Abstract
Short-range planning often fails for a simple reason hidden in plain sight: when blasthole data are incomplete, many operations project available holes vertically into the bench below, even though mineralization rarely follows a vertical path. The result is a short-term block model that appears operationally usable yet systematically misrepresents grade continuity where ore-waste decisions matter most.
This presentation shows how resource geologists, mine planning engineers, and data scientists worked together to replace that assumption with an orebody-informed machine-learning block model at a large open-pit operation producing more than 60,000 tonnes per day.
Rather than treating short-range estimation as only a geostatistical problem, the team treated it as a learning problem. Historical blasthole and resource model data were used to train a Gradient Boosting Decision Tree model to predict grade distribution one or two benches ahead of blasthole or grade control drilling. Geologists defined the mineralizing controls and tested whether outputs respected deposit geometry. Mine planners defined the operational requirements, timing constraints, and decision points the model needed to support. Data scientists built the training, validation, and deployment pipeline so predictions could be delivered fast enough to influence active mine planning.
The resulting block model was generated in minutes and could be evaluated by geologists without specialist geostatistical configuration. At the study operation, the machine-learning model achieved an F1 ore-waste classification score of 0.90, compared with 0.76 for the unadjusted resource model and 0.92 for the final ore control model. Numeric reconciliation between the machine-learning short-term planning model and the final ore control model reached ±2% variance.
The broader implication is that the persistent reconciliation gap between short-range planning and ore control is not inevitable; it is a product of an assumption that can be replaced. This case study demonstrates how geological interpretation, operational planning, and data science, when genuinely integrated, can produce a workflow that is technically credible, operationally usable, and transparent to decision-makers. Under the integrated lens this conference champions, that kind of cross-disciplinary collaboration is not a soft skill. It is the mechanism through which reserve confidence is built and defended.