To generate winter averages for each season, the data were seasonally adjusted to the 1st January using an average winter growth curve derived from the data. The seasonally adjusted data were averaged onto 100 × 100 km grid cells on a polar stereographic projection. We estimated the errors in ice thickness for each grid cell as follows: The error in the ice freeboard was calculated using equation (1) from Giles and Hvidegaard [2006], with an additional term to account for the error in the velocity of the radar signal through the snow pack. The error in the estimate of ice thickness in each grid cell was then calculated from equation (3) of Giles et al. [2007], with errors in the ice and water densities taken to be 5 kg m−3 and 0.5 kg m−3 respectively [Wadhams et al., 1992], the snow density error taken to be 3 kg m−3 [Warren et al., 1999], and snow depth error taken to be the interannual variability (IAV) in snow depth of 0.03 m from Radionov et al. [1996]. For each winter season we then calculated ice thickness anomalies for each grid cell, which contained data for all years, by removing the six-year mean thickness for that cell. The error in the anomaly for each grid cell was also computed. Our grids of ice thickness anomalies were then averaged over the Arctic for each winter season, inversely weighted by the error on each grid cell, to generate regional or circumpolar averages.