A large portion of the global population lives in areas of extraordinary geographic size and population density commonly referred to as megacities. Maintaining social stability and ensuring citizen safety in these megacities is a formidable challenge for both civilian and military forces. Disruptive technologies, such as mobile communication devices and social media, can result in near-instant social volatility. Expanded situational awareness, continuous realtime monitoring, and improved reaction and response methods are essential. Several U.S. Department of Defense (DoD) organizations are working to define how military operations will address megacity-related challenges and what is required to train for these operations. It is unreasonable to define all the megacity geospatial requirements upfront for simulation-based training. Megacities include manmade and natural structures (surface, subsurface, and super-surface); utility networks; complex building interiors; patterns-of-life and emergency response data; sophisticated building usage information; and detailed transportation networks. This paper presents an agile terrain development process to support simulation-based training. Agile frameworks support iterative product development. When enhancing a product iteratively, the product team does not start with a full specification of requirements. Instead, development begins by specifying only the well-known requirements and implementing just part of the product, which can then be reviewed to identify further requirements. The team repeats this process, producing a new version of the product that builds upon prior versions. The agile terrain development process assumes there is a terrain data repository that consists of low-fidelity geospatial data for the world and higherfidelity data for some areas. By leveraging this process, features can be iteratively added to the repository to meet training requirements, reduce cost, and reduce the time to deliver terrain databases to the warfighter. The paper also applies the process to two training examples to demonstrate its utility.