Ocean Dynamics (2019) 69:1217–1237 1227 Fig. 6 Oxygen at the ocean surface averaged over May 2012 on both model grids. The upper row shows the experiment FREE. Super- posed to the model field are the in situ observations displayed as squares. The bottom row shows the mean difference of the experiments WEAK-FREE, i.e. the change in oxygen caused by the data assimila- tion. The model underestimates the oxygen in particular in the Baltic Sea where the assimilation increases the concentrations for the experiment FREE for both model grids. The in situ data values are plotted on top of the model fields. In the Baltic Sea, but also in the German Bight in the North Sea, the model mainly underestimates the oxygen concentration. The bottom row of Fig. 6 shows the difference between the oxygen concentrations from the WEAK and FREE experiments averaged over May 2012. The dynamic reaction of the model on the assimilation is to increase the oxygen concentration by up to 18 mmol/m3 in the Baltic Sea, which reduces the model bias. The dynamic reaction on the assimilation is much smaller in the North Sea with increases and decreases up to 5 mmol/m3. Figure 7 shows the comparison between the model concentrations and the in situ data as scatter plots. Consistent with Fig. 6, the main influence of the assimilation is to increase concentrations that are above 340 mmol/m3 in the experiment FREE. For the group of data points at about 350 mmol/m3 in the coarse grid, this leads to a slight overestimation of oxygen. Since also larger concentrations that are generally too low in the model are further increased the overall assimilation effect is positive. Thus, the assimilation reduces the RMSE and the amount of bias with statistically significance (at 95 % probability). However, the correlation between the model and the situ data remains essentially unchanged. The overall assimilation effect is similar in April and June, while it is lower for July. 5.3 Strongly coupled assimilation effect on the biogeochemical model ?elds In the strongly coupled data assimilation experiments STRONG-lin and STRONG-log, all BGC model fields are