Prospectivity
cu-v3.1
Sediment-hosted copper, Kalahari Copperbelt
All runs Ranks ground
Mineral system
Cu Copper Definition 2.0
Sediment-hosted copper, Kalahari Copperbelt
Signed by T. Kgosi, Pr.Sci.Nat on 02 Jul 2026
Gradient-boosted trees, redox-constrained · 250 m
Ran 19 Aug 2026, 172 min
Run by: Amini modelling team · Accepted by: T. Kgosi, Pr.Sci.Nat
Supersedes cu-v3.0 Superseded by cu-v3.2Capture efficiency
The top 10% of ranked ground holds 68% of the known deposits.
Search-space reduction 94.6%
The share of the modelled area a competent person can set aside for this commodity.
Validation
Blind-site hit rate 83%
Threshold agreed at MS1 70%
Block cross-validation 6 folds, 40 km blocks
AUC 0.83 ± 0.06
Physical constraints
- Prediction is confined to the D'Kar / Ngwako Pan redox contact and its permissible stratigraphy.
- Cover thicker than 120 m caps the class at medium.
Features and importance
| Feature | Critical process | System role | Importance |
|---|---|---|---|
| Distance to interpreted redox contact | Trap: reduced host against oxidised red-bed source rock | Trap | |
| Magnetic low, graben margin | Transport: basin-margin faults route brine to the contact | Transport | |
| Cu-Ag-Co pathfinder residual | Source and transport: leakage of metal-bearing brine to surface | Transport | |
| Red-bed thickness, modelled | Source: oxidised sediment supplying copper to solution | Source | |
| Kalahari cover thickness | Preservation and detectability: cover attenuates every surface proxy | Preservation |
Inputs
| Dataset | Version | Confidence filter | Coverage |
|---|---|---|---|
| Ghanzi aeromagnetics | 2004 levelled | A and B only | 94% |
| National gravity | 2021 | A and B only | 71% |
| KCB regional soil geochemistry | 2019 | B or better, levelled | 38% |
| Geology 1:250k with Pre-Kalahari interpretation | 2023 | all, lineage recorded | 100% |
KCB regional soil geochemistry covers 38% of the modelled area and is the layer that governs the bounds.
Validation
| Withheld site | Withheld by | Class | In the high class |
|---|---|---|---|
| BS-CU-01 Zone 5 extension | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-CU-02 Boseto west | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-CU-03 Ghanzi ridge south | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-CU-04 Mahumo | T. Kgosi, Pr.Sci.Nat | Medium | Missed |
| BS-CU-05 Plutus | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-CU-06 Banana Zone north | T. Kgosi, Pr.Sci.Nat | High | In the high class |
Outputs
- Prospectivity surface 3.1 · COG, 250 m
- Confidence surface 3.1 · COG, 250 m
- Feature-importance maps 3.1 · COG per feature
Parameters
- learningRate
- 0.05
- maxDepth
- 5
- trees
- 800
- classCut
- high ≥ 0.80, medium ≥ 0.55
- negatives
- 3:1, drawn outside permissible stratigraphy
Limitations and completeness
- No petrophysical calibration: the 2024 Boseto batch had not been loaded.
- Gravity coverage is 71% of the modelled area.
Against the previous edition
Trust changed between these editions: Does not rank to Ranks ground.
| cu-v3.0 | cu-v3.1 | Change | ||
|---|---|---|---|---|
| Blind-site hit rate | 50% | 83.3% | +33.3% | Improved |
| Capture at 10% of area | 55% | 68% | +13% | Improved |
| Block-CV AUC | 0.79 | 0.83 | +0.04 | Improved |
| AUC spread across folds | 0.09 | 0.06 | -0.03 | Improved |
| High-class ground | 11,400 km² | 7,980 km² | -3,420 km² | Improved |
| Search-space reduction | 92.3% | 94.6% | +2.3% | Improved |
| Grid resolution | 500 | 250 | -250 | Improved |
Inputs that arrived
- National gravity · 2021
Inputs revised
- KCB regional soil geochemistry: 2019 → 2019
What moved the model
- Red-bed thickness, modelled 0% → 15%
- Cu-Ag-Co pathfinder residual 27% → 20%
- Distance to interpreted redox contact 34% → 29%
- Kalahari cover thickness 16% → 12%
- Magnetic low, graben margin 23% → 24%
Targets from this run
0 targets are ranked from this run.