Current seasonal forecast models involve simple schemes for representing sea ice, such as imposing climatological values. The spread of ensemble forecasts may in principle be biased due to common boundary conditions prescribed in the high latitudes. The degree of sensitivity in the 2-metre temperature, associated with seasonal time scales and the state of the June–August sea ice, is examined through a set of experiments with a state-of-the-art coupled ocean-atmosphere model. Here we present a suite of numerical experiments examining the effect of different sea ice configurations on the final ensemble distribution. We also compare the sensitivity of the 2-metre temperature to sea ice boundary conditions and sea surface temperature perturbation in the initial conditions. One objective of this work was to test a simple scheme for a more realistic representation of sea ice variations that allows for a spread in the Polar surface boundary conditions, captures the recent trends and doesn't smudge the sea ice edges. We find that the use of one common set of boundary conditions in the polar regions has little effect on the subsequent seasonal temperatures in the low latitudes, but nevertheless a profound influence on the local temperatures in the mid-to-high latitudes.