The inversion of hydrogeological parameters is an essential step of groundwater numerical simulation. A simulation-optimization method is an effective method to solve the hydrogeological parameters inversion problem. However, in the process of simulation-optimization, the groundwater numerical simulation model will be utilized thousands of times, which results in a large computational burden. To overcome this disadvantage, an adaptive dynamic surrogate model is proposed and applied to the inversion of hydrogeological parameters in a water source area in China. Firstly, the modified Morris method is introduced to screen out the sensitive parameters of the groundwater simulation model, which can reduce the dimensions of the parameters inversion. Then, an initial static surrogate model and the optimization model are constructed. Finally, the adaptive sampling method is used to iteratively update the surrogate and optimization models until the optimal inversion parameters are obtained. The results show that the adaptive dynamic surrogate model effectively improves the accuracy of the surrogate model with only 11 newly added samples compared with the static surrogate model. Validation results demonstrate that the adaptive surrogate-model-based simulation-optimization framework effectively reduces the error between the simulated and measured hydraulic heads, thus improving the accuracy and feasibility of hydrogeological parameters inversion.