A low-pass filter was applied to the daily polar sea ice extent data to remove short scale “weather” variability. This provides a reasonably smooth annual cycle giving a better indication of the general state of the ice than selecting a single daily max/min data point.
The zeros of rate of change of this filtered data were used to determine the “max/min” turning points in ice extent.
The interval between the two extremes was taken as the duration of the melting / freezing seasons respectively. Since the melting periods alternate in each hemisphere, the plotted antarctic “melting” is taken as the complement of the similarly calculated antarctic freezing period : ( 1 year – the freezing period ). In this way both are estimating the same 6 month interval.
3-sigma gaussian low-pass filters were used. A range of increasing values were tested to find the point at which the pattern of the results stabilised. This basically means removing sufficient short term variability to ensure a single change of direction at each extreme. The retained values of 13 and 20 days are indicated in the legend of the graph. The lighter filter being more sensitive to change.
There has been a notable shortening of the Arctic melting season since the “catastrophic” melting of 2007. Taking the average of the last two years as a typical recent value and comparing to 2005, the derived melting season has shortened by 9 days.
It is notable that there was an almost identical and symmetrically opposite change in the Antarctic in the years 2008 to 2011, inclusive.
Though not a perfect match, opposing tendencies in northern and southern hemispheres are also seen from 1985 to 2007.
Data for each hemisphere are provided in three files. Data up to 30th Sept 2013 in “final” files , data since Jan 2013 in “nrt” files given as “anomalies”: variations from an annual average “climatology”.
Data from “final” was taken until end Dec. 2012 then from “nrt”. Data before 1987 is at 2 day intervals and was linearly interpreted to provide equal spacing for the filters, as was a break in the satellite coverage for 42 days at the end of 1987. The resulting series for northern and southern hemispheres were then processed as outlined above.