Work

KenAI: Geology Informed Uranium Prospectivity and Spatial Simulation

Year: 2026
Type: article
Abstract: Mineral prospectivity mapping requires geologically meaningful predictors, explicit assumptions and validation that reflects spatial dependence. This study formalises K-SUPI, a five-component index for sandstone-hosted uranium systems, and evaluates its computational behaviour against an equal-weight index and logistic regression. The index represents source potential, transport, redox conditions, geometry and independent supporting evidence through a weighted geometric combination. A deterministic simulation generated 3,072 unitless grid cells, 15 noisy proxy features and 384 positive labels from interacting latent process surfaces. Logistic regression was evaluated with five spatial block folds and a two-cell training buffer; fixed indices were scored on the same labels. The K-SUPI ROC-AUC was 0.985, compared with 0.948 for equal weighting and 0.983 for logistic regression. K-SUPI recalled 71.4% of generated positives in the highest-ranked 10% of cells. Sensitivity and group-ablation analyses characterise model dependence. Because the labels are produced from assumed process relations, these statistics establish internal simulation performance only. No Kazakhstan field data, verified deposit labels or discovery claims are used. A separate protocol specifies the evidence required for geological application.
Language: English
Open Access status: green
OpenAlex ID: W7235151165
FWCI:
Cites:
Cited by:
Related to: 10