Prospectivity
u-v1.0
Sandstone-hosted uranium, Karoo palaeochannels
All runs Ranks ground
Mineral system
U Uranium Definition 1.0
Sandstone-hosted uranium, Karoo palaeochannels
Signed by T. Kgosi, Pr.Sci.Nat on 29 Aug 2026
Gradient-boosted trees, palaeochannel model · 250 m
Ran 12 Sept 2026, 105 min
Run by: Amini modelling team · Accepted by: T. Kgosi, Pr.Sci.Nat
Capture efficiency
The top 10% of ranked ground holds 60% of the known deposits.
Search-space reduction 94.8%
The share of the modelled area a competent person can set aside for this commodity.
Validation
Blind-site hit rate 60%
Threshold agreed at MS1 60%
Block cross-validation 5 folds, 35 km blocks
AUC 0.80 ± 0.08
Physical constraints
- Prediction is confined to Karoo sandstone within a modelled palaeochannel; basement cannot score.
- A redox front is required: oxidised channel sand alone does not score high.
Features and importance
| Feature | Critical process | System role | Importance |
|---|---|---|---|
| Modelled palaeochannel axis | Transport: the channel that carried oxidised uranium-bearing water | Transport | |
| Uranium channel residual | Source: surface expression of the mineralised sequence | Source | |
| Carbonaceous horizon proximity | Trap: organic matter reduces and precipitates uranium | Trap | |
| Cover thickness | Preservation and detectability: cover attenuates the radiometric signal | Preservation |
Inputs
| Dataset | Version | Confidence filter | Coverage |
|---|---|---|---|
| Radiometrics, northern tiles | 1997 | C or better, levelled | 58% |
| Borehole logs, Karoo sequence | 2025 compilation | B or better | 36% |
| Geology 1:250k | 2023 | all, lineage recorded | 100% |
Borehole logs, Karoo sequence covers 36% of the modelled area and is the layer that governs the bounds.
Validation
| Withheld site | Withheld by | Class | In the high class |
|---|---|---|---|
| BS-U-01 Serule central | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-U-02 Gorge | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-U-03 Letlhakane east | T. Kgosi, Pr.Sci.Nat | Medium | Missed |
| BS-U-04 Dukwe margin | T. Kgosi, Pr.Sci.Nat | High | In the high class |
| BS-U-05 Foley | T. Kgosi, Pr.Sci.Nat | Medium | Missed |
Outputs
- Prospectivity surface 1.0 · COG, 250 m
- Confidence surface 1.0 · COG, 250 m
- Feature-importance maps 1.0 · COG per feature
Parameters
- learningRate
- 0.05
- maxDepth
- 4
- trees
- 600
- classCut
- high ≥ 0.75, medium ≥ 0.50
- negatives
- 3:1, basement and non-channel Karoo
Limitations and completeness
- Radiometric coverage of 58% is the binding gap; the northern channels are modelled from boreholes alone.
- Palaeochannel geometry is interpreted at 1:250k and will move with any new drilling.
Against the previous edition
This is the first edition of this model.
Targets from this run
1 targets are ranked from this run.
| Target | Score | Class | Competent person |
|---|---|---|---|
| LET-U-004 Serule Palaeochannel | 0.69 | Medium | M. Mokgatlhe, MAusIMM CP |