Protection against sea submersion is a key point for the management of coastal areas. Operators need tools to monitor the aging of the sea dikes in order to reduce the risk of catastrophic events such as the Xynthia storm that occurred in 2010 (west of France). Long-term monitoring of these structures can be done using the ambient seismic noise produced by a combination of natural sources (e.g. the impact of swell on the structure, water currents, the wind force on ground-anchored constructions, etc.) and/or nearby anthropogenic sources (e.g. road/pedestrian traffic, coastal activities, etc.).Continuous ambient noise recordings can be used for monitoring structures by detecting small variations in seismic velocities related to localized degradations that cannot be detected visually. However, this monitoring technique requires sufficient energy in the proper frequency range for the intended application (typically at wavelengths of the order of the size of the structure or less). Additionally, the distribution of the sources must be relatively stable over time in order to interpret the velocity variations explicitly. The case of sea dikes is particularly challenging as the seismic noise is highly variable due various factors, like the water height variations during the tidal cycle, the variations in tidal intensity during the year, or the effects of the climatic conditions on the direction and intensity of the swell.This study is conducted in the framework of the SEEWALL project, which aims to develop a system for monitoring sea dikes using ambient seismic sources. A dike located on the Noirmoutier Island (France) has been instrumented with permanent accelerometers along with several geophysical and meteorological probes which have been recording continuously for about one year. This contribution focuses on the identification of the essential properties of the ambient seismic noise recorded in this setting and seeks to evaluate how these properties affect our ability to measure temporal variations of the seismic waves velocity. In other words, our objective is to identify the noise sources that contribute most favorably to the approximated empirical Green’s functions and/or that are highly repeatable over time, in order to develop methods that can be applied to various dikes and to optimize the type and amount of seismic data required to monitor such structures.