Quantify the subsurfacebefore you commit a drill rig.
Strataprobe fuses Sentinel-2 spectral indices, structural lineament mapping and supervised ML to produce ranked drill-target inventories — delivered as a reproducible, auditable report.
A focused exploration toolkit, not a black box.
Each layer is independently inspectable. Geologists keep the final call; the pipeline removes the manual drudgery of band math, masking and target triage.
Spectral indices
Iron-oxide, gossan, clay, carbonate, mafic and laterite indices computed from atmospherically-corrected Sentinel-2 L2A scenes.
Statistical anomaly detection
PCA / Crosta band ratios with Isolation-Forest residual scoring isolate spectral outliers across the licence.
Structural lineaments
Edge and Hough-transform extraction of brittle structures, intersected with hydrology to flag fluid-pathway candidates.
Supervised target ranking
XGBoost classifier trained on labelled occurrences scores every pixel; non-maximum suppression yields a discrete target list.
Spectral angle mapping
Match candidate pixels against a curated mineral spectral library (heterogenite, malachite, gossan, dolomitic shale, …).
Reproducible reports
Every deliverable ships with the YAML config, model version, scene IDs and seed — re-runnable on demand.
The same band math you’ll see in the report.
Switch between six commonly-used spectral indices. Drag the threshold to isolate the strongest pixels. The tile below is procedurally generated — regenerated client-side, no real imagery is fetched.
Ferric iron staining; classic gossan / hematite indicator.
Pixels above the threshold retain ramp colour; below pixels are desaturated. P5–P95 stretch removes outliers.
Eight stages, one deterministic build.
Hover any node to inspect the inputs, dependencies and tooling. Every run is hash-pinned; re-running the same AOI on the same scene set produces a byte-identical report.
From licence boundary to ranked targets.
A four-stage pipeline. Every intermediate artefact is preserved so an independent geologist can trace any target back to the underlying scene.
Define the licence
You provide the concession boundary (KML / shapefile / WKT) and target commodity. We pull the latest cloud-free Sentinel-2 mosaics and DEM tiles for the AOI.
Spectral & structural processing
Atmospheric correction, cloud/shadow masking, computation of mineral indices, PCA decomposition, edge detection and lineament extraction.
ML target ranking
Isolation Forest flags spectral outliers; XGBoost ranks every pixel by occurrence-class likelihood; SAM matches against a curated mineral library.
Auditable deliverable
A static, mobile-friendly HTML report plus a GeoPackage of ranked targets, the YAML config, scene IDs, model version and random seed.
A report you can hand to a geologist.
Every project ships with a fully reproducible PDF + GeoTIFF + GeoJSON bundle. Toggle layers below to see how the prospectivity raster, mapped structures, drainage and ranked targets compose into one decision-ready view.
Eight minerals. Every absorption labelled.
Tap a card to see synthetic 400–2400 nm reflectance curves with diagnostic features marked. Click a second card to compare via Spectral Angle Mapper.
Synthetic curves built from gaussian absorption features at published band centres. θ = spectral angle in radians (lower = more similar).
Held-out, leave-one-licence-out, blunt about errors.
Every model ships with a confusion matrix and ROC curve from a held-out licence — never the one you hired us for. Numbers below are the rolling mean across the last twelve regional analogues.
False positives are cheaper than false negatives in early-stage targeting — we tune the operating point to favour recall. Numbers shown are the published rolling mean; per-project metrics ship inside the report.
Move the sliders. See what targeting is worth.
Inputs are yours. The model assumes a percentage uplift in your historical hit rate from ranked targeting. Numbers are illustrative — calibrate to your own programme.
Illustrative model. Strataprobe does not guarantee discovery; ranking improves prior probability, not certainty.
A licence boundary (KML, shapefile or WKT), the target commodity / deposit style, and any prior geological data you’re willing to share (occurrences, soil geochem, prior drilling). That’s it.
Ready to focus your drilling budget?
Send us your licence boundary and target commodity. We’ll come back with a fixed-scope proposal and an estimated turnaround within two business days.