Ruffled Feathers But No lessons Learned

My previous post on the flaws of Goddard's colouring in and counting project seems to have ruffled a few feathers. Within a few hours he made several posts on sea ice extent, but has not learned from his mistakes - in fact he keeps making the same errors over and over.

In this post he is still colouring in but he doesn't rotate the image this time - what was all that about?

And Mr Goddard now proudly states;
"My pixel counts agree with Cryosphere Today numbers, which show more excess ice in the Bering Sea, than missing ice in the Barents Sea."
 I am at a loss as to why he has gone to the trouble of repeating the same flawed exercise to try and get his pixel count to agree with  Cryosphere Today. Why not just accept the information on Cryosphere Today? Perhaps because it doesn't suggest that the data "should be showing Arctic ice extent slightly above normal, for the first time in nearly ten years", his original and erroneous claim.

But he is still making the same mistakes. I hand you over to  Peter Ellis, a real sceptic and someone who has noticed problems with Goddard's analysis and similar problems with this map as I did in my original post. This is his comment to Steven Goddard;

Peter Ellis says:
April 19, 2012 at 3:33 pm

You count does not agree with Cryosphere Today numbers, since they show a current overall area anomaly of -0.103 million sq km. You can’t just pick two regions and ignore the rest. As far as your pixel count goes, even by eye I can see you missed some pixels in the Greenland sea. But that omits a larger issue: the pixels that comprise the orange line itself. These (being the border of the ice) presumably are ice-covered in the historical mean data. Since we cannot see through the line, we do not know how many of them are ice-covered today. The orange line is several pixels wide and thousands of pixels long. Your count therefore has a minimum error of several thousand, which makes your entire exercise moot.

Regarding the NSIDC figure, the historical average line is not strictly shifted, it’s actually a subtly different shape, since it’s now based on a 5-day average rather than a 9-day average. If you look carefully you’ll see the line is now slightly less smooth.

What does that do to the overall shape? Well, after normalising and taking the current date as day 0, the April 16 graph shows the average of days (-6,-5,-4,-3,-2,-1-0,1,2) , while the April 18th graph shows the average of days (-4,-3,-2,-1,0). That is, there’s no “shift” in that both are centred on day -2, but there is a difference in the size of the smoothing window.

How does the inclusion of days -6,-5,1 & 2 affect the average? Not at all during parts of the year when the rate of gain or loss is ~ linear. However, during this part of the spring, melt is accelerating, and so days 1&2 are further below day -2 than days -6&-5 are above day -2. So, during this part of the year, the 9-day average will be fractionally lower than the 5-day average.

Taken together, that means that going from the 9-day average to the 5-day average, even when correctly centred, will cause the “average” line to rise fractionally. It’s how smoothing works. The reverse will hold true as the melt slows down in August/September, so as we approach the summer minimum the “new” line for the historical average will be fractionally lower than it was on last year’s graphs.
Perhaps it is time Steven Goddard put his crayons away.


  1. Have You noticed, that the picture is labelled as "median" and the graph as "average", and they are not the same thing.

  2. No I didn't, and a good call sir. I wonder why they do it that way and how that will effect the outcome? I suppose extreme outliers will have less an effect with median than average. But this is just another example of Goddard's 'science'.