Accurate estimation of hydraulic parameters, such as hydraulic conductivity, recharge rates, and layer elevations, is essential for groundwater modeling. Qatar’s karst limestone aquifers present challenges due to their heterogeneity, including cavities, sinkholes, and underground channels. Additionally, distorted hydrology images and the computational intensity of calibration processes hinder accurate simulation, particularly for historical groundwater data from 1958, representing steady-state conditions. This study introduces a two-stage framework integrating machine learning (ML) and zonation-based optimization to improve groundwater modeling. ML techniques, including Random Forest and K-Nearest Neighbor models, preprocess hydraulic data from scanned maps, converting them into structured raster formats. These outputs enhance the interpolation of observed hydraulic heads, supporting the calibration of Qatar’s aquifers. The second stage applies zonation-based calibration using MODFLOW. Instead of computing gradients for all model cells, K-Means clustering partitions the domain into representative zones, optimizing calibration by computing gradients at single points per zone. Iterative parameter updates refine hydraulic conductivity and recharge rates, ensuring convergence with minimal error. This framework enhances computational efficiency and accuracy in groundwater modeling. Results highlight hydraulic conductivity variations in Qatar’s aquifers, offering insights into flow dynamics. Future work will extend the method to transient-state conditions and assess artificial recharge impacts.