Modelling the phytoplankton community in a front: a Mediterranean Sea case study.

In the ocean, fine scales (1-100 km) are short-lived structures (days to weeks) that drive ocean physics, chemistry, ecology, and can influence climate. Among them, fronts are ubiquitous fine-scale physical features that separate different water masses and create gradients of biogeochemical contents. Fronts are often associated with vertical mixing of the water column, allowing the availability of nutrients that support phytoplankton dynamics. However, how such structures affect phytoplankton distribution is not well understood, especially in oligotrophic regions. We hypothesize that the phytoplankton community observed in the frontal zone is a mixture of communities observed in the adjacent water masses, plus another community.

Here, we are interested in the community composition based on nine phytoplankton functional types (PFTs) observed by flow cytometry in a front, in the western oligotrophic Mediterranean Sea. During the PROTEVSSWOT MED cruise (doi:10.17183/protevsmed_swot_2018_leg2), south of the Balearic Islands, high-resolution measurements allowed us to collect samples in the front and the two adjacent water masses along a strong salinity gradient.

Our objective is to model the frontal phytoplankton community as a finite mixture of the adjacent water mass communities A and B, and a new community C. In this model, we specified that the communities in the adjacent water masses and the new community can arise from a discrete mixture of multivariate normal distributions. First, we estimated the parameters and number of components in the finite mixture for the adjacent water mass communities A and B using an Expectation Maximization algorithm. From a larger dataset, we estimated the parameters of a set of likely communities for C. Then, we developed a hierarchical Bayesian model to estimate the weight of each component of the discrete mixture. Finally, the hierarchical Bayesian model was run a second time using only the most significant components for community C.

One component was sufficient to model community A (North to the front), while communities B (South to the front) and C were modeled with two components. The new community C explained a significant part of the frontal community. With very few observations in the frontal zone (n=11), our Bayesian approach highlighted the spatial distribution of the phytoplankton community around the front. Our result suggests that local environmental conditions in the front allow the emergence of a new community. This work is a first step in understanding frontal zones in an oligotrophic region, representative of the global ocean. Our modeling approach will be further applied in a larger dataset (BIOSWOT-MED cruise, doi:10.17600/18002392). In these further analyses, environmental data will be included to disentangle the physical-biological processes that shape phytoplankton distribution.

This work was funded by the Institut des Mathématiques pour la Planète Terre which supports collaborations between mathematicians and life and earth sciences.