deposit type. The three case studies—spanning two commodities, three geological environments, and a range of data quality levels—provide empirical support for the claim that the approach is transferable across geological contexts without fundamental methodological modification. Input selection varies between deposits as a function of PSA outputs; the framework itself does not. The methodology is likely to deliver greatest value at sites where three conditions coincide: high within-domain grade variability, structural geological complexity, and substantial volumes of unassayed or under-utilized logging data. Underground IOCG, orogenic gold, and intrusion-related deposits are representative examples. Extension to VMS, porphyry, and sediment-hosted systems is a logical next step for evaluation. 6. CONCLUSIONS This study presents and validates a DL-based resource estimation methodology applied across three geologically distinct deposits, demonstrating consistent improvement over conventional Kriging-based estimation, with the magnitude of improvement scaling with geological logging data quality. At the Chilean IOCG site, the DL model incorporating proxy logging integration (Method 3) reconciled 1.67 times more mineralization than the site Kriging model at an equivalent false positive rate. A targeted 2,200 m underground drill program verified 7.7 kT of in-situ Cu (approximately USD 64M) in volumes previously classified as waste, with all 11 drill holes intersecting high-grade intercepts and drilling costs 25% lower per kiloton verified than the site baseline. Block-level confidence scores correctly distinguished reliable predictions from uncertain ones; high-confidence target zones achieved a 100% drill intercept rate while low-confidence zones were withheld from the drill program and directed to supplementary data acquisition. At the Australian and Kazakh orogenic gold sites, the approach confirmed economic mineralization in volumes undetected by conventional estimation, with the PSA adapting to local data quality conditions. The principal operational implication is that mines are routinely collecting geological logging and core observation data that contain genuine predictive value for resource estimation and that this value is currently not being captured by conventional workflows. The technical tools required to extract this value have been developed and validated across multiple deposit types. No new data collection programs are required for implementation; the marginal cost of application is therefore substantially lower than the marginal cost of conventional grade control drilling. Identified priorities for future research include: expansion of prospective validation to additional deposit types, specifically VMS, porphyry, and sediment-hosted systems; incorporation of geochemical and geotechnical data streams beyond lithological logging; and the development of standardized geological logging quality assurance protocols designed to support ML model integration, recognizing that model performance is bounded above by the information content and consistency of the logging data on which it trains. Continued accumulation of operational track record also remains essential—the methodology demonstrates technical transferability across geological contexts, but institutional confidence in model outputs is built site by site through demonstrated field performance.
RkJQdWJsaXNoZXIy MTM0Mzk2