Evaporation pond brine chemistry estimation for lithium operations
Spectral reflectance of lithium evaporation ponds shifts measurably as brine concentrates. Sentinel-2 and Landsat-9 time series let operators track pond-by-pond chemistry progression remotely, though ground-truth calibration remains non-negotiable.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands (B2–B8A), 5-day revisit at the equator with both satellites. The four 10 m bands and four 20 m red-edge bands cover the spectral range most responsive to suspended mineral and brine colour changes. Free and open archive from 2015.
- Landsat-9 OLI-2: 30 m resolution across six reflective bands including coastal aerosol (Band 1, 443 nm) useful for shallow-water and brine optical depth work. 16-day revisit, but combined with Landsat-8 the effective cadence halves to 8 days. Archive continuity to 1972 via earlier missions gives long-baseline change context.
- Planet SuperDove: 3–5 m resolution, daily revisit over most salar latitudes, eight spectral bands including red-edge. Resolves individual small ponds and inter-pond berm detail that Sentinel-2 blurs. Commercially licensed; adds cost but is the right instrument when pond areas are under roughly one hectare.
- DESIS (DLR Earth Sensing Imaging Spectrometer, ISS): Hyperspectral imager covering 400–1000 nm in approximately 235 bands at 30 m resolution. Narrow contiguous bands allow direct spectral unmixing of brine mineral assemblages rather than ratio proxies. Revisit is irregular due to ISS orbit, so it supplements rather than replaces time-series sensors.
What a pond's colour is actually telling you
Lithium brine evaporation operates in stages. Fresh brine pumped from beneath a salar is largely transparent with low total dissolved solids. As water evaporates, halite and then sylvite precipitate, brine density rises, and the surface shifts from blue-green through green-yellow to near-white as salt crusts form. This colour progression is not cosmetic. It reflects real changes in the optical properties of the water column: increased scattering from precipitated minerals, changes in chromophoric dissolved organic matter, and altered absorption coefficients tied to ionic concentration.
Published research on the Atacama Salar and Salar de Uyuni has linked visible-band reflectance ratios, particularly the green-to-blue and NIR-to-red ratios derived from Sentinel-2 and Landsat, to broad brine concentration stages. A 2019 study published in Remote Sensing (MDPI) demonstrated that Sentinel-2 band ratios could distinguish early-stage, intermediate, and late-stage evaporation ponds with reasonable accuracy when calibrated against contemporaneous field samples. The word 'reasonable' matters here: spectral proxies track concentration stages, not precise molality values. They are a monitoring tool, not a laboratory.
The calibration problem nobody should skip
Every spectral-to-chemistry relationship is site-specific. The Atacama brines are chemically distinct from those of the Bolivian Altiplano or the Chinese Qaidam Basin. Magnesium-to-lithium ratios, sulphate content, and co-precipitating minerals all alter the spectral response at a given lithium concentration. A model trained on one salar and applied naively to another will give wrong answers with high confidence, which is worse than no model at all.
Calibration requires contemporaneous field samples taken across the pond network on the same day as a clear-sky satellite overpass. Minimum sample density is a function of pond heterogeneity; a single grab sample per pond is rarely sufficient. Atmospheric correction is equally critical: the thin, high-altitude atmosphere over most salars (the Atacama sits above 2,300 m) reduces aerosol loading, but sun-angle effects and adjacency reflectance from bright salt flats can still distort surface reflectance by several percent, which is large relative to the spectral signal being tracked. Sen2Cor and LaSRC are the standard atmospheric correction chains for Sentinel-2 and Landsat respectively, and both have documented performance over bright targets.
Cloud cover is a genuine constraint in the dry season over the Atacama, but the altiplano wet season (roughly December to March) brings persistent cloud that can interrupt time series for weeks. Operators should plan field campaigns around the dry-season window and treat wet-season imagery as supplementary.
Running a time series: what changes and how fast
At 5-day Sentinel-2 revisit, a full evaporation season of six to eight months yields 30 to 50 usable clear-sky images over most Atacama ponds. That cadence is sufficient to track pond turnover events, detect unexpected dilution from rainfall or berm failure, and map the spatial gradient of concentration across a multi-pond system.
The practical spatial resolution limit matters operationally. At 10 m, Sentinel-2 resolves ponds larger than roughly 0.01 km² cleanly. Smaller ponds, or the narrow brine channels between ponds, are mixed pixels and should be treated with caution. Planet SuperDove at 3–5 m resolves those features but requires a commercial licence and introduces its own radiometric calibration considerations. The choice of sensor is therefore a function of pond geometry, not just budget.
Change detection between time steps is more reliable than absolute concentration retrieval. A pond that shifts two spectral stages in a fortnight is flagging something worth investigating, whether that is faster-than-expected evaporation, a pump failure, or a berm breach letting dilute brine in. Relative change is the operationally useful product; absolute chemistry still needs the laboratory.
Hyperspectral as a periodic diagnostic, not a workhorse
DESIS on the ISS offers something multispectral sensors cannot: continuous spectral coverage across the visible and near-infrared at roughly 2.5 nm spectral resolution. That allows spectral unmixing approaches that can separately quantify the contributions of halite, sylvite, carnallite, and brine solution to a mixed pixel. Published mineral mapping work on evaporite systems has used similar hyperspectral data to map potassium chloride precipitation zones, which are a useful proxy for the lithium concentration stage that follows.
The limitation is revisit. The ISS orbit does not repeat on a fixed schedule over a given ground target, and cloud-free acquisitions over a specific salar may occur only a handful of times per year. DESIS is therefore best used as a periodic calibration anchor for the Sentinel-2 time series rather than as the primary monitoring instrument. Future hyperspectral missions, including ESA's CHIME (planned for the late 2020s), may change that calculus, but they are not yet operational.
From spectral index to operational decision
The practical workflow runs from image acquisition through atmospheric correction, water masking, band-ratio computation, and comparison against a calibrated lookup table derived from field samples. Outputs are typically a per-pond concentration-stage classification (early, intermediate, late, crust) updated on each clear-sky overpass, plus a time-series chart of the dominant spectral index for each pond polygon. That chart is where operational value lives: it shows whether a pond is progressing on schedule, stalling, or regressing.
Satellize runs this workflow on open Sentinel-2 and Landsat-9 archives and can add commercial Planet tasking where pond geometry demands finer resolution. The analytic approach is the same published band-ratio methodology used in peer-reviewed salar remote sensing studies, applied systematically across a client's pond network with client-supplied calibration samples. The Tonga crop-estimation programme established the team's approach to calibrating spectral indices against in-situ ground measurements; the underlying statistical methods transfer directly to brine chemistry staging.
One honest note on what satellite data cannot replace: it cannot tell you lithium concentration in milligrams per litre. It can tell you which ponds are ahead of or behind schedule, which have experienced an anomalous optical event worth investigating, and how this season's progression compares to the previous three years of archive. That is a meaningful operational input. It is not a substitute for the assay laboratory.
Typical figures
| Spatial resolution (primary) | 10 m (Sentinel-2 MSI visible/NIR); 30 m (Landsat-9 OLI-2); 3–5 m (Planet SuperDove, commercial) |
| Revisit cadence | 5 days (Sentinel-2A+B combined); 8 days (Landsat-8+9 combined); daily (Planet SuperDove) |
| Spectral bands used | Blue (443–490 nm), Green (560 nm), Red (665 nm), Red-edge (705–783 nm), NIR (842 nm); DESIS 400–1000 nm continuous at ~2.5 nm resolution |
| Atmospheric correction | Sen2Cor (Sentinel-2), LaSRC (Landsat-9); high-altitude salar sites reduce aerosol error but bright-target adjacency effects require verification |
| Minimum resolvable pond area | ~0.01 km² cleanly resolved at 10 m (Sentinel-2); ~0.001 km² at 3–5 m (Planet) |
| Cloud limitation | Atacama dry season (April–November) typically yields 30–50 clear-sky overpasses per sensor; wet season (December–March) can interrupt time series for weeks |
| Archive depth | Sentinel-2: from 2015; Landsat: from 1972 (30 m multispectral from 1984 with TM) |
| Delivery formats | GeoTIFF reflectance mosaics, per-pond spectral index time series (CSV/JSON), concentration-stage classification polygons (GeoJSON/Shapefile), PDF monitoring reports |
| Ground-truth requirement | Contemporaneous field samples required for calibration; site-specific lookup tables; model not transferable between salars without recalibration |
Analytics Satellize can run
| Per-pond concentration-stage classification | Sentinel-2 band-ratio thresholding (Green/Blue, NIR/Red) calibrated against field samples; published methodology from salar remote sensing literature | GeoJSON polygon layer updated per clear-sky overpass, with stage label (early/intermediate/late/crust) and confidence flag |
| Spectral index time series per pond | Normalised difference water index and custom brine-ratio indices computed from atmospherically corrected surface reflectance | CSV and interactive chart showing index value per pond per date across the full archive and current season |
| Anomaly alert: unexpected optical change | Statistical process control on rolling spectral index; flags ponds deviating more than two standard deviations from seasonal baseline | Email or API alert with pond ID, date, magnitude of deviation, and thumbnail image chip |
| Season-on-season progression comparison | Multi-year Sentinel-2 and Landsat archive analysis; pond-level phenological curves compared across years | Annual PDF report with per-pond progression curves overlaid across up to five seasons |
| Berm and pond boundary change detection | Supervised classification of water, salt crust, and berm pixels; polygon differencing between epochs | GeoJSON diff layer showing pond area gains, losses, or berm breaches with date of first detection |
| DESIS hyperspectral mineral-stage anchor | Spectral unmixing of DESIS imagery to identify halite, sylvite, and carnallite fractions per pixel; used to calibrate multispectral classification | Mineral-fraction raster (GeoTIFF) delivered as a periodic calibration supplement to the Sentinel-2 time series |
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.