Species Distribution Models (SDMs) have been used to estimate one component of biodiversity, namely species richness. They aim to predict the geographic range of species (sometimes over time) based on observation data. However, species observation data result from complex and dynamic processes, and are often sparse in space and time, heterogeneous, imbalanced and inconsistent due to the variety of collection protocols, each introducing different biases. These characteristics pose a major challenge for data manipulation and interpolation/extrapolation, as well as for model evaluation. A new modeling approach, called Spatial Implicit Neural Representation (SINR), that jointly estimates the geographic range of a large number of terrestrial species on a global scale, based on presence data alone, was proposed in 2023, along with four geospatial benchmark tasks. It addresses some of the challenges posed by data in environmental science, making it a promising basis for the development of a model that would also incorporate the temporal dynamics of species distribution and interspecific interactions. However, transferring the model from terrestrial to marine data is not straightforward, as the influence of ecological factors on species distribution patterns can vary among environments and groups of organisms. For example, dispersal is more critical on average for terrestrial species, while ontogenetic changes are more prevalent in marine species. Furthermore, as the surface area to be covered is larger and less accessible, the amount of data available for marine organisms is much lower than for terrestrial organisms, and certain biases, such as sample selection bias, are exacerbated. All these aspects can affect the model's performance when applied to marine data. Here, we propose to qualitatively and quantitatively assess the validity of the SINR model for marine ecosystems. To do this, we use geolocated records of marine animals from the GBIF database, on a global scale, supplemented by other non-opportunistic data sources as the Jellyfish Database Initiative (JeDI). We also adapt the geospatial assessment tasks to marine data.