Weather forecasts are increasingly made by earth system models that couple the atmosphere, ocean, sea ice and land surface. These coupled forecasts are best initialised from a coupled data assimilation system, especially if this uses satellite radiances that are simultaneously sensitive to surface and atmosphere. This is more consistent and optimal than using separate geophysical retrievals. Working towards this, microwave imager radiances sensitive to the ocean, sea ice and atmosphere have previously been used in atmospheric data assimilation, with retrieval of sea ice concentration (SIC) at observation locations as a by-product. The current work further activates the assimilation of these SIC retrievals, derived from Advanced Microwave Scanning Radiometer 2 (AMSR2), into a coupled ocean and sea ice model, using an outer loop coupled data assimilation approach. Improvements in modelled SIC are demonstrated indirectly in the marginal ice zones (MIZ) by the 12 h forecast fits to scatterometer wind and radar wave height measurements, where errors are locally reduced by up to 10 %. In the Antarctic summer, comparison is made to Ocean and Land Colour Instrument (OLCI) visible radiances and SIC estimates. These comparisons confirm the quality of the AMSR2 SIC retrievals but expose limitations in the modelled sea ice. The data assimilation also suffers spurious sea ice loss of up to 10 % in some Arctic winter locations, though effects on the atmosphere are locally contained, and sea ice losses are rapidly corrected by the forecast model. More generally, the atmospheric forecasts are improved in polar regions in the early forecast range, particularly in the winter seasons, confirming overall benefit from coupled SIC assimilation, likely due to improvements in the MIZ. Atmospheric forecast improvements propagate to midlatitudes by day 4, where wind and temperature errors are locally reduced by up to 1 %, indicating that sea ice assimilation can improve weather forecasts also outside polar regions.