AbstractGroundwater contamination source characterization is the prerequisite for the rehabilitation and remediation of contaminated aquifers. This study demonstrates the computational feasibility and application of an efficient linked simulation-optimization based methodology for optimal characterization of multicomponent reactive contamination sources in a complex contaminated aquifer. The computational encumbrance of the complex reactive transport simulation is ameliorated using sufficiently accurate multioutput support vector regression (MSVR)–trained surrogate models as embedded constraints within the simulation-optimization framework. Three different metaheuristic optimization algorithms, i.e., genetic algorithm (GA), particle swarm optimization (PSO) , and generalized simulated annealing (GSA), were used, and their comparative performances were evaluated for a contaminated aquifer resembling a filed-scale contaminated mine site located in Northern Territory, Australia. The performance evaluation result based on an actual contaminated aquifer site along with synthetic concentration measurements indicated that the performance of the GSA-based source identification model is superior compared to PSO and GA. The reliability and robustness of the GSA model were further investigated for erroneous concentration measurement scenarios. The preliminary findings demonstrated the computational efficiency and accuracy of the proposed MSVR surrogate model–assisted optimization for simultaneous identification of multicomponent reactive sources at complex geochemically contaminated field-scale aquifers.