This study aims to minimize water scarcity through the joint management of surface water (dam reservoir) and groundwater (aquifer) resources within the Astaneh-Koochesfahan aquifer system. First, the groundwater level was simulated over a 20-year period at a monthly time step using aa Least Squares Support Vector Regression (LSSVR) machine learning model. Subsequently, an integrated optimization model for surface and groundwater allocation was formulated and solved by employing a novel metaheuristic technique, referred to as the Reptile Search Algorithm (RSA). In the final stage, an intelligent machine learning model, the Group Method of Data Handling (GMDH), was developed to predict the optimal rate of groundwater extraction by drawing upon the results of the preceding stages and available data, including water demand and surface water availability. These variables were structured into nine distinct input patterns to enhance the model’s predictive performance. The simulation results indicated that the LSSVR model was able to predict groundwater levels with a high level of accuracy, yielding an MAE of 0.045 m and an RMSE of 0.24 m. The results of the integrated optimal management model demonstrated that, even under severe drought conditions representing the worst-case scenario of surface water availability, approximately 60% of the total water demand could be satisfied. Moreover, the model prioritized the allocation of water to the domestic and industrial sectors via the aquifer, while surface water resources were primarily allocated for agricultural purposes. In the third part of the study, the GMDH model demonstrated effectiveness in estimating the optimal rate of groundwater withdrawal. This approach is not constrained by spatiotemporal limitations and can be applied to other regions comprising both surface and groundwater systems.