SST shows consistency among the Arctic ocean-sea ice reanalyses (Chevallier et al., 2017). We compare INTAROS-opt with the TOPAZ4 reanalysis concerning the root mean square of SST anomaly difference (Figure 8a) and the mean difference (Figure 8b). The SST differences of the mean state and of the variability are located along the strong currents and the variable ice extent regions.  
We took the SST time series (Figure 8c) averaged over a 50 km ยด 50 km box near Fram Strait (black box in Figure 8a) as an example to examine and explain the differences between the reanalyses and observations. During September-November 2012, SIC observations averaged over this box (blue line in Figure 8d) show sea ice appearing from 27 September to 26 October accompanied by declining SST (solid and dashed blue lines in Figure 8e). Both INTAROS-opt and the TOPAZ4 reanalysis simulate a similar process as the observations (black and green lines Figure 8d, e), starting from 11 October to 23 October and from 27 September to 23 November, respectively. Based on INTAROS-opt, we diagnose SIC changes (DA) due to advective divergence (Sdiv), atmospheric thermodynamic effects (Satm), and oceanic sources (Soce, Figure 8c). Figure 8c reveals that the advective convergence ( ) effect dominates the accumulation of sea ice in this region. The underlying water cools as the ice is moving in. It melts the ice from underneath, which is shown by the negative contribution from the thermodynamic effect of the ocean. The small contributions from the thermodynamics suggest that the colder water moves together with the ice into the box, while surface fluxes have a minor impact. In the TOPAZ4 reanalysis (green lines in Figure 8d, e), sea ice emerges at a date similar to that of SIC observations, but more sea ice is simulated in the middle of October. During the decay, SIC is reduced (corrected) during each step (every seven days) of the data assimilation before building up biases again between two analysis cycles. In general, no dynamics can be inferred behind the SIC changes. The decay process in the TOPAZ4 reanalysis takes more time.
We conclude that overall, the adjoint method improves the SST both in its variability and its time mean. INTAROS-opt is consistent with the TOPAZ4 reanalysis, but SST differences remain along sea ice extent regions and strong current regions, which are likely to be related to the fast ocean-sea ice interaction processes and mesoscale processes.