Compared to the TOPAZ4 reanalysis, INTAROS-opt simulates less freshwater content over the Arctic continental shelves and slightly more freshwater in the Canadian Basin (Figure 10a). One may argue that INTAROS-opt increases the freshwater content in the Canadian Basin at the expense of degrading the Arctic continental shelves. However, the mean freshwater content increment after data assimilation (Figure 9a) shows almost no changes in the Arctic marginal seas.
 Based on observations (Proshutinsky et al., 2009), the WOA18 atlas (Zweng et al., 2018), the PHC atlas version 3.0 (Steele et al., 2001), INTAROS-ctrl, INTAROS-opt, and the TOPAZ4 reanalysis, we computed the freshwater content in the Beaufort Sea (Figure 10c), the Laptev and East Siberian Seas (Figure 10d). INTAROS-opt changes the mean freshwater content without altering its variability. The mean freshwater content in the Beaufort Sea is increased from 12 km3 to 16 km3 after data assimilation (Figure 10c), but changes in the marginal seas are small except for the first year (Figure 10d).
Freshwater content (Figure 10a) and SSS (Figure 10b) remain different between the TOPAZ4 system and our reanalysis, especially over the Arctic marginal shelves. SSS in the INTAROS-opt is much higher than the TOPAZ4 reanalysis in the marginal seas, resulting in less freshwater content (Figure 10a). In the TOPAZ4 system, SSS is relaxed to a combined climatology of the WOA05 and the version 3.0 of  PHC (Steele, 2001) with a timescale of 30 days to complement limitations of seasonal river discharge and relatively coarse atmospheric forcing. In our reanalysis,  the salinity is mainly changed by adjustment of initial salinity and atmospheric forcing. In the marginal seas, the SSS restoring term is more efficient in changing SSS, while adjusting atmospheric forcing seems not as efficient. However, it may not improve freshwater content efficiently in the marginal seas (Figure 10d) and also damps the seasonal freshwater content variability (green lines in Figure 10c, d).
The WOA18 data remains an essential source of hydrographic observations to constraint the model’s climatology in this study. However, the differences between different hydrographic atlases remain significant (Figure 10c,d) regarding mean state and variability. Although the PHC atlas data (Steele et al., 2001) is more popularly used in studies of the Arctic Ocean, we note that the WOA18 atlas is closer to the observations (Figure 10c) concerning the freshwater content in the Beaufort Sea. Moreover, WOA18 reveals the secondary maximum of freshwater content occurring from May to July (see Figure 5 in Proshutinsky et al., 2019). Increasing the number of Arctic Ocean hydrographic observations and improving the quality of the Arctic Ocean climatology are required for further improving the model simulation through data assimilation.