Figure 3. RMSE vs. True Inter-hole Spacing. The analysis identifies a clear "Value Window" at inter-hole spacings between 30m and 60m. In this range, the hybrid model outperforms Kriging by approximately 30%. This corresponds to the Infill Drilling stage (converting Indicated to Measured resources). The data suggests that using GeaAI could allow operators to expand drill spacing (e.g., from 30m to 40m) while maintaining the same estimation quality, resulting in significant CAPEX savings. 4.3 Robustness and Outlier Management A sensitivity analysis comparing GeaAI (trained on raw data) versus Kriging (trained on capped data) revealed that top-capping actually degraded the Kriging performance (RMSE increased from 33.9 to 39.9). This indicates that high-grade outliers in this deposit carry structural information. GeaAI, with its tree-based architecture, successfully utilized these raw high-grade samples to define ore shoots without smearing them, eliminating the need for arbitrary top-capping.
RkJQdWJsaXNoZXIy MTM0Mzk2