Abstract
Five environmental block models were developed for the proposed KSM project in the Golden Triangle district of northwestern British Columbia, Canada. These block models incorporated two key conceptual principles that reflect predicted long-term waste material environmental behaviour, based on a multi-year geochemical characterization program: 1) waste material ARD potential and 2) lag time to acidity, both of which can be adequately predicted by NPR value as supported by humidity cell observations. The innovative approach involved a Python data cleaning pipeline, including rule-based data validation, imputation, unit standardization, and addressing geochemical closure. Subsequently, machine learning-based methods projected environmental behavior from the environmental database onto the exploration database, providing sufficient resolution for block model development. Specifically, unsupervised multivariate exploratory data analysis was completed to identify environmental domains representing potentially segregable and manageable waste rock volumes with known environmental characteristics. These domains can then be predicted for each interval in the exploration assay database using supervised multivariate proxy analysis. Operators can leverage this flexible approach during mine planning and development, aiming to reduce misclassification and material dilution during operations and achieve significant long-term reductions in environmental liabilities.