This thesis describes how transmission of the 2009 A/H1N1 influenza pandemic in the United States varied geographically, with emphasis on population distribution and age structure. This is made possible by the availability of medical claims records maintained in the private sector that capture the weekly incidence of influenza-like illness in 834 US cities. First, a probabilistic method is developed to infer each city's outbreak onset time. This reveals a clear wave-like pattern of transmission originating in the south-eastern US. Then, a mechanistic mathematical model is constructed to describe the between-city transmission of the epidemic. A model selection procedure reveals that transmission to a city is modulated by its population size, surrounding population density, and possibly by students mixing in schools. Geographic variation in transmissibility is explored further by nesting a latent Gaussian process within the mechanistic transmission model, revealing a possible region of elevated transmissibility in the south-eastern US.

Then, using the mechanistic model and a probabilistic back-tracing procedure, the geographic introduction sites (the transmission hubs') of the outbreak are identified. The transmission hubs of the 2009 pandemic were generally mid-sized cities, contrasting with the conventional perspective that major outbreaks should start in large population centres with high international connectivity. Transmission is traced forward from these hubs to identify basins of infection', or regions where outbreaks can be attributed with high probability to a particular hub.

The city-level influenza data is also separated into 12 age categories. Techniques adapted from signal processing reveal that school-aged children may have been key drivers of the epidemic.

Finally, to provide a point of comparison, the procedures described above are applied to the 2003-04 and 2007-08 seasonal influenza outbreaks. Since the 2007-08 outbreak featured three antigenically distinct strains of influenza, it is possible to identify which antigenic strains may have been responsible for infecting each transmission hub. These strains are identified using a probabilistic model that is joined with the geographic transmission model, providing a link between population dynamics and molecular surveillance.