This paper proposes a method using satellite stereopairs to reconstruct the bathymetry. The proposed method comprises three major steps: (1) image matching, (2) elevation calculation, and (3) refraction correction. Semi-global matching (SGM) is employed in the first step. Two matching cost assessments, namely, Normalized Cross Correlation (NCC) and Mutual Information (MI) are compared in this study. The matching starts from an initial bathymetry model. We calculate the disparity by SGM for each pixel pair. Then the elevation calculation is employed for 3D positioning via the disparity. The correction for refraction at the air/water interface is needed. Thus, we revise the bathymetry derived from elevation calculation through the incidence angles of satellite images at the final step. Experimental results show that the proposed method could yield the accuracy around 0.7 m for the reconstruction of shallow water bathymetry when WorldView-2 satellite image pairs were tested. The results also indicate that the MI performs better than NCC.