Recent work has vastly expanded the known viral genomic sequence space, but the seasonal dynamics of viral populations at the genome level remain unexplored. Here we followed the viral community in a freshwater lake for 1 year using genome-resolved viral metagenomics, combined with detailed analyses of the viral community structure, associated bacterial populations and environmental variables. We reconstructed 8950 complete and partial viral genomes, the majority of which were not persistent in the lake throughout the year, but instead continuously succeeded each other. Temporal analysis of 732 viral genus-level clusters demonstrated that one-fifth were undetectable at specific periods of the year. Based on host predictions for a subset of reconstructed viral genomes, we for the first time reveal three distinct patterns of host–pathogen dynamics, where the viruses may peak before, during or after the peak in their host’s abundance, providing new possibilities for modelling of their interactions. Time series metagenomics opens up a new dimension in viral profiling, which is essential to understand the full scale of viral diversity and evolution, and the ecological roles of these important factors in the global ecosystem.