<p>In this paper, we investigate the potential of using a probabilistic modelling approach in the prospect of ocean colour data assimilation. The main objective of the study is to assess the benefits of using error covariances based on an explicit simulation of model uncertainties. The relevance of this approach is evaluated by considering 3D observational updates of the ensemble (one update at one time step) performed every 5 days (over one year) using the statistics of a North Atlantic coupled NEMO/PISCES stochastic ensemble simulation involving 60 members, as previously described in Garnier et al, 2016. </p><p> In this experiment, SeaWIFS ocean colour data are used to update the ensemble with a low rank ensemble Kalman Filter analysis scheme. The non-Gaussian behaviour of the model variables is taken into account using anamorphic transformations. Comparisons between the updated ensemble and the MERIS satellite observations shows that the integration of high resolution SeaWIFS data significantly improves the representation and the ensemble statistics of chlorophyll concentrations. We also show that these improvements consistently cascade in the water column chlorophyll distributions and on non-observed variables closely linked with the primary production. </p><p>In addition, we present first results illustrating the potential of our approach for biogeochemical forecasts. The objective is to examine the model response to data assimilation in the perspective of future operational applications. For this purpose, we perform a 60 member simulation initiated from updated biogeochemical states. This forecast simulation shows that ocean colour data assimilation would be skillful considering integration cycles of the order of a day. Finally, the intend of this article is to point out the feasibility of operational biogeochemical data assimilation in the near future.</p>