Gold assays often arrive days to weeks after core is logged, while decisions on sampling priority, drilling and short-term planning cannot wait. Secondary information such as multi-element geochemistry, handheld XRF, magnetometer readings, and geological logging is usually available earlier. This study asks whether such proxies can provide quantitative gold guidance before final assays are received. The study used 3,270 drill core samples (from 3,471 in the database) from an orogenic gold deposit. Antimony, arsenic and sulphur were selected as predictors on the basis of their correlation with gold in normal-score space (ρ = 0.55, 0.55 and 0.41). Antimony adds information largely independent of arsenic (ρ = 0.24). Supervised regression models, both linear and polynomial, were trained as the machine-learning core of the workflow. They were evaluated using 10-fold cross-validation and benchmarked against MultiGaussian conditional expectation and co-kriging on the same folds. The multiple linear model explained 49% of the variability of normal-score gold, and polynomial models raised this to 56%. In cross-validation, the best non-linear regression came close to co-kriging, with an RMSE of 7.92 against 7.88 g/t, at a fraction of the modelling effort, as it needs no variogram or co-regionalization model. Back-transformed regression estimates were biased low, which the MultiGaussian conditional mean corrects. For sample prioritization, samples were ranked by predicted gold, using the geochemistry plus logged lithology and weathering, on drill holes held out from the fit. Of the top-ranked tenth, 84% were high grade (≥ 4.2 g/t), against 1% of the bottom tenth. Assaying the first 20% of the ranked list recovered 60% of the high-grade samples, three times a random order. The result was stable across 20 different drill hole splits. Lab assays of As, Sb and S were used as a proof of concept. The intended application is any variable obtainable before gold, including handheld XRF, logging and vein density. While the model cannot replace laboratory assays, it shows that readily available proxies can provide quantitative guidance for faster decisions, sample prioritization and exploration targeting in time-sensitive environments. The workflow is designed to be retrained as each assay batch returns.
Innovation in Resource and Reserve Modelling
Can We Predict Gold Before the Assay Comes Back?
3 views