Sydney Harbour, Australia, provides a model system for investigating the dynamics of microorganisms across multiple physicochemical gradients, and their response to anthropogenic disturbance in a highly dynamic environment. Here we have comprehensively profiled microbial communities from a range of habitats inside the harbour using ribosomal tag sequencing and shotgun metagenomics. We demonstrated strong spatial and temporal shifts in the taxonomic composition and functional gene potential of bacterioplankton communities. Different regions of the harbour, containing variable amounts of anthropogenic input and tidal flushing, contained clearly partitioned microbial communities, with an overall shift from marine Synecococcus and SAR11 clades to a community dominated by Flavobacteriaceae, Rhodobacteriacaea and Microbacteriaceae at the riverine end of the system. Using network analysis, we have identified the environmental drivers of community partitioning (nutrient concentrations and salinity) and determined the key taxa and metabolic pathways that define ecosystem states in the system. Using an in silico genome binning approach, we linked taxonomy to function to define the role of individual organisms within the community. Combined these results lead to a more detailed understanding of the roles and dynamics of bacterioplankton across nutrient, salinity and contamination gradients, and provide insight into the microbial ecology of urbanised marine ecosystems.