Information overload is a critical problem in mobile mapping applications where users are impeded by several device limitations. Locating the relevant content and generating a personalised version of the map, tailored towards users' preferences would benefit downloading time as well as readability on a small screen. While existing applications propose solutions that rely on explicit user input, we adopt an implicit profiling approach. In this paper we describe a multimodal mobile GIS that implicitly records all user movements, as well as interactions between users and maps, to dynamically recommend information and to infer persistent spatial preferences. Modeling preferences in this manner allows us to recommend personalised context-aware spatial content to users whenever they request maps. Providing a multimodal interface further improves a user's geospatial experience as each user has the ability to freely switch between different input modalities, including speech and gesture, depending on which mode best suits their current task and environment.