Abstract Carbon Capture and Storage (CCS) provides a mechanism by which CO2 can be removed from the atmosphere and stored in reservoirs. Regulations and stakeholder assurance require monitoring to show storage is robust. The marine environment is heterogeneous and dynamic, and baselines are extremely variable. Hence, distinguishing anomalous CO2 from natural variability is challenging. Monitoring schemes must be designed to identify releases early and with certainty, whilst being cost effective. A key question is how to deploy the smallest number of sensors to ensure effective monitoring? We approached this problem through a 3D hydrodynamic model (FVCOM) coupled to a carbonate system. The unstructured grid resolution ranges from 0.5 km to 3 m and simulates seabed release scenarios ranging from 3 t d−1 to 300 t d−1 using the Goldeneye complex as an exemplar test bed. This configuration allows us to characterise and analyse the fate of CO2 in the water column, with the spatial and temporal CO2 patterns shown to be affected by both tides and seasonal mixing/stratification. A weighted greedy set algorithm is used to identify the positions within the model domain which yield the greatest combined coverage for the smallest number of sampling stations, further limited by selecting only a feasible number of sample sites. The weighted greedy set algorithm incorporates the effect of the variable grid spacing as well as the proximity of the sample locations to the Goldeneye complex. The weighted greedy set can identify releases sooner, with a stronger signal than a regular sampling approach.