Ocean changes
In the filter approaches, ocean states are changed by updating the state through assimilating observations and the propagation of the analysis increment by the forward model. The adjoint method adjusts the atmospheric forcing and the initial state to reduce the model-data misfits. These adjustments to the control variables are calculated by the information that is propagated by the adjoint model and the forward model. In this section, we concentrate on ocean changes after data assimilation and compare them with the TOPAZ4 reanalysis.
SST improvements
In INTAROS-ctrl, both the mean SST (Figure 7c) and SST anomaly (Figure 7a) show significant errors, especially in seasonal sea ice extent regions. The normalized root mean square of SST anomaly difference (Figure 7b) is significantly reduced with values ~1 in vast areas of the pan-Arctic Ocean. Near the central Arctic Ocean, normalized root mean square of SST anomaly difference can be as large as 4, indicating that the model-observations difference is still significant compared with the observation uncertainties.
SST is not observed by satellites under the sea ice, but we assume SST is -1.96 °C where sea ice is observed but not simulated in the model. INTAROS-opt reduces the mean SST errors by about 1 °C over the ice-free regions and by as much as -2 °C under the sea ice-covered area.