Underwater acoustic sensor networks (UASN) can enable many potential oceanic applications. For the environment monitoring and event detection services, the sensor networks will record a large amount of data. But due to the harsh sea conditions and node energy constraints, real-time data delivery to the ground data center may not be feasible. Thus in-network data storage becomes a possible alternative. To enable efficient and frequent data access services, data-centric storage (DCS) protocols have been proposed for terrestrial sensor networks. However, the adverse underwater environment challenges the DCS protocol in two aspects. First, the unreliable underwater channel requires more robust design of long-distance multi-hop reliable data transport. Secondly, the high node failure rate demands higher reliability of the stored data. On the other hand, fountain codes have been studied for the advantages in data transport and storage. To adapt fountain codes into underwater DCS, we design decomposed Raptor codes (DRC) with threelayer encoding. In addition, a DRC-assisted DCS (DCS-DRC) protocol is proposed for reliable underwater in-network data storage. Analyses and simulations are provided to verify the performance and benefits of the DRC scheme and the DCS-DRC protocol.