Maritime data can be defined as all data generated at maritime domain, including information about ships entering and leaving ports, cargo transported, and the VTS information that monitors and manages vessels. This maritime data is sent and received in a variety of formats and is multimodal, with each data type being closely related to the other, making it difficult to integrate and manage. Moreover, utilizing maritime data for AI systems requires knowledge of the data domain, which limits the ability of non-experts to utilize the data. In this paper, we proposed a data architecture that can effectively manage multimodal maritime data by utilizing the correlation of data. The proposed management architecture includes a preprocessing procedure for multimodal data, a graph database based on associations, and an object storage space for unstructured data, which can automatically extract and store associations from the collected data. In addition, we examined the feasibility of the data management system through the example of building a database of VHF data, and confirmed that it can improve the user's understanding of the data and improve data utilization compared to the existing data management system.