A thorough characterization of aquifer parameters is crucial for long-term predictions of the ATES system's functioning. Single well tests, also known as push-pull tests, have been widely used to identify effective porosity, flow velocity, decay constants, sorption coefficients, and heat storage capacity of the aquifer. For more than fifty years, multiple analytical and numerical approaches have been developed to validate push-pull test data and to identify model sensitivity. Despite the relatively straightforward approach, the main bottleneck of the push-pull test calibration is the non-uniqueness of the inverse problem solution. Especially in a deep ATES system data scarcity induces the parametric uncertainty and thus calls for the stochastic parameter optimization. To address this issue, a sensitivity-acknowledging surrogate modeling-based optimization technique for stochastic parameter optimization has been developed. Based on the analytical solution for heat and conservative tracer, a surrogate modeling-based optimization approach was developed to identify the heat storage from the push-pull test data. The optimization procedure has been validated against a synthetic dataset with parameter ranges from one of the ATES sites in Berlin. Results confirm that doing a push-pull test with heat and conservative tracer together enables uncertainty reduction. At the same time, sensitivity acknowledging optimization results in a much narrower posterior parameter distribution than the instant fusion of all available data. The modeling procedure highlights that objective function selection, as well as measurement accuracy, define the confidence interval and calibration precision.