Numerical flood modelling is an essential tool for predicting coastal flood hazards such as from storm surges and the outputs of these models are heavily influenced by the bathymetry data that goes into them. However, in shallow regions near the coastline such as coastal lagoons where good bathymetry is critical for reliable model outputs, the influence of uncertainty in bathymetric data on flood model predictions is poorly understood. In this study we use a 2-D hydrodynamic storm surge model to quantify variations in predicted inland flood heights across four bathymetry datasets of varying resolutions nearshore, including a newly developed satellite-based dataset, across the coastal lagoon system of Belize. We find that greater lagoon depths result in lower inland flooding in our model for all lagoon bathymetries except the shallowest, which shows the opposite. We also compare model outputs across three population datasets of varying resolution and show that the higher resolution bathymetries predict more flooded populations regardless of the resolution of the population datasets. Our results show the importance of variations in shallow water bathymetry on flood predictions and suggest that very shallow lagoons may behave differently with regard to storm surge propagation compared to deeper lagoons.