Steve: why would you say that I’ve argued the “falsity” of AGW? I’ve focused on proxy studies and what can be concluded from them. I haven’t attempted to analyze modeling literature as I have only so much time and energy.

But again, this can’t be checked one way or the other until Isaksson releases the long-overdue Svalbard data.

One down and three to go. Get them all and I’ll have to say some guys get all the breaks.

So my first question would not be, why does sea ice exhibit this behavior, but rather, why does the tree ring index (or d18O series) exhibit this behavior?

Or maybe there really were such breaks in climate, affecting d18O, TR, and sea ice together.

I agree completely and realized what you are contending here after I had written most of my post. I posted it as originally worded anyway with the thought that a reconstruction of sea ice extent if it has any meaning should be responding to its explanatory variables and further that I needed to test the attention span of potential responders out there in CA land.

On thinking about this further, I realize that you and Craig Loehle have published a reconstruction free of tree ring proxies and here we have in the Fauria 2009 paper data used that would allow a rather direct comparison of a tree ring versus non tree ring proxy over 800 years. I would suggest that you contact Fauria with an offer to coauthor a paper using the tree ring and dO18 data not together but in opposition as a test one on the other. Just determining the t-x lags to use could fill half the paper.

Actually sea ice extent, tree ring indexes and dO18 are all assumed to be related to SST and, therefore, a look at breakpoints of the tree ring indexes and dO18 would be in order – if the authors had provided that data or if I could find it.

I gather from Aslak’s remarks that co-author Elisabeth Isaksson controls the Svalbard ice core data. It sounds like she isn’t about to release it until she has coauthored a separate article based on every century of it. I don’t know who funded her.

The Fennoscandia series may be available somewhere, however, as “Fennoscandia” has been discussed on CA at great length already. I’m not sure what series are involved, or which one is which, but check out “Tornetrask Digital Version — Hooray!” and “New Light on Old Fudge” for starters. The fact that Briffa is not a co-author suggests that this is archived data.

My point was simply that a linear combination of an ice core series and a tree ring series does not constitute sea ice extent, even if is purported (unverifiably) to correlate with sea ice extent. I see no reason why the ice core data would exhibits the sudden shifts you show, but a tree ring series that splices together data from several trees might show such breaks, entirely without any relation to sea ice. So my first question would not be, why does sea ice exhibit this behavior, but rather, why does the tree ring index (or d18O series) exhibit this behavior?

Or maybe there really were such breaks in climate, affecting d18O, TR, and sea ice together.

But first I need to ask why you decided on five break points. Looking at the diagram I find it difficult to follow how what I regard as a very marked event at 1913 or thereabouts seems to have been by-passed by R. Is it perhaps because you specified only 5 break points?

Robin, there are 4 breakpoints and it is not I, but rather the program in R that decides how many breakpoints reach the level of significance by, I think, doing a least squares calculation as would be performed in a single trend except here multiple trends are allowed.

Hu M was merely providing me with a reality check – I assume. Actually sea ice extent, tree ring indexes and dO18 are all assumed to be related to SST and, therefore, a look at breakpoints of the tree ring indexes and dO18 would be in order – if the authors had provided that data or if I could find it.

However, I am fascinated by your plots, Kenneth (#82) and have some comments to make in the light of my own trials on the data. The data loaded into my software instantly and produced exactly the same plot as your diagram, without the fitted values that you show in red, of course.

I am not a user of R (hangs head in shame) because I find that such a dense language is too heavy for my aged brain to handle. I use my own “easy to drive” software that works very nicely indeed on a dataset such as this.

However, my methods for attempting to identify and quantify regime changes are totally unrelated to those provided by R, and as a result give totally unrelated estimates of change points! Unfortunately I can’t provide the diagrams here (how does one upload GIFs to this thread?) so I’ll just mention some prominent ones giving their tentative dates.

But first I need to ask why you decided on five break points. Looking at the diagram I find it difficult to follow how what I regard as a very marked event at 1913 or thereabouts seems to have been by-passed by R. Is it perhaps because you specified only 5 break points?

Based upon a cumulative sum chart followed by dummy variable regression my tentative estimate for the step change at around 1913 is about 210 units and is followed by a possible rising trend (slope 0.6, t=1.6). I also propose five other tentative regime changes, based upon quite simple reasoning. I could supply full details and plots, but preferably by email. Is it “safe” to provide an email address here?

I identify other “break points” (dates are approximate) at 1295 (upward, following a highly significant decline) to a stable regime lasting to about 1500. Then a downward step (80 units?) to another stable state lasting about a hundred years (~1600), an upward step of about 160 units followed by a very significantly declining period, slope -1.5/year, t=5.0, again for about one hundred years, to around 1700. Then an upward step of about 120 units leads to a stable period ending abruptly around 1913 as noted above.

In a diagram it all looks very plausible!

But, if Hugh is correct, it’s all to no good purpose

Internet connection down today. This will have to wait. Robin

Now restored (obviously!)

I forgot to add that it would be interesting to have the dO18 and tree ring data used in Fauria et al. (2009) available for breakpoint analysis.