-pixel abundance GeoTIFF with uncertainty layer, clipped to area of interest |
| Bare-soil fraction mask | Normalised Difference Vegetation Index (NDVI) thresholding on Landsat or EnMAP VNIR bands to flag pixels where spectral proxy is unreliable | Binary GeoTIFF mask delivered alongside abundance map |
| Landsat OLI iron-oxide screening layer | Band-ratio compositing (OLI band 4/band 2 for iron, band 6/band 7 for clay/carbonate) across multi-year Landsat archive to identify persistent spectral anomalies | Multi-date composite GeoTIFF for regional screening; used to prioritise hyperspectral tasking |
| Ranked candidate sampling zones | Intersection of spectral anomaly layer with residential land-use data and population density; ranked by anomaly intensity and exposure potential | GeoPackage polygon layer with attribute table ranking and spectral evidence scores; CSV export for field teams |
| Seasonal bare-soil acquisition window analysis | Time-series NDVI analysis on Landsat archive to identify optimal low-vegetation acquisition months per site | Site-specific acquisition calendar report, PDF or JSON |
| Change detection for active demolition or disturbance | Bi-temporal spectral change analysis on Landsat OLI to flag newly exposed bare soil at demolition sites within the study area | Alert GIS layer updated on each cloud-free Landsat pass; delivered as GeoTIFF or web-map tile feed |
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.