The increasing availability of high-resolution satellite stereo imagery has enabled large-scale 3D reconstruction for urban morphology analysis and monitoring change. However, conventional semi-global block matching methods are highly sensitive to the minimum disparity parameter, particularly in complex urban environments characterized by dense buildings, severe occlusions, and abrupt elevation variations, often leading to mismatches and extensive holes that degrade the quality of digital surface models and point clouds. To address this limitation, we propose a hole-driven hierarchical framework for adaptive minimum disparity estimation. The method first derives a globally adaptive minimum disparity to generate an initial disparity map, followed by the identification of invalid regions and progressive refinement through a coarse-to-fine strategy guided by local statistical characteristics. This transition from global constraint to local adaptation effectively reduces matching failures and improves disparity completeness while preserving geometric consistency. Experiments on multiple satellite stereo datasets, validated against airborne light detection and ranging point clouds, demonstrate significant improvements in the geometric completeness, structural consistency, and detail preservation of reconstructed urban surfaces. The proposed approach provides a robust and scalable solution for accurate 3D urban morphology reconstruction and supports the reliable analysis of fine-scale urban changes.