Optimal Identification of Concealed Copper Mineral Deposit Locations Utilising Isotope Signatures in Groundwater Systems

This study presents a synthetic proof-of-concept surrogate-assisted simulation–optimisation framework for identifying potentially mineralised locations using groundwater copper-isotope observations. The methodology adapts established groundwater contaminant-source characterisation principles, in which unknown source release histories are reconstructed from down gradient concentration measurements, to the investigation of concealed copper mineralisation. An isotope-informed MODFLOW–MT3DMS model was used to generate synthetic groundwater concentrations of CuTot (total copper), 63Cu, and 65Cu for hypothetical mineralisation scenarios. Model responses were evaluated at 20 observation wells over five stress periods. These synthetic data were used to train, validate, and test a Gaussian Process Regression (GPR) surrogate model. The GPR model replicates the numerical simulator responses during testing. The test results show R2 values exceeding 0.9998 and corresponding RMSE and MAE values of 0.03022 and 0.00571, respectively, indicating high predictive accuracy within the simulated parameter space. The trained surrogate was subsequently coupled with an Adaptive Simulated Annealing (ASA) algorithm to solve the inverse source-characterisation problem. The optimisation formulation contained 20 decision variables, representing time-varying source fluxes at four predefined candidate locations over five stress periods. Three candidates represented active synthetic mineral sources, whereas the fourth was intentionally assigned as an inactive dummy location to evaluate whether the framework could distinguish active from non-mineralised candidates. The optimisation converged in approximately 1080 s under the reported computational configuration. The reconstructed flux histories identified the three active candidates while assigning a near-zero flux to the dummy candidate. These results demonstrate that, under controlled synthetic conditions, isotope-informed inverse modelling can discriminate between predefined active and inactive candidate locations and reconstruct their temporal source strengths. However, the results do not constitute validation of a field-ready mineral-exploration method, because both the training and evaluation data were generated using the same underlying numerical framework. Subject to validation with independent field observations and more realistic geological, hydrogeochemical, and uncertainty representations, the proposed methodology could provide a computational screening and hypothesis-testing tool for prioritising prospective mineralised zones for further investigation.