To efficiently evacuate an area threatened by volcanic disaster, onsite observation of active volcanoes by remotely controlled mobile robot systems is desired. Issues involved with developing such systems include planning a safe path for the robots. In this research, our objective was to realize a safe path planning method based on a digital elevation map (DEM) of volcanic mountain fields that considers the mobility of a mobile robot. We assumed that the DEM is obtained through airplane laser measurements beforehand. Because the target environment is vast, obtaining a DEM with sufficiently high resolution is difficult. Even if this is possible, path planning based on such a high-resolution DEM in vast environments significantly increases the computational load. Therefore, we propose a path planning method that can be applied to any DEM resolution; path planning is seamlessly performed roughly in a global scale and precisely in local scales. We extended the general DEM into 3-D space by adding an axis to denote the discrete heading direction of a mobile robot, which we call an extended elevation map (EEM). In the 3-D space EEM, the transition-cost from the start position is derived for each voxel by considering the mobility of the mobile robot. The transition-cost is sparsely propagated from the start position, and the sparsely valued field derives a single path with the lowest transition-cost to reach the goal position. The proposed method was implemented, and simulation experiments using DEMs of real volcanoes were performed to confirm its validity.