Computational cost is a critical issue for large-scale water resource optimization problems that often involve time-consuming simulation models. This issue is compounded when optimizing under uncertainty, since Monte Carlo simulations are often required to evaluate objective function values over multiple parameter realizations. In order to improve computational efficiency, we propose a dynamic surrogate modeling approach to approximate and replace the time-consuming numerical models within a noisy genetic algorithm (GA) optimization framework. The surrogates are trained to predict the distribution of the objectives online, using Monte Carlo simulation results created during the GA run. The surrogates are then adaptively updated to improve their prediction performance and correct the GA's convergence as the search progresses. Latin Hypercube sampling method is used to efficiently sample parameters for the Monte Carlo simulation and the sampling results are archived so that the estimation of the objective function distributions is progressively improved. The GA is modified to incorporate hypothesis tests to produce reliable solutions. The method is applied to a hypothetical groundwater remediation design case study, where the primary source of uncertainty stems from hydraulic conductivity values in the aquifers. Our preliminary results show that the technique can lead to reliable and cost-effective solutions with significantly less computational effort.