1. Ice extent time series are available from the National Snow and Ice Data Center (NSIDC) based on the application of the NASA team algorithm (used here) and a bootstrap algorithm to the passive microwave brightness temperatures (http://nsidc.org/data/seaice/). Trends computed from both are negative in all months but those from the bootstrap series are slightly smaller (which yielded a September trend of –7.9% per decade). Trends are computed from anomalies referenced to means over the period 1979 to 2000. Surface melt in summer contaminates the passive microwave signal resulting in the underestimation of ice concentration. Use of ice extent (a binary ice–no ice classification) largely circumvents this problem.
2. Trends for all months are significant at the 99% confidence level based on an F test with the null hypothesis of a zero trend. Trends are also significant (exceeding the 95% level) based on the approach of Weatherhead et al. ( 3 ) which computes the trend significance from the variance and autocorrelation of the residuals.
3. Factors affecting the detection of trends: Statistical considerations and applications to environmental data
4. J. C. Comiso, Geophys. Res. Lett.33, L18504 (2006).
5. Ice thickness can be described from a probability distribution which has a peak at about 3 m. Although ice at the peak of the distribution is predominantly multiyear ice that has survived one or more melt seasons and thicker than younger first-year ice (representing a single year's growth) ridging can result in very thick first-year ice (up to 20 to 30 m).