Acoustic data collected by multibeam echosounders (MBES) are increasingly used for high resolution seabed mapping. The relationships between substrate properties and the acoustic response of the seafloor depends on the acoustic angle of incidence and the operating frequency of the sonar. These dependencies are often treated as confounding factors for seabed mapping, yet they can also provide increased information useful for discriminating benthic substrates or habitats. Using a multi-frequency case study from the Bedford Basin, Nova Scotia, machine learning methods are explored that enable increased utilization of MBES data collected at multiple frequencies over a range of incidence angles for seabed mapping. Multiple imputation is used to augment the angular multi-frequency MBES data, which enables accurately modelling distributions of substrate properties at a high spatial resolution. In addition to facilitating continuous spatial prediction, the high-resolution imputed angular models performed favourably compared to alternative approaches.