2010-09-29 01:16:44Craig Loehle Post
Robert Way

I just noticed Loehle's new post... this one is gonna have to be dealt with. It is ridiculous.



I can't believe they let him re-center it like that. Clearly dishonest. The new reconstruction reminds me of moberg which makes me happier we've been using moberg.

I may feel obliged to go Tamino'esque and plot up the data myself. Perhaps we should consider writing something?
2010-09-29 02:04:52agreed
Dana Nuccitelli

Yes this is the sort of WUWT post which warrants a "response".  One which discusses a new peer-reviewed study, but which is misrepresented on WUWT.  My suggestion would be to make a passing mention of the WUWT article at most and simply do our own analysis.

Loehle's conclusion in that blog post is absurd.

"No unprecedented warming in recent decades, just a repeat of what looks to me like a periodic pattern of warming and cooling."

Based on this figure:

I don't know about you, but that nearly vertical spike from the instrumental temperature record looks pretty "unprecedented" to me.  And a periodic pattern?  You can't make a conclusion about a periodic pattern based on 2 cycles!  Totally ridiculous.  Oh gee it went down then up then down then up, I guess it must be periodic!  What a doofus.

I think a good SkS post would be to compare Ljungqvist (2010) to Moberg and perhaps Mann '08.

I love Tamino's plot of Loehle's "vindication":

2010-09-29 02:10:48NH temps
Dana Nuccitelli
By the way, the first commenter on Tamino's post noted that perhaps it would make more sense to use the NH rather than global instrumental temperature record since it's a NH reconstruction.  I think that makes sense too.
2010-09-29 03:09:42


OK, I just put up a blog post on this. 


Feel free to disagree with how I've handled it -- in particular, I am not convinced by Tamino's argument about centering.  (I wrote this on my own, having started it before Tamino's post was up.)


I think that Loehle's methodology is junk, and his conclusions are typically spun in a very favorable way by the "skeptics".  But there's not actually a huge difference in the results between his reconstruction (after the 2008 correction) and those of Mann, Moberg, etc.


I think the main points are


(1) that Ljungqvist 2010 vindicates Mann (and others)

(2) the MWP existed, but temperatures are warmer today

(3) Loehle's commentary on WUWT carefully avoids dealing with the fact that Lj 2010 is northern hemisphere only while his own work was supposedly global.

2010-09-29 03:25:16


Dana, i replied to Zeke that he's using NH data:

reading from the data file at NOAA: “Instrumental temperature data from the variance adjusted CRUTEM3+HadSST2 90–30°N data set”

sooner or later it will show up.

2010-09-29 03:34:01



It looks to me like Lj 2010 did their instrumental comparison using NH land+ocean data.  One might think that, since it's a land reconstruction, it would be more appropriate to compare to NH land only instrumental temperatures.  For my post on SkS I included both (CRUTEM and HADCRUT) so people can pick whichever they prefer.

2010-09-29 07:41:56comment
Robert Way

I'm a little confused by Ljunqvists commentary over on WUWT. He says amplitudes are underestimated in his and other reconstructions and for me that makes it sound a little like it is not clearly as warmer now.
2010-09-29 09:49:08Thanks for this blog post, Ned
John Cook


Great that you could get this posted so quick, usually I get nasty surprises when I wake up and check my website in the morning. Important to hit this kind of disinformation hard and fast which has happened at several blogs.

So great to wake up and find a pleasant surprise! Have just tweeted, facebooked and sent your blog post to the mailing list.

2010-09-29 21:06:58


Thank you, John.  I'm still nervous about posting stuff without running it by you first, so I'm glad to hear you don't disapprove.


Robert, yes, I noticed that commentary.  Part of the issue may be that multiproxy reconstructions don't have the temporal resolution to detect individual cold/warm years, just longer-scale variability.  The main thing to keep in mind, however, is that Lj's reconstruction is for mid- to high-latitude NH only, and we know that the MWP and LIA were mostly anchored in the mid- to high-latitude NH (see, e.g., Greenland ice cores) so a larger MWP in this reconstruction doesn't mean the MWP was all that warm globally.


As for whether it's warmer today ... well, in Lj's reconstruction, if you look at the range between the upper uncertainty bound during the MPW, and the lower uncertainty bound during the LIA, that gives you a worst-case scenario range of about 1.2C between the peak of the MWP and the bottom of the LIA.  If you then double that to a range of 2.4C (highly improbable IMHO) by adding 0.6 to the MWP and subtracting 0.6 from the LIA, that still leaves the MWP a smidge cooler than current temperatures based on the NH land-only instrumental record. 

2010-09-29 22:16:48comment
Robert Way

Thanks for the response Ned.

Myself personally, I've always held the opinion that post-1980s is unprecedented but that the early century warming could be part of the same 1000 year cycle that caused previous warm periods. Viau et al (2006) identify a 0.2 degree cycle that peaks every 1050 years roughly and has shown itself as the RWP and MWP. I certainly think the MWP was not as warm in the rest of the world as it was in high latitudes. Actually the MWP has the tell-tale signatures of being due to an intensification of the THC. (i.e. warm SSTs in the North Atlantic, warm Greenland and Scandinavia, warm Labrador). I think there's a recent study (can't remember its name) which agrees with this and finds a predominantly positive AMO during the MWP and a negative one during the LIA. Obviously these things don't force themselves so something must have forced it (solar?). Either way thank you for your great post and your explanation here.
2010-09-29 23:13:43


"Actually the MWP has the tell-tale signatures of being due to an intensification of the THC. (i.e. warm SSTs in the North Atlantic, warm Greenland and Scandinavia, warm Labrador). I think there's a recent study (can't remember its name) which agrees with this and finds a predominantly positive AMO during the MWP and a negative one during the LIA."

Robert, that would be a great subject for a post ...

2010-09-30 03:07:34comment
Robert Way

Well me and John talked about putting together an "it's the AMO" argument which could undoubtedly deal with this. My problem with doing it is that i'm waiting for to get some more information on this subject before I proceed further. I'm slowly gathering it up and such. I'm at a conference this weekend and am presenting twice so after this weekend i'll get ahead of things

I do have a bit of a technical question for you since you seem to have a pretty good grasp of statistics and such. If I'm trying to run a multiple regression on a dataset, do I use the whole dataset for the training period? For example, I use excel's data analysis toolbar and run a regression of temperature versus say 2 (X) factors. Do I run it for the whole period? I've had a professor tell me that is wrong and that you must run only part of the time period, then find out the equation and input the variables and predict the next part of the data. I find this a little confusing because I thought the multiple regression based model did exactly that, tested the relationship and tried to predict it using the variables it had?

Sorry if its a bit technical. IF you prefer not to answer publicly my email is rway019@gmail.com
2010-09-30 10:26:27


Well, there are different ways of doing this.  If you have lots of data, you might randomly assign part of it to a calibration or training set and the other part to a validation set.  However, there are lots of cases where we don't have huge volumes of data to work with, and it's not really necessary because the statistics associated with the regression model should give you a pretty good idea of how well it will perform.   The dangerous situation is when the sample data you have in hand for use in developing the model are not entirely representative of the data you'll encounter when you try to apply the model later.  This is most likely to occur when the data set used for model development is small or it fails to adequately cover the full range of conditions over which you'll want to apply the model later.

I don't know if this is helpful (and there are probably others in this group with deeper expertise than I in statistical modeling).  If you're uncertain, try doing something, then post it up here and let people take a look at it.  Or, if it's complicated, consider emailing Tamino and asking whether he'd be willing to "check your work" before you post it (he's a statistician, after all....)