1232 Ocean Dynamics (2019) 69:1217–1237 Fig. 9 Chlorophyll concentration on the 1st of May, 2012 from exper- iment (top left) WEAK, (top right) STRONG-lin without vertical localisation, (bottom left) STRONG-log without vertical localisation and (bottom right) STRONG-log with vertical localisation of 5 m. While the vertical localisation improves the field, there remains an unrealistic high-concentration spot in the eastern Gulf of Finland concentrations are changed to a statistically significant extent. This change in the oxygen concentration can be mainly attributed to the changed temperature that changed the solubility of oxygen. Actually, for July 2012, the change in oxygen concentrations has nearly the same pattern, but reversed sign, as the temperature change in the bottom row of Fig. 5. Other BGC variables did not show a clear improvement. Mainly, we expect that the processes in the ERGOM model would react to the changed temperature. Thus, the growth of the phytoplankton groups is modified which affects the nutrient concentrations. The assimilation did not directly modify the vertical velocity so that the vertical entrainment of, e.g. nitrate is not modified. Anyway, this effect should only be present in the Baltic Sea and the Norwegian Trench, while the North Sea is shallow and usually well mixed. Given that the error in the BGC model state without data assimilation is rather large, and the dynamic reaction is small, the changes in the BGC state induced by the data assimilation are also small compared to its error. The strongly coupled assimilation resulted in larger changes of the BGC model fields. In particular, oxygen was further improved. However, the dependence of oxygen solubility in temperature makes it well (anti-)correlated to temperature. This correlation is expected to be represented by the ensemble; hence, the strongly coupled assimilation should improve oxygen. The dependence of other BGC fields on temperature is not that direct. For example, the nutrients will depend more strongly on the changed growth of the phytoplankton. Whether the ensemble-estimated covariances can improve, the model state also depends on the initial error in the BGC fields. Generally, the