Localization of a mobile robot is a very important task for autonomous navigation. However, with only an initially inaccurate map, a mobile robot cannot estimate its pose robustly because of the inconsistency between the real observations from the environment and the predicted observations on the inaccurate map. The main map used for outdoor environment is DSM (Digital Surface Model) which consists of 2-D grids with elevation information on each grid. In this research, the inaccurate DSM is updated using both estimated robot pose and a local elevation map built by laser range data. In order to match the reference DSM with the local elevation map, ICP (Iterative Closest Points)-based scan matching technique with COAG (commonly observed from air and ground) features is used. Also, the robot pose is estimated by MCL (Monte Carlo localization). Experimental results show that the updated DSM yields better performance in localization compared to non-updated DSM. Error analysis of estimated paths from each map is presented with respect to the ground truth.