Maps with different representations play an essential role in the development of automotive intelligence. In order to enhance the capacity of autonomous trucks in surface mine, an extensible terrain mapping system based on LiDAR is proposed in this paper. Point cloud map, 2.5D grid map, and mesh map are integrated into a unified and extensible map-building framework. In order to adapt to the unique characteristics of surface mine, terrain mapping methods are proposed based on existing approaches. Each map-building method builds a local robot-centered map for time-sensitive tasks. Local maps are fused into global maps in the cloud for non-time-sensitive tasks. The construction method of the point cloud map can avoid the loss of information when updating the map by computing convex hulls. The 2.5D grid map can model the unstructured and rugged terrain of mines. The mesh map is built based on Poisson reconstruction, which is conducive to human-truck interaction. In addition, the map maintenance method in the framework is proposed. Experiments are conducted with datasets collected in real-world scenes.