Wednesday, September 15, 2010

Again Arctic Ice

"Researchers say projections of summer ice disappearing entirely within the next few years increasingly look wrong.

At its smallest extent, on 10 September, 4.76 million sq km (1.84 million sq miles) of Arctic Ocean was covered with ice - more than in 2007 and 2008, but less than in every other year since 1979." (BBC news story)

Long term readers of this blog will remember my posts a while back on the subject of the area of arctic sea ice. A web page produced by NASA/JPL claimed that the latest data showed it continuing to decrease. The actual data, available on the web from the National Snow and Ice Data Center, showed that the decline had reversed, at least temporarily, about two years earlier. Commenters attempting to defend the NASA/JPL claim argued, in effect, that the reversal was only random variation, and the trend was still down.

That was somewhat over a year ago. As the quote above suggests, the evidence so far suggests that they were wrong. The minimum sea ice extent for this year is below last year's but above the figures for the two previous years. It is below the figure for still earlier years, due to the previous decline, but so far there seems no reason to believe that that decline has continued—which was the claim that I challenged.

And, if the BBC story is to be believed, researchers in the field have begun to adjust their predictions accordingly.


Tim Lambert said...

If you had said that the decline stopped in 2009 and started again in 2010, you would be wrong but at least consistent. Apparently your new definition of decline is synonomous with "set a record low this year".

This is not at all the conventional way of looking at trends in statistics. Whether you use frequentist or Bayesian statistics, the Arctic ice melt this year is evidence that the long-term decline is continuing.

David Friedman said...

It might be evidence that the decline is resuming, although pretty weak evidence. But "continuing" ought to mean on trend line. I believe you were the one arguing that two years of increased ice area counted as evidence that area was continuing to decrease.

Tim Lambert said...

Observations close to the trend line are evidence that the trend is continuing. In Bayesian speak they increase the probability that there is a trend increases. In frequentist speak the confidence increases.

You can, if you wish, you use own definitions of what a trend is, but you seem to be inconsistent in your own usage and it is wrong to accuse the JPL of dishonesty for using the definition you find in statistics texts.

Anonymous said...

Well the popular media has never been a good host for scientific ideas.

Francis said...

Mr. Lambert;

I'd like to ask you: what would be an annual value of the ice extent that would NOT confirm the trend?


Jonathan said...

Researchers are looking for summer ice in the wrong place. Just go to the nearest bar...

Anonymous said...


I don't think you understand trend lines very well. It's very much possible for the next sample in a trend line to be higher than the previous AND have that sample continue a downward trend.

Here is a simple example to illustrate, using a trend line constructed as the trailing average of three points

sample trend
110 110
100 105
90 100
80 90
70 80
81 77

Francis said...

Well, Anonymous, I don't think you understand trends very well. I don't know how you calculated what you call "trend", above, but it doesn't seem to make any sense at all.

Francis said...

By the way, Anonymous, I just remembered that, I think, Mr. Friedman has a Ph.D. in Physics. So presumably, he should be able to understand 3-point trailing averages. Now, in what sense can a 3-point trailing average can represent "a trend"?

Anonymous said...

David, have you taken a look at the PIOMAS Arctic ice volume analyses?

In any case, wishing that the Arctic sea ice is recovering to its historical values based on a very short-term trend (3 or 4 years) is a mistake.

Think of the numerical values as being "$ per share" instead of "millions km^2". Would a company whose stock price was numerically identical be considered to be "recovering"? I don't think so.