Background:
High-throughput pathogen sequencing can strengthen outbreak surveillance, but collecting and analysing complete genomic datasets is often not feasible during large outbreaks. Empirical evidence is needed to determine how much genomic sampling can be reduced while preserving key signals required for surveillance and evolutionary analysis, such as maintaining sensitivity to detect important new variants.

Methods:
We evaluated a lineage-based proportional subsampling strategy using 38,217 quality-controlled SARS-CoV-2 genomes from Victoria, Australia (2020–2022). Genomes were sampled at 1%, 5%, 10%, 20%, 50% and 100% per lineage, with 100 bootstrap replicates. Subsampling was designed to retain at least one genome per lineage. We assessed Shannon diversity, Pielou’s evenness, Jensen–Shannon divergence, and preservation of temporal signal using Wuhan-rooted root-to-tip regression.

Results:
Diversity and representativeness improved rapidly between 1% and 10% sampling, with diminishing gains beyond 10%. At 10%, Shannon diversity closely approached the full-dataset value, and variation across replicates decreased substantially. Lower sampling fractions resulted in greater instability and distortion of lineage composition. Jensen–Shannon divergence increased sharply below 10%, indicating reduced representativeness. Temporal signal was retained in independent 10% subsamples, with regression slopes closely matching the full dataset, indicating stable evolutionary rate estimates under moderate subsampling.

Conclusion:
Moderate genomic subsampling can preserve lineage diversity, representativeness, and temporal signal while reducing dataset size for downstream analysis. These findings provide practical guidance for designing scalable genomic surveillance systems and support the use of fit-for-purpose sampling strategies when sequencing or analytical capacity is limited.