1234 Ocean Dynamics (2019) 69:1217–1237 The right panel of Fig. 10 shows the development of the chlorophyll concentration profile in the Gulf of Finland. Here, the Baltic Sea is rather shallow and the profile is initially homogeneous, even though with rather high concentrations of about 40 mg/m3. However, on the 28th of April, the profile becomes more variable with a maximum concentration at the surface and a minimum at around 16 m of depth. Afterwards, the profile jumps to unrealistically high concentrations with a strong gradient from below 13 m and very low chlorophyll at the bottom. This gradient becomes even steeper in the following analysis steps. The high concentrations of chlorophyll are caused by high concentrations of flagellates, while the concentrations of diatoms and cyanobacteria remain low. The temperature increments by the data assimilation between the 20th and 30th of April in the eastern Gulf of Finland are always negative. The step-wise increase of the flagellates (and hence chlorophyll) concentration shows that the concentration is negatively correlated with the temperature during this time period. Given the larger assimilation effect with logarithmic concentrations, the unrealistically high concentrations develop. Actually, this effect is, to a lower extent, also visible in the experiment STRONG-lin with actual concentrations when all fields of the BGC model are updated by the data assimilation. In STRONG-lin, the concentrations increase to 170 mg/m3 in the eastern Gulf of Finland until the 15th of May (the top right panel of Fig. 9 shows increased concentrations already on the 1st of May). So, also in this case, the concentrations are not fully realistic. However, they are much lower than the concentrations obtained for STRONG-log and relax to realistic concentration levels until end of May. Overall, the assimilation in the experiment STRONG-lin behaves stable, while in the case of STRONG-log, the concentrations grow to extreme values and do not recover from this. However, if the phytoplankton variables are excluded from the assimilation update of STRONG-lin, their concentrations, including those of the chlorophyll, remain realistic. Thus, the cross-covariances between SST and the phytoplankton fields are not sufficiently well estimated to generate a realistic assimilation update at all times. This might be due to the larger errors in the BGC model state so that the linear regression between the SST and the concentrations fails. 8 Conclusion In this study, the effect of assimilating satellite sea surface temperature (SST) data into a coupled ocean- biogeochemical model for the North and Baltic Seas has been studied. The model uses nested model grids to better represent the circulation in the German coastal areas. The assimilation is successful in constraining physical ocean fields, which has been assessed with in independent situ data for surface temperature and salinity. With regard to the biogeochemical (BGC) fields, both weakly and strongly coupled data assimilations have been assessed. With the weakly coupled assimilation, the assimilation only directly updates the physical variables while the BGC fields react dynamically on the changed physical conditions during the following forecast phase. In this case, most BGC model fields are only slightly changed, e.g. oxygen by up to 5 %. The changes are particularly small in the North Sea. In the Baltic Sea, the phytoplankton concentrations and the chlorophyll and oxygen are slightly increased as a response to the assimilation. The validation with in situ data did only show small changes in the BGC fields. However, over the full experiment from April to June 2012, the improvements of oxygen concentrations were statistically significant. In case of strongly coupled assimilation, both the physical and BGC model fields are directly updated by the data assimilation method. When the actual concentrations of the BGC fields are used in the state vector, the assimilation behaves stable. The changes to the BGC fields are, as expected, larger than for the weakly coupled assimilation. Quite high concentrations of phytoplankton and hence also chlorophyll appeared in the eastern Gulf of Finland between end of April and middle of May if all BGC fields are updated by the assimilation. These high concentrations disappeared until the end of May, and the assimilation was overall stable. In contrast, the concentrations remained realistic if the phytoplankton variables are excluded from the assimilation update, so that only the nutrients and oxygen are directly updated. Thus, only updating the nutrients and oxygen when assimilating SST data appears to be the recommended approach. The strongly coupled assimilation was also performed using the logarithm of BGC field concentrations, which is the common choice when satellite chlorophyll observations are assimilated. In this case, the assimilation becomes unstable and local patches of unrealistically high or low concentrations developed. This was mainly the case in the Baltic Sea but also in the Norwegian Trench. The development of the chlorophyll was examined at two locations in the Baltic Sea, where particularly high concentrations developed. Vertical profiles showed that in the Gulf of Bothnia, the assimilation resulted in an unrealistic subsurface maximum of chlorophyll around 40 m of depth, caused by high concentrations of diatoms and flagellates. Ultimately, this maximum also influenced the concentrations at the surface. In the shallow eastern Gulf of Finland, the assimilation increased the concentrations of flagellates, and hence chlorophyll over most of the upper part of the water column. When a vertical localisation was introduced, so that the assimilation increments are linearly reduced as a function of depth until they are set to zero