Stage 03 · Variography → Kriging → JORC/KCMI Classification
Resource estimation · 3D block model with JORC/KCMI classification
Goal: Estimate 3D resources from Stage 02 composites with JORC/KCMI classification (Measured / Indicated / Inferred).
Flow: Domain-tagged composites → Variography → Block model → IDW/OK/NN → Cross-validation → Classification → KCMI report.
Disclaimer: The kriging engine is suited for demo & teaching scale (n≤50 neighbors). For production use (commercial production software replacement), cross-validate with industry software.
Pipeline overview, completeness checklist + workflow guide.
Domain-tagged composite from Stage 02 or custom CSV.
Select element, domain, coordinate system, declustering.
Experimental + model fit (Spherical/Exponential/Gaussian) per direction.
Extent + block size XYZ + grid generator.
IDW + Ordinary Kriging + Nearest Neighbor in parallel.
Leave-one-out — MAE / RMSE / R² scatter.
Mean grade per X/Y/Z slice for bias check.
Measured/Indicated/Inferred via kriging variance + slope.
GT curve, cutoff sensitivity per domain.
Plotly volume rendering, slice planes, drill traces.
Resource statement + variogram summary + narrative.
Tip: Each tab has an "Auto-pilot" button that fills in conservative parameters based on the data. Good for first-pass; manual tuning for final estimate.
Stage 03 — Orebit Resource
Estimate 3D block-model resources with variogram fitting, kriging, JORC/KCMI classification.
Output: classified block model CSV + KCMI 2017 Table 1 §3 report
Tip: Start with the Tour button in the top-right header for a guided walkthrough.
Format: CSV with header row. Columns are auto-detected (case-insensitive). Requires hole_id, from_m, to_m, midx, midy, midz, and domain for estimation.
Source: Export Master CSV from Stage 02 (Assay → Domain tab) after domain-tagging your composite intervals.
Privacy: All processing runs client-side. Data is never sent to a server.
Click or drag & drop a CSV file here
Accepts composite CSV from Stage 02 (Orebit Assay)
| Field | Aliases (auto-detected) | Required? |
|---|---|---|
| hole_id | HoleID, BHID, DHID, HOLE, hole | ✓ |
| from_m | FROM, from, From, depth_from, comp_from | ✓ |
| to_m | TO, to, To, depth_to, comp_to | ✓ |
| midx, midy, midz | x, y, z, easting, northing, rl, elevation | ✓ (for 3D est.) |
| domain | DOMAIN, zone, rock_type_code | ✓ (for variography) |
| au_gpt | AU, Au, au_ppm, gold | optional |
| cu_pct, pb_pct, zn_pct, fe_pct | Cu, Pb, Zn, Fe, copper, lead, zinc, iron | optional |
| density | SG, density_t/m3 | optional (QA/QC) |
Restore the embedded sample data (1,241 composites · Thalanga RGMWP with 3 K-Means domains).
Select target element, domains to estimate, and declustering parameters. Default = auto-pilot (conservative parameters).
Experimental variogram + model fit per direction. Nugget ratio < 80% is considered healthy (Sinclair & Blackwell 2002).
Extent from data + block size XYZ. Rule of thumb: block size ≈ ¼–½ × nearest data spacing.
Three parallel methods: IDW (inverse distance weighted, p=2), Ordinary Kriging (variogram-based BLUE), NN (nearest neighbor for validation). Search ellipsoid = 1.5× variogram range.
Leave-one-out validation. Target: R² > 0.4 for OK, slope ≈ 1.0, mean residual ≈ 0 (no global bias).
Spatial bias validation. Drillhole grade vs estimate per X/Y/Z slice. Similar trends = unbiased; divergent = re-tune variogram/search.
Automatic classification based on kriging variance, neighbor count, and distance to nearest composite. Logic: Sinclair & Blackwell 2002 + Rossi & Deutsch 2014.
Tonnage & mean grade sensitivity vs cutoff. Primary output for economic study (Stage 04 Mining Studies).
Visualize blocks with color-coded grade + composite drillholes as lines. Filter by classification or cutoff.
Resource statement + variogram summary + classification rationale. Format: KCMI 2017 Table 1 §3.
8-page PDF: cover, variogram, block model, grade estimation, swath validation, classification, cut-off sensitivity, and KCMI commentary.
Every estimation step applied this session (variography, block model, kriging, classification). Embedded in the PDF report so the reported numbers are always traceable.
First 10 rows of the block model / composite data that will be exported.
Audit trail of every download this session (max 20). Persists in browser localStorage.