<p>Accurate and continuous prediction of ore hardness and its spatial variation is crucial for optimizing downstream processes in open-pit mining. However, due to the scale of mine operations, regularly measuring rock hardness across different sections and calculating its spatial variability is a challenging, laborious, and time-consuming task. This study investigates the application of close-range hyperspectral imaging (HSI) as a non-invasive, rapid, and scalable technique for predicting rock hardness at active mine sites. The paper discusses the adaptation, upscaling, and evaluation of two laboratory-scale predictive models, random forest regressors trained on (I) band ratio spectral features (RFR_BR) and (II) absorption peak spectral features (RFR_AP), for their proper use at actual mine site conditions. This includes addressing challenges associated with spectral distortions, atmospheric effects, and scale differences, as well as retraining the RFR_BR and RFR_AP models. Afterwards, the adapted predictive models were applied to hyperspectral scans of five rock stockpiles. For the sake of validation, forty-six rock samples were sourced from the five stockpiles scanned, and the rock hardness values were measured by performing the Leeb rebound hardness (LRH) test. The results indicate that the RFR_BR model, with an average error of 5%, a mean absolute error (MAE) of 12%, a root mean square error (RMSE) of 101.2, and an average runtime of approximately 130&#xa0;s, outperforms the RFR_AP models, which exhibited an average error of -7%, an MAE of 21%, an RMSE of 149.4, and an average runtime of 1650&#xa0;s. This research represents a novel field implementation of spatial characterization of rock hardness using close-range HSI data, highlighting the potential of the proposed approach for the automated real-time hardness characterization in active mining operations.</p>