Stochastic Exploration Models

The top of the analytics ladder. Every map so far gave you one answer per pixel; a stochastic model gives you a distribution. It runs a Sequential Gaussian Simulation (SGS): your ground samples anchor a trend fitted on the scene’s spectral bands, the spatial structure of the residuals is captured in a variogram, and the platform then simulates the mineral surface N times. Where the simulations agree, you can trust the estimate; where they disagree, the model is telling you it’s guessing — and that distinction is exactly what drill-hole planning needs.

What you need #

• An Area with sample points carrying assay values (properties like Au_PPB, Cu_PPM — see Zones, Geochemical & Geophysical Data).
• A raw reflectance export of that Area in your Cloud — the model conditions on the scene’s bands.
• A reasonably sized Area (the dice icon’s tooltip tells you when it’s too large).

Running a simulation #

In Cloud → Exports, the dice icon (Run Stochastic Exploration) appears on raw reflectance files. The form:

• Target mineral — the assay key from your samples to simulate. Keys ending _PPB/_PPM automatically enable the log₁₀ transform (recommended for grade data — it tames the heavy right tail).
• Realisations — how many independent simulations (default 20; more gives smoother uncertainty estimates at linearly higher cost — 50+ for publication-quality statistics).
• Variogram model — Auto (AIC) fits spherical, exponential and gaussian and picks the best; keep it unless you have a strong prior. An Advanced section accepts manual range/sill/nugget if you do.
• Summary bands only — drop the individual realizations from the output and keep just the summary bands (~10× smaller file for big runs).
• Exceedance cutoff (optional) — adds a per-pixel probability of exceeding your cutoff (e.g. 100 ppb) to the output.
• Random seed — fix it for reproducible runs.

The scene’s spectral bands are used as covariates automatically. Click Run simulation: real runs execute in the background — you’ll see “Stochastic run queued”, and a few minutes later “Stochastic run completed”, when a new file named after your mineral (tagged [stoch]) appears in the same folder.

The dashboard #

Open the result with the chart icon (Open stochastic dashboard). Exploring the dashboard is free — only the simulation itself consumes usage. The map (left) shows one derived layer at a time:

• P50 (median) — the central estimate (default view).
• Mean — the average across realizations.
• P10 / P90 — the pessimistic and optimistic scenarios.
• σ (std) and P90 − P10 — where the model is uncertain.
• P(score > threshold) — drill-ready probability of exceeding the threshold.

Score, uncertainty and probability layers use deliberately different color families, so you always know which kind of map you’re reading — and the Layer interpretation card explains the active layer in one sentence.

Interactions:

• Click any pixel → the Pixel inspector shows all N simulated values as a histogram (narrow = confident, wide or two-humped = uncertain), the P10/P50/P90/σ summary, and the pixel’s percentile within the whole Area — is this spot hot relative to everything else?
• Draw AOI → sketch a polygon (a tenement, a Zone) and the AOI statistics card re-computes inside it: valid area, share of pixels above the threshold, mean uncertainty.
• Top targets → the highest-probability pixels, ranked and spatially separated so you don’t get ten neighbors of the same anomaly. Click a row to inspect it, and triage with the N/R/D pills (new → reviewed → discarded — the marks last for your session). The lat/lon columns go straight to the field plan.

A geologist’s session, end to end #

1. Samples uploaded, raw reflectance exported.
2. Dice icon → pick Au_PPB, keep defaults, cutoff 100 → Run simulation.
3. Coffee. “Stochastic run completed.”
4. Dashboard: read P50 for the story, σ for the caveats, P(>t) for the shortlist.
5. Click through Top targets, vet each in the inspector, mark R or D.
6. Draw AOI over the ground you actually control and check the stats hold.
7. Export the coordinates to the field team — see The Cloud for downloads.