Real-time strategy (RTS) games represent a mainstream genre of video games. They are also practical test-beds for intelligent agents, which have received considerable interest from Artificial Intelligence (AI) researchers, in particular game AI researchers. Terrain knowledge understanding is a fundamental issue for RTS agents and map decomposition methods can help AI agents in representing terrain knowledge. These contributions support AI agents’ path finding and combat strategy. In some RTS games, such as StarCraft, all terrain information is provided to AI agents at the beginning of the game. This presents an unfair advantage, as human players do not have access to this information. We propose a terrain analysis framework, in which AI agents gather terrain knowledge by managing scouts to explore game maps. This framework is part of my Ph.D. study that is investigating scouting strategies for RTS games. We developed an extension to the StarCraft system, called terrain engine that releases terrain information in small chunks rather than providing the full map, to investigate human-like techniques for scouting. Within the terrain analysis framework, we present a reconnaissance (recon) algorithm to guide individual scout units in recon tasks. Then, we identify the factors for terrain exploration planning model, which will be implemented as part of our future work.