This study presents a hybrid computational framework for 3D modeling of subsurface fluid filtration in oil and gas reservoirs using satellite remote sensing data and machine learning techniques. The proposed approach integrates multispectral imagery from Sentinel-2 and Landsat-8 with a convolutional neural network (CNN) to characterize subsurface heterogeneity and permeability distributions. The CNN model was trained on 1200 labeled satellite image patches linked with borehole and seismic data and achieved a validation accuracy of 91 percent. The predicted permeability fields were coupled with a Darcy-based numerical flow model to simulate fluid transport in porous media. Quantitative evaluation demonstrates that the proposed method improves pressure prediction accuracy by 18 percent and reduces the number of required exploratory wells by 20 percent compared to conventional geological modeling approaches. The results confirm that the integration of remote sensing, deep learning, and physics-based simulation provides a robust, non-invasive, and cost-effective framework for reservoir characterization and exploration planning.