In December last year, reader Phi brought the Briffa MXD data from his 2013 paper to my attention. He showed the following graph of MXD data vs 3 different temperature series. Needless to say, it shows an impressive correlation between trees and temperature:
While Phi made the claim that the trees bear out UAH lower troposphere data over ground temps, I don’t see a single instance of a better fit of tree data to one temp series over another as particularly solid. This is particularly true considering that there are known divergent datasets. Still, it seemed reasonable that Phi had picked out an excellent dataset from the literature to look at. I took my time and downloaded UAH and RSS satellite data, the tree data from Briffa 2013 and found gridded data from CRUTEM4 using the google world map application from this RealClimate™ post. I actually went over to that blog to see if there was anything humorous to tease them about and found a very workable application – so shame on me! Of course, shame on them for having so much to mock but that is for another post.
It took a bit of fiddling with the calibration and filtering but I was able to reproduce a reasonably similar result to Phi. All temperatures presented below are from summer (June to August) averages of the 67.5N 67.5E gridcell closest to the tree data.
What I find amazing is how good a fit this data actually is to historic temps in the recorded period. First, recall that I made this red series above by simply aligning and averaging the data. I did this simple process with the understanding that some of the variance we see in these MXD series is from a statistically significant age related signal, so this series average is not as good a representation of annual tree MXD as it could be. Still, the age correction won’t make much difference and even the oldest portion of the data doesn’t diverge terribly from the black observed temperature curve. One of the main contentions I have with treemometers, besides massive non-linearity, is that the high frequency components and low frequency components aren’t necessarily governed by the same relationship and that those relationships with environmental conditions will change over time. e.g. how does the same tree respond to temperatures in low water vs high water conditions?
Anyway, I looked at lower frequency response in the following plot:
The data from this set is truly fantastic compared to some we have looked at but you can see a large divergence of temperature above tree latewood density in recent years and a similar problem in 1880-1990. We could shift the graph up and down to try for a better fit but it seems pretty obvious that the trees are reacting to other environmental conditions than temperature as years go by. The visual correlation is still amazing though.
While it may be tempting for climate scientists to take this kind of data and paste temperature onto it, calling it a development or something, they may not find the whole reconstruction that exciting. The rest of the data is fairly interesting to those of us skeptical of the general exaggerations pervading the science of global warming doom. Below is the full MXD reconstruction with temperatures to 2006 overlaid. 
Like this years RedWings, it seems to be a hockey stick without a blade. I’m still amazed at the quality of the fit to CRUTEM though and have decided to continue this study and take the next step of correcting by the average growth curve and perhaps the pith offset as well.
For completeness, and so the alarmist climate community doesn’t have a heart attack, if we extend CRUTTEM and UAH past the end of the reconstruction to 2012, the graph looks like this:
As always, I intend to make the code available. Unless someone is interested, I will clean it up and post it with my future calibrated reconstruction post.
the Air Vent 
I’ve always maintained that tree rings can tell you one thing: whether there was a good growing season or not.
Weather is part of that, so there should be some connection with temperature. Among a number of other things.
An idea I just had: what if apparent correlations with temp are actually correlations with sunlight, that is, cloudiness? In general, clear days may be warmer, but that won’t always be the case.
TTCA, are you suggesting that somehow trees grow using sunlight? Next you’ll be claiming that they eat air!
Yep – just the carbon dioxide from the air mixed with sunlight and water …
http://en.wikipedia.org/wiki/Photosynthesis – they take next to nothing else from the soil. Without CO2 all plant life dies.
He he.
Are you sure the mxd data is actual tree ring data and not CRUTEM with a little noise thrown in?
At the risk that you are joking, yup. Downloaded myself.
I wasn’t suggesting you made a mistake. I was suggesting the original source made stuff up. Desperate scientists have done worse.
With the Global Precipitation Mission being launched today it is timely to note that precipitation holds the key to the answer to the trillion dollar question which is “Does gravity induce an autonomous temperature gradient in all solids, liquids and gases?”
Josef Loschmidt first postulated that it would in the 19th century. Dr Hans Jelbring worked on it for his PhD and published a paper about a decade back. Now physicists are starting to realise that it is indeed a reality, and this can be shown using the Second Law of Thermodynamics in conjunction with Kinetic Theory.
But, most compelling of all is the empirical evidence which I have presented in a book “Why it’s not carbon dioxide after all” being released late April. Temperature and precipitation records are used to show that regions with higher precipitation do in fact have lower mean daily maximum and minimum temperatures than drier regions at similar latitudes and altitudes.
This means water vapour cools. And this is evidence that gravity produces a “dry” gradient (aka lapse rate) at the molecular level (not requiring a hot surface or upward convection) and water vapour then reduces that gradient (as is well known) due to inter-molecular radiation (not well known) and this leads to lower surface temperatures.
The greenhouse is smashed,
Seems that CO2 fertilization is supported in these graphs also!! 8>)
Under CO2 fertilization trees continue growth into older age and can restart growth after seemingly stopped and are more resistant to droughts. They also appear to more efficiently use available nutrients and soil chemistry makes more available. The Idsos are a wealth of information on these issues collecting studies by others along with their own.
A great example was the recent announcement of the older Redwoods restarting growth here in California. It was, of course, attributed to Gorebull Warming!!
http://www.wunderground.com/news/climate-change-may-help-redwood-trees-thrive-20130818
Jeff Id,
Congratulations for this great work. I am pleased that one is interested in these data, they deserve it and I think they have much to teach us.
The graph at the top was a first version that unfortunately has some holes in CRUTEM data, I could partially fill it afterwards: http://imageshack.us/a/img21/1076/polar2.png
This is not completely trivial, we see in particular that the correlation with UAH is truly superior.
I calibrated MXD on the temperatures of the lower troposphere because I have no confidence in stations data : the likelihood of perturbations is important and the methodology of adjustments is in my opinion untenable. Comparing correlations seems to justify this.
Regarding the multisecular reconstruction, I feel that your process suppress the long-term variability. The simple average of densities has not this issue, but we can see the problem of selection bias: http://imageshack.us/a/img22/2053/44qm.png
So I decided to take a look at the UAH LT data and the CRU ground data in the region. I downloaded from Climate Explorer, so unfortunately I had to get UAH v5.5, instead of 5.6, but it shouldn’t make a significant difference. At any rate, I went with 65-70 N 65-70 E for both-if I am understanding the region in question correctly? Anyway, I then baselined to 1979-2013, and finally took the average of 11 and 13 month centered averages:
Note: red is CRU, blue is UAH.
Okay, major red flag: the land surface data shows much greater variance than the data for the lower troposphere, in addition to a lower trend.
Several approaches to correcting for this-or rather, finding the correct factor to multiply/divide LT data by to match the surface to identify potential long term trend bias.
First: assume no bias: just regress the raw anomalies. Best estimate coefficient: ~.54*CRU = UAH, or UAH/.54 = CRU. Same with annually smoothed: ~.43*CRU=UAH or UAH/.43 = CRU.
Second, don’t assume no bias: assume bias in either data set is removed by removal of linear trend: regress detrended anomalies and detrended annually smoothed anomalies:
~.54*CRU = UAH or UAH/.54 = CRU, ~.41*CRU = UAH, or UAH/.41 = CRU.
Finally, ratio of standard deviations, for raw and detrended: CRU-STD/UAH-STD = ~0.63, CRUD-STD/UAHD-STD = ~0.63.
Okay, so taking largest and smallest coefficients (and mean coefficient ~0.53) and dividing UAH anomalies by each. Now take the annual smooth. And finally let’s graph up all three normalized versions of UAH and also the CRU for comparison, and put trend lines through them:
As you can see, once you apply a variance correction, the trend in this region is basically the same measured at the ground, or by satellites.
It is worth keeping in mind that if you did something similar on a global scale, it would imply CRU or whoever does have a warming bias over the period. But in this particular region, there is no evidence for a warming bias in the surface temperature data.
I’m not a mathematician … but it appears that the mxd data underestimates temperature spikes way more than it overestimates cold periods.
So if 50 years in row the MWP was 2C or 3 warmer, the mxd data would imply it was only a little bit warmer. Optimum growth does NOT occur when it is hottest.
Treemometers, for the most part, always underestimate really hot years.
Thinking about it a little bit more, drought decades would appear to be slightly colder than normal too.
Is there rainfall data for the location / period in question? We could create a table.
UAH | Rainfall | MXD
————————–
Hot | Low | Low
Optimum | Optimum | Highest
Cold| Wet| Low
etc
There are a number of precip datasets at Climate Explorer. Tricky thing is whether there is good data for this gridcell-also, precipitation is very “spotty” compared to temp, so even if there is a rain gauge within a few kilometers, it may not be representative of the location of the trees.
Still, might be worth looking into, although a lot of questions arise as to what data to use, given the number of different datasets.
I agree with most of the comments here. Tree response to temperature is not linear so I suspect variance and peeks of tree temps are muted by that factor as well as what low frequency signal that might exist by simple averaging math. Sunshinehours1 also noted that the HF variance seems stronger to the negative side which hints that trees can only grow so healthily on the positive side but they can completely die the other way.
It’s just a pile of treemometer data but it is fun to see good correlation.
Here’s something kind of telling: it’s the CRU precipitation data, annual totals in mm, for 65-70 N and 65-70 E:
I see some similarities with the tree data.
TTCA,
I haven’t looked at precip data but would like to do a comparison. Do you have a link handy for a lazy guy?
I downloaded from here:
http://climexp.knmi.nl/select.cgi?id=someone@somewhere&field=cru311_pre
Just put in the coordinate ranges-I usually say 0% of missing data just to get the most temporally complete record possible.
thanks
JeffID, it is good that you are willing to look at results and analyze those results which, at least at first glance, appear to show a temperature proxy reasonably well correlating with temperature under some filtering conditions. Do you have a link to the MXD data?
Are there station temperature data available within a reasonable distance from the trees used for MXD?
It would seem that using grid data would be called for only if the nearby and nearly complete station data were not available. Obviously to compare satellite and land temperatures you have to work with grids.
I assume that it was raw MXD data that was used here and thus any change in how the data were manipulated would not account for the better correlation of proxy response to temperature.
I did not see a calculation of the correlation, r, of MXD to whatever annual months you chose for temperature. Also what would a sensitivity test show using more months of the year for temperature. Did you use mean temperatures? Why should we expect the MXD to correlate better with UAH data than CRU data? How is the correlation with GISS and GHCN temperature data? Have you calculated trends for the MXD response and temperature over the instrumental period?
The dilemma of those doing temperature reconstructions, and particularly with dendro proxies, is that one can sometimes obtain reasonably good higher frequency correlations between proxy response and temperature but at the same time produce very different trends between temperature series and proxy response series. That is often the case with the divergence problem. In your plots the divergence problem is evident with the CRU data.
Proxy responses and temperatures are mostly used when constructing reconstructions to determine trends and that means that it is the lower frequency correlations that are most important. That is also why the divergence problem is a reconstruction wrecker.
I find it of interest that the temperature and MXD response for this area show no trend over the entire 1900-current time series. That is surprising given the Arctic polar amplification and the faster warming in the higher latitudes seen as a general case. Are there any explanations or conjectures for this exceptional case.
Lots of questions and I don’t have a lot of answers. I have done a bunch of the correlation checking but nothing post-worthy yet. I’m interested in correcting for age variance and in looking into the selection bias Phi alludes to.
Correlation to UAH is not expected, but Phi seems to think it proves that UAH is better than ground data, which I flatly don’t agree with for many of the divergence reasons you point out here. The 25 year crutem (Figure 3) does have some trend and fitting a line to recent years produces quite a steep upslope so it looked reasonable to me. I did stop it at the same year as the reconstruction (2006) and there is more uptick at the recent years so that reduces some of the HS effect.
Phi is right that averaging the way I did will repress long term variance so I have been considering modifying Roman’s hammer code to regress the pile together. We could publish a paper on that in dendro probably because I haven’t seen anyone do it that way. While I’m impressed with the HF correlation, I’m not sold on the data yet.
I found the polar.mxd here: http://www.cru.uea.ac.uk/cru/papers/briffa2013qsr/PY_Chrons.prn . Also the CA discussion on Briffa(2013) is here: http://climateaudit.org/2013/05/24/briffa-2013/ .
Yup, sorry I didn’t get back sooner.
This graph may help to understand why I trust more in lower troposphere data for regional surface temperatures (at least in this particular case).
The study of TTCA is very interesting but there is no tropospheric attenuation in JJA.
To clarify : for JJA detrended, UAH = ~0.86 * CRU.
In terms of trends, and always for JJA 1979-2012:
HadCRUT4: 0.445 ° C per decade
UAH: 0.032 ° C per decade
Difference in trend: 0.413 ° C per decade.
What I have been trying to explain, and what graph 2 demonstrates, is that there is enough error in the difference between thermometers and trees that you cannot rely on trees to be thermometers. You must realize that it is a stretch to say that trees are lower troposphere sensors simply by the fact that the lower troposphere is functionally not physically in contact with the tree. I realize that you don’t trust the ground thermometers which I don’t blame you but to reach your conclusion trees cannot respond to anything except temperature. I sure as heck don’t trust treemometers to NOT respond to factors other than temperature.
It is true that I am interested enough to try to prove myself wrong with this particular data but nothing I have ever seen convinces me that it can be done. Instead Figure 2 is fairly convincing that the trees are responding to more than temp. –as one would expect.
Jeff Id,
MXD does not respond to low troposphere temperature. The problem is not there. It is clear that regional surface temperatures profoundly influences maximum densities. There are other factors but obviously the link with this temperature is strong. MXD are until proven otherwise a good proxy for regional temperature. TLT is also apparently a good proxy of the same temperature (seem poorer for high latitudes winter months). The main question is as follows: are temperatures of stations a good proxy for regional temperature ?
Sometimes yes, sometimes no. What is clear is that station data are not a the gold reference. An objective consideration of the problem does not allow a priori to put the divergence on the back of MXD. In our particular case, there is no divergence MXD-TLT.
And for good measure, I give once more links to strikings evidences of the superiority of stations thermometers :
Thermometer of Geneva corrected by Giss http://img833.imageshack.us/img833/166/giss.jpg
Thermometer of Geneva corrected by MétéoSuisse http://img715.imageshack.us/img715/261/natva.jpg
Themrometer of Brussels corrected by Giss http://img805.imageshack.us/img805/9535/uccleg.jpg
I’ll look into whether there is damping in summer. I kind of find the suggestion surprising and maybe not something that would be obvious or clear.
I do wonder, based on other work I have done, whether snow cover might bias the brightness temperature somewhat. It’s worth further study.
Could be the antisymmetric of the hot spot. Something that models seem to reproduce quite well.
Only if the surface temperature data are strongly warm biased in the tropics.
Which, considering the quality of data I’d expected to come from those regions…Yeah, I’d by that.
And it’s not even a general problem, it is in fact sufficient to eliminate the land values to get back on its feet, it’s ok with oceans.
I have recently browsed through Briffa (2013) and found some interesting and telling statements in this paper:
(1) The authors state that the strength of the underlying common variability (correlations) in tree-ring chronologies is greatest at short time scales and with longer time scales generally showing lower levels of common variability. That phenomena is what I pointed to in an earlier post in this thread with reference to the divergence problem.
(2) The authors note that it is the growth rate of trees, and not simply age, that is critical to the proper adjustments that the RCS process is expected to make. The authors additionally note that trees used for coring are often the larger and older trees that as a result are faster growing trees. These trees if adjusted based on age alone would bias the modern warming period to higher apparent temperature responses. The authors indicate that they have added to the RCS procedures to take into account this phenomenon, but do not show directly (which may not be possible or difficult to accomplish) that those adjustments are completely successful.
(3) The authors note that both, TRW and MXD, require standardization to avoid including the effects of changing tree geometries in the longer term climate related trends.
(4) The authors have evidently made what they judge are necessary adjustments to TRW and MXD through a succession of fine tuning methods that would appear to be in general subjective. An outside observer of all these problems and required adjustments noted by the authors of Briffa (2013) might well judge that the art/science of dendroclimatology is a work in progress. I would also think that a scientist outside the influence of the climate science community doing temperature reconstructions like, Jim Bouldin, should be taken seriously when he states the current state of dendroclimatology is a mess.
http://climateaudit.org/2013/05/24/briffa-2013/
(5) Figure 2 in Briffa (2013), if I correctly interpret it, shows spaghetti graphs of the separate Yamal sites of standardized chronologies with the top 2 graphs showing the results of chronologies obtained separately from the sites and a bottom 3 graphs showing the results of chronologies obtained using pooled data from the entire Yamal region and the a further spline smoothing of that result. The top 2 graphs show no modern warming while the bottom 3 show a curious modern warming as a plateau higher than the preceding times starting around 1920 and ending at the end of the series around 2006. That 1920 starting date is much earlier than the 1970s normally used as the start time for significant increases in green house gases (GHGs) and resulting AGW. The increase in warming attributed to this increase in GHGs would obviously not be a plateau.
(6) Most disconcerting to me about Briffa (2013) is strong evidence that the paper has let advocacy get the better of science in being deceptive about the manner in which the results are presented and discussed. An example is the authors referring to an unprecedented 100 years of warming over past warming without pause to discuss how that warming and its shape in their graphs fits with GHGs. In order to obtain that warming plateau the authors have used spline smoothing which is known to have spurious series end effects – depending on the spline parameters selected. The authors claim the spline smooth is used for ease of graph viewing and then turn around and use those spline smooths as evidence of the unprecedented 100 years of modern warming.
I also have to look deeper into what data was used here in the JeffID/Phi analyses and how it was filtered. Any comments on that would be helpful.
The Briffa (2013) paper and SI (SM) are linked here:
http://www.sciencedirect.com/science/article/pii/S0277379113001406
Kenneth Fritsch,
The polar.mxd file contains the raw maximum density. To eliminate selection bias I have simply normalized each series on a reference period.
V : all of the values in a series,
V’ : values for the years 1951 to 1980
m’ : average on V’
s’ : standard deviation on V’
Vr : resulting values of the series.
1. The series is eliminated if there are not 30 values in V’.
2. Vr=(V-m’)/s’.
The reconstruction is simply the result of annual averages of all Vr.
I chose a 30-year period to obtain m’ and s’ significant enough and I started from 1951 to not remove too many series in the twentieth century.
Otherwise, your comment on the smoothing is quite relevant. This trick is unworthy of a scientific study, unfortunately, this is a rule rather than an exception in the field of climatology.
Jeff, I am a bit confused that you label some MXD graphs as Yamal. I believe that Briffa(2013) states that Yamal was all done with TRW and that Polar Urals used both TRW and MXD. The entire extended region for the Briffa reconstruction is called by the authors, Yamalia. Polar Urals is centered at approximately 67N and 68E and Yamal is centered approximately at 67N and 70E. The nearby climatological station is Salekhard at near the same latitude and longitude as Polar Urals.
I will keep it straight in coming posts. These were from the polar.mxd file.
Looking at the 1000 yr data plot, it strikes me that this is what I would expect for a tree that has an upper growth rate limited by biology. Boreal trees can’t grow like bamboo or a rainforest tree. So all the upper points line up for good years. But the lower limit for a bad year can be very far down from the mean, close to zero growth (true zero will kill it). This means that truly warm years will not show up.
I was thinking in somewhat this same vein in noting that the instrumental data and TRW show little or none of the expected accelerated warming in the modern period for this region which would normally start in the 1970s. TRW can track the high frequency variation of temperatures reasonably well but the problem as stated in Brifaf(2013) is the longer term variations. Since the TRW responses and temperatures appear to plateau.in the 1920-2006 time period there is no longer term variation in temperature for the TRWs to (fail to) respond.
I really need to finish my look at the Briffa (2013) data for Yamal TRWs and Polar Urals TRWs and MXD before it is treated and filter as it was for their paper.
I have a couple more observations from that paper. The TRWs for Yamal and Polar Urals both show that 1920-2006 plateau, while the Polar Urals MXD shows no modern warming in any form. For Polar Urals either the MXD or TRW would have be wrong. The fact that this difference is not discussed in this paper again makes me wonder about advocacy getting the better of science. I will credit the authors of this paper stating in the paper concerns about selecting proxies after the fact of comparing the proxy responses with temperature. They are not clear what problem they were avoiding and whether it was the selection after the fact problem. I rather doubt this to be the case since that most of the proxy data were previously measured and these authors are merely fine tuning the chronology methods. Also the observed temperature series in the paper are not shown until the second to last page – excluding reference pages. Notice also that on that page in Figure 13 the fine print noted that the spline smoothing process can make the end of series presentations uncertain due to end effects. This avoids I suppose discussing the issue of using the choice spline parameters to spuriously depict a wide range of ending flourishes. In the case of Briffa (2013) the nod went to showing a dramatic ending upward trend. Whoda thunk? I think I will make part of my analysis of Briffa (2013) the showing of how different spline smooth parameters can change the ending time series curve. I have seen this use of a spline smooth in other climate science papers and without any warning.
This is a good point Craig. Just because tree lines are temperature limited does not mean that that temperature is the limiting factor in all years for all trees. Think of the set of all possible sample trees in a given temperature limited area as plotted mxd vs temp on a scatter plot. It is only the upper envelope of this data cloud that would represent the true temperature limited trees. Sample points below the upper envelope edge would represent growth limited by something other than temperature which by definition lies below (and never above) the edge of the envelope.
I have sometimes thought that some type of data envelope analysis method could be done by sub-setting tree age during the instrumental period to identify age related differences in growth. If there are truly age related differences in growth then this should be reflected in differing upper envelopes in the growth vs temp scatter plots.
JeffID, I have linked below my initial look at the Polar MXD series for the period 1889-2006 by calculating correlations with the gridded June July CRU Tem4 and GISS 250 km temperature series and trends for all 3 series for the two periods of 1889-2006 and 1950-2006. What I found was correlations of around 0.60 that were very much the same for CRU versus Polar MXD and GISS versus Polar MXD. The CRU and GISS trends were high and in line with what one would expect from this region of the globe. The trends for Polar MXD were quite low and very different than those for CRU and GISS. I have not double checked these results but I intend to move on to the UAH and RSS satellite temperatures versus Polar MXD.
At this point this exercise presents an excellent example of what I was referencing when I mentioned that one can obtain reasonably good high frequency correlations with these tree proxy and temperature series and yet have very different trends, i.e. low frequency responses.
JeffID, I have linked a second version of the correlation and trend data. Please note that I corrected the MXD trends to be in line with temperature by way of a simple regression of temperature versus MXD units for the 1889-2006 period. The MXD trends change very little, the negative trend 1950-2006 for MXD remains and the large temperature trends, of course, stay the same. These MXD and temperature data were all previously put into anomaly form using 1889-2006 as the base period.
I should also note that the 10 nearest meteorological stations to the Polar (and Yamal) site all had 1991-2004 missing and some ended at 1991. That year corresponds to the breakup of the Soviet Union. That is why I am using gridded data. GISS 250 km extrapolates up to 250 km to infill missing data, but I do not believe either GHCN or CRU extrapolate for infilling.
I have just noted that the raw.zip file from the link below at Briffa (2013) has temperature data from Salehard that has the data for 1991 and beyond. Now I know that HadCRUT obtains their data for Russia from GHCN and at KNMI that data are missing for Salehard. That Briffa (2013) page also has some adjusted temperature data for the for the Yamal Polar Urals area. At this point I do know how that data were adjusted, but before moving on to the satellite data I will do correlations and trends with the temperature data from the Briffa page.
http://www.cru.uea.ac.uk/cru/papers/briffa2013qsr/
Jeff, I went back and did the same calculations from the Briffa (2013) materials page using their adjusted and Salehard temperatures as I did above using the gridded GISS 250 km and CruTem4 and obtained the same results. That shows that my results are not different because of the temperature series I used. The MXD series is the same as you used in this thread because it matches exactly. I note here that I receive correlations of around 0.60 for temperature to MXD while Briffa (2013) reports correlations around 0.80-0.85. I do not know at this point if the authors would correlate using smoothed data but I would hope not.
Anyway using these data I find that the MXD proxies are showing very definite divergence at the end of the series. Next I’ll look at the satellite series.
Jeff, I looked at the UAH temperature(65.0-67.5N and 67.5-70.0E) versus MXD for Polar Urals for the 1979-2006 time period in the manner I did above for GISS and Cru temperatures 1889-2006 and 1950-2006. What found was an even worse case of divergence with UAH having a trend of 0.87 degrees C per decade versus the equivalent for MXD of 0.01 for the 1979-2006 time period. The correlation for MXD versus UAH was 0.48.
What I see from my calculations is very different than what I see from the graphs in this thread???
I do want to look at the Briffa (2013) TRW versus temperature in the same manner as I did above for MXD.
I’m intrigued. Your values may be consistent with the graphics of Jeff Id but much less with mine and with the values of Timetochooseagain (# 6).
Mine are as follows for UAH 5.6 JJA 65-67.5 E 65-67.5 N:
1979;-0.813, 1980;-2.057, 1981;1.493, 1982;-0.143, 1983;0.587, 1984;0.450, 1985;-0.797, 1986;-1.403, 1987;-0.047, 1988;0.400, 1989;1.007, 1990;1.227, 1991;0.160, 1992;-1.753, 1993;1.520, 1994;1.440, 1995;-0.283, 1996;-0.863, 1997;-2.827, 1998;1.073, 1999;-1.683, 2000;1.167, 2001;0.433, 2002;-1.327, 2003;1.237, 2004;0.277, 2005;-0.033, 2006;0.027, 2007;0.720, 2008;-0.323, 2009;-0.827, 2010;-0.903, 2011;-0.850, 2012;1.360
With that, I have a trend of 0.032 ° C per decade.
Do you have other values?
Sorry, this is UAH 5.5 but there is little difference with version 5.6.
I’m sorry Phi, can you give a link where you obtained your UAH series? I used the gridded monthly data and downloaded the whole planet.
JeffID, we cross posted but Phi’s data is the same as I used. The huge difference is in the period used. My data were from KNMI.
I also tried different periods which made a big difference for me as well. I left my software in a different set of months from the post but it is possible that I grabbed an extra month between june and august somewhere. Rss has a ton of trend compared to UAH
Phi, I went back and checked my calculations which were correct but I found that instead of using the months June and July as suggested in Briffa (2013) I inadvertently had used May and June. What I have done was calculate the UAH temperature trends for 1979-2006 and correlations of UAH versus MXD for the 3 cases: May/June; June/July; June/July/August . The sensitivity to period selected was amazing.
The results were for UAH trends: MJ =0.87; JJ =0.10 ; JJA =0.10. The JJ and JJA trends were not significantly different than zero.
The results for the correlations UAH to MXD were: MJ =0.48; JJ=0.70; JJA= 0.80.
Using the data you listed above for UAH is consistent for what I found for JJA and thus we are using the same data. The difference is the extreme sensitivity to time period used. I am going back and do the GISS and Cru versus MXD for all these time periods.
I get a UAH JJA of 0.15 +/- 0.33C/Decade and rss 0.24 +/- 0.364. Both had low autocorrelation so the CI isn’t bad.
JeffID, I have gone back and did the trend and correlations for the months May/June; June/July and June/July/August for UAH (1979-2006) and CRUTem4 (1950-2006) and linked the results below. Using the months JJA reconciles my calculation with what you show in your graphs here, Jeff. It also shows that the correlation of MXD response and temperature is better with CRUTem4 data than UAH for periods analyzed.
I need to look further into the very different picture painted using May/June versus June/July/August. Obviously one could note that the JJA period of the year perhaps hits the sweet spot for tree growth and manifestation of temperature correlated MXD. That situation would put the MXD as thermometer in an even better light. On the other hand, the good high frequency correlations
occur here where the temperatures show no significant warming and thus there is no low frequency trends to follow. The Briffa (2013) authors state that the higher frequency responses of TRW and MXD to temperature is more reliable than the lower frequencies. I need to go back and compare the detrended temperature for May/June with the MXD to insure that the higher correlations in the JJ and JJA are not simply artifacts of being compared with little or no trend in the temperature series.
I also did correlations of the CRUTem4 temperatures for the period 1889-2006 for JJA versus annual and obtained r = 0.53 which indicates that the JJA variation accounts for approximately 30 % of the variation in the annual temperatures. Also interesting is that the JJA temperature trends in the Arctic region for recent times would not be larger than what we see here. The Cowtan Way (2013) paper recently published and discussed at the Blackboard produced gridded temperature series that showed considerably higher temperatures for Arctic region than did CRU, but those temperature difference were greatest in the winter and spring.
Jeff, I went back and looked at the correlations of MXD to CRUTem4 (1950-2006) and UAH (1979-2006) detrended for the months May/June and the correlations improved from 0.63 to 0.68 and 0.48 to 0.55, respectively. While improving the correlation, the detrending does not get you to the 0.80 and 0.85 levels of the JJ and JJA months.
I did another calculation of trends of the entire year for CRUTem4 (1950-2006) and UAH (1979-2006) to compare with those using the months JJA. For CRU we have using the entire year 0.30 degrees C per decade and very significantly different than 0 trend and using JJA the trend is 0.14 and not significantly different than zero. For UAH the trends were 0.20 for full year and 0.10 when using JJA and neither of these trends are significantly different than zero – due to noise and fewer degrees of freedom.
My point with CRUTem4 is that a trend derived in the polar region and during a time of Arctic polar amplification is going to miss the annual warming trends by using the June/July/August months. I need to go back to the MXD Schweingruber proxy series, that Mann (2008) cutoff 1960-current time in order to avoid the divergence problem there, and see how the location and handling with those proxies differed from those used in Polar Urals in Briffa (2013).
Yeah, if you look into it, there is something really interesting going on with the “amplification”: it’s really a warming concentrated in cold, dry, high pressure air masses.
So there is a distinct difference between winter and <a href="http://i23.photobucket.com/albums/b370/gatemaster99/GISSNHWARM.png"summer.
What’s more, in the US at least I know for sure that the warming is strongest on the coldest days of the year-and by that I don’t mean the 15th of January or something like that. I mean if you rank each day within a year, and then ask what the trends are for 1st coldest, second coldest, and so on, by far the largest trends are in the most extremely cold days. Evidently what’s cold, warms more. Veeeery interesting.
Jeff, I found a great example of what I have been talking about with regards to good high frequency correlations between time series that have very different trends. I used the Polar Urals MXD series (the one you reported on in this thread) from Briffa (2013) for the time period 1920-2006 and the gridded temperatures series CRUTem4 and GISS 250 km for the months June/July/August for the same time period and calculated difference series of CRU-F1*MXD (D1)and GISS-F2*MXD ( D2) with F1 and F2 being factors derived by regressing the temperature series versus the MXD series in order to scale the MXD series to the temperature series. From the difference series I calculated trends and p.values for those trends.
Those trends and p.values for D1 were: 0.37 degrees C per decade and 0.0008, respectively and for D2 were: 0.45 degrees C per decade and 0.00004, respectively. The autocorrelations of these difference series were barely significant and small.
Correlation of MXD versus CRU JJA for the 1920-2006 time period was r = 0.81.
Correlation of MXD versus GISS JJA for the 1920-2006 time period was r = 0.78.
From these calculations and observations it is easy to see the weakness of the MXD proxy for lower frequency responses to temperatures (decadal to centennial trends) and how much better it does for
high frequency response as indicated by annual correlations (r). I have seen these differences referenced in several dendro temperature reconstructions papers, but with little details on how limiting that makes these proxies.
Here are some links to papers and discussions of those papers involving MXD and TRW temperature reconstructions.
CA discussion of the Schweingruber MXD series used in Mann (2008):
http://climateaudit.org/2008/09/22/the-new-rain-in-maine/
Below is a good link to an SI on TRW and MXD and trends and correlation to temperatures and temperatures from Siberia area, Salehard, etc.
Click to access Esper_2010_GCB_sup.pdf
CA take on the paper linked immediately above:
http://climateaudit.org/2009/11/03/esper-et-al-2009-on-west-siberia/
The Esper paper is critical of dendro proxies while at the same time attempting (unsuccessfully in my view) of explaining the divergence problem. Esper is the one in this field who puts forth the issue of using the unadjusted temperatures for calibration in dendro temperature reconstructions much as Phi has advocated. Obviously if one had a divergence problem with which to contend and the adjusted temperatures were adjusted downward some of it would have to go away using unadjusted temperatures. That, however, does not make the unadjusted temperatures somehow more valid than the adjusted ones. Unadjusted temperatures were collected for most of these time periods covered by temperature reconstructions, not for the purpose of use in these studies, but rather for purposes requiring less precision and continuity. That does not make the adjusted temperatures without sources of errors and uncertainties but no doubt better than the unadjusted ones.
I need to correct the below values as I made a dumb mistake in adding JJA values instead of taking averages. The correlations are not effected. The trends in the difference series for the time period 1920-2006 remain significant and substantial but not as a large.
Uncorrected values:
Those trends and p.values for D1 were: 0.37 degrees C per decade and 0.0008, respectively and for D2 were: 0.45 degrees C per decade and 0.00004, respectively. The autocorrelations of these difference series were barely significant and small.
Corrected values:
Those trends and p.values for D1 were: 0.08 degrees C per decade and 0.05, respectively and for D2 were: 0.15 degrees C per decade and 0.00005, respectively. The autocorrelations of these difference series were barely significant and small.
As Jeff notes the MXD series are very flat for most of the time period of the instrumental temperature and particularly for the last 90 years. MXD series do have good high (annual as indicated by r) frequency correlation,, but is noted even by those using MXD in temperature reconstructions as having a poor low frequency response to temperature – as in trends, I have noted that several papers have concentrated on the Yamal and Polar Urals region. That region (un)fortunately shows historical JJA temperatures that also are more or less flat and particularly when considering the latter part of that time period is the modern warming period. There is no steep warming for the MXD series to diverge from in that period.
Phi, since the temperatures from later in the series do not have trends significantly different than zero, we cannot make a good case there for divergence. We can only look at graphical divergence for what that is worth, although when climate scientists attempt to sell the idea of accelerated warming in the past few decades graphical appearances can be important and graphical divergence worth hiding. That is why I wanted to find other MXD reconstructions in other regions of the globe where the JJA temperatures might show the modern warming.
The great interest in these specific MXD data (because they have been completed until 2006) is to be able to compare them seriously to satellite temperatures. There is no divergence in this case. This finding is a strong argument in favor of the hypothesis of a divergence originating from problems with stations data.
There are many other arguments in this direction, I recalls somes :
– Surprising lack of tropospheric amplification in the tropics (hot spot)
– Excellent correlation MXD-glaciers melting,
– Good anticorrelation MXD-snow,
– Bias of the dicontinuities in raw temperatures series.
Adjustments are not the primary cause but I would add that the fact that homgeneizations techniques have been developed in the context of the IPCC problematic is absolutely not an argument in their favor, quite the contrary.
I show substantial divergence from RSS.
Phi, and I think Jeff will agree with me here, that the satellite data and MXD would be for only 1979-2006 JJA and would include only 28 data points. With noisy data it is difficult to show statistically significant trends. From curiosity and what Jeff commented here I might look at RSS JJA and MXD using a difference series as I did above for CRU JJA and GISS JJA for the period 1920-2006. My guess is that I could see substantial trends in the difference series but that are not statistically significant. Phi, the CRU JJA correlation with MXD from 1950-2006 and 1920-2006 is better than the UAH JJA correlation to MXD was for 1979-2006. I saw no divergence for using the CRU or UAH data for their respective time periods, and because, although the temperature trends were greater than the MXD trends, none of the trends were statistically significant.
Jeff Id,
RSS would be an argument against, at least partially. There may be others. Must take into account all that is known and estimate the strength of each. Those that I mentioned are, in my opinion, very strong , they should be taken into account too.
Kenneth Fritsch,
There is a clear divergence CRUTEM-UAH from 1990. MXD follow UAH, not CRUTEM.
MXD follow a flat line through their whole history Phi. The flat line is the lack of variance so common with treemometers (think HS handle). The fact that one temp series also has a flat line doesn’t necessarily validate the other.
Jeff Id,
The lack of variance of MXD is a legend due to calibration on false data. Where is this lack of variance?
I do not see it.
The lack of variance is visible in the long term “reconstruction” of the graph above (Figure 4 from the top). Consider that in the past 2000 years, no density values exceeding recent temperatures and the fact that the mean is so close to the max values yet far from the min. Dr. Loehle makes the point above about the trees appearing growth limited. If we just consider for a moment that there is a point where more temp doesn’t give more growth, it becomes apparent that this plot might be exhibiting that exact behavior. That growth limitation won’t limit the high frequency correlation of the signal because the short term amplitude overwhelms the long term. Just acknowledging that the growth limitation exists should be enough for you to pause in your argument and take a more skeptical look at the “trend” of this curve.
The lack of variance in your graph does not come from MXD but from your methodology. Simple averages do not show this behavior:
“Dr. Loehle makes the point above about the trees appearing growth limited.”
TRW represent a growth index, not MXD. MXD and TRW can move in opposite direction:
However, as for any physiological phenomenon, it is certain that MXD are also bounded. The question of whether these bounds are detectable in the series available is still open.
In the particular case, there is no evidence suggesting that these bounds are reached. The characteristic low point around 1998 is obviously not concerned. It is however consistent with the whole period 1979-2006.
Phi,
Averaging doesn’t affect the HF variance asymmetry I was pointing out. We don’t need to keep on about this though. I will keep messing with the data a bit longer and see what we can find.
And there?
We see many things but certainly not a lack of variance of MXD.
MXD clearly fails to reproduce the magnitude of the peak in the 40’s, and the peak in the beginning of the data. Why should one assume it it would be correctly capturing a recent peak?
You’re right, it could be a upper limit to the sensitivity of MXD. At least for the peak of the 1940s. However, it must be nuanced: the weakness of this period si mainly due to the exceptional value of 1947 and other factors could be invoked for 1860-1870.
What I think is important in this case, it is mainly the fact that MXD (despite their weaknesses) are closely confirmed by glaciers melting. There is no divergence MXD-glaciers. This divergence (compared to stations data) is about 0.15 ° C per decade since the early twentieth century. For the same region, the divergence between winter temperatures and snow is also about 0.15 ° C per decade (http://img69.imageshack.us/img69/3867/jndjfm.png).
I do not conclude from this that MXD are a perfect proxy but that stations temperatures are certainly unreliable.
The trouble with glaciers is similar. They aren’t thermometers and respond to many factors other than temperature.
The variable used by Huss is the melting anomaly; no physical quantity other than temperature is known which could significantly influence it. Huss et al. 2009 try with shortwave but it is not satisfactory, not supported physically and poor correlation.
The solution is actually quite simple in this case. Huss et al. 2009 use the reference station of Davos. Davos temperatures, raw or adjusted taking into consideration Hansen et al. 2001, show a remarkable correlation with the melting anomaly : http://img837.imageshack.us/img837/5687/fontea.jpg.
Phi,
glaciers also “melt” due to lack of snow. Kilimanjaro is an excellent lesson in this. The temps are almost never above freezing but the glacier sublimated faster than the small replacement precip. It is now regrowing due to more precip.
phi said
March 5, 2014 at 4:23 pm
The lack of variance in your graph does not come from MXD but from your methodology. Simple averages do not show this behavior:
That graph could very easily be interpreted as the MXD proxy not being capable of responding to the modern warm period and thus missing those periods in the past – prior to the instrumental period. You have to make an independent case for the temperature record being wrong in that locale and the one you use is correct.
Kenneth Fritsch,
The use of graphics is certainly sensitive and there is a large place for subjectivity. But it is also a very powerful analytical tool.
“That is why I wanted to find other MXD reconstructions in other regions of the globe where the JJA temperatures might show the modern warming.”
I have already presented this graph (http://img38.imageshack.us/img38/1905/atsas.png) for the Alpine area from the Schweingruber database (http://www.ncdc.noaa.gov/paleo/treering-wsl-data.html
). I regret that this database is so little exploited for regional comparisons. It would be great if you or others could be interested to study other regions.
“That graph could very easily be interpreted as the MXD proxy not being capable of responding to the modern warm period and thus missing those periods in the past – prior to the instrumental period.”
I introduced this graph to show that there was probably no big issue with MXD variance in general. Otherwise, I would say it mainly shows the huge selection bias that affects Briffa et al. 2013.
In the link below is a paper on the divergence problem with Rob Wilson as a co-author. I mention Wilson because he is one of the very few dendroclimatologists who have appeared at any of these skeptical blogs. The paper has an underlying theme of the divergence problem being an unprecedented one and human caused which probably has more to do with the leanings of these dendro to maintain the credibility of trees as valid indicators of historical temperatures preceding the instrumental record. Rather obvious is an alternative explanation that could be stated as all these different conjectures put forward to explain divergence could be merely an accounting of all the variables that could affect TRW and MXCD responses and that can overwhelm the temperature signal – at least on the low frequency scale.
The paper points to temperature reconstructions wherein the TRW/MXD responses follow the recent warming (they do not talk about responses that exceed the modern warming but we know those exist also) and where the responses diverge downward from the recent warming. I have noticed that all these observations are further complicated by the use of different months of the year to be used with the TRW and MXD responses. Part of this could be rationalized by different growing seasons for different tree sites, but the months selected range from: MJ, JJ, JA, AMJJAS and AJJ. Obviously if these various month combinations have very different modern day temperature trends then the selection, without independent and science based reasons, can amount to curve fitting.
I can also see that analyses like this one have an underlying faith in trees as thermometers and that the analyses never take the approach that a more skeptical party might. This paper noted that the divergence is greater in the northern latitudes. One could ask whether that situation could be explained by accelerated warming in modern warming period in the northern latitudes versus those further south and thus the divergence is merely more noticeable in the northern latitudes. I doubt very much that a paper with such a skeptical approach would be likely to get published given the current politicized environment.
I am curious enough to want to look at temperature trends at various latitudes for selected months.
Click to access DArrigo_etal.pdf
Jeff, I used 45 individual proxies in the Briffa (2013) polarall mxd file, with mostly complete data from 1900-1990, in order to determine how much these proxies vary when standardized by subtracting the mean and dividing by the standard deviation and then comparing the individual proxy to that of the mean of all the proxies standardized by the same method. I then calculated trends of the difference series scaled to degrees C per decade along with p.values and correlations of the individual standardized proxy series to the standardized proxy mean. I did the correlations 2 ways using the series data and using the residuals from a linear regression versus time. I also calculated the trends and p.values for the standardized individual proxy series and the standardized mean proxy series. These results are reported in the linked table below.
The average correlations for the 45 series using the series and linear regression residuals were 0.75 and 0.78, respectively. Using residuals brings a number of the lower correlations using the series data up closer to the average.
The calculations in the linked table show that while the Briffa (2013) paper discusses the average of several individual proxies, as most temperature reconstruction papers do, you do not see the variation in the individual proxies unless you dig deeper. That variation is much greater than one would expect from measurement errors using thermometer when considering the variation in individual series trends and I am not at all sure that variation is included in the uncertainty reported in the papers. It is trends that are paramount in temperature reconstruction papers. The individual proxies do not exhibit a common trending and, as a result, would provide different pictures of divergence effects. One might want to Monte Carlo the uncertainty of the average series trend by leaving randomly selected number of proxies out of the calculation. The correlations of the individual series, on the other hand, show very good high frequency coherence with the mean proxy series and by inference with each other. It is this correlation that is often reported in reconstruction papers and from what I see in deference to reporting the trend differences amongst the individual proxy series.
I wonder how to translate summer temperatures to yearly temperatures.
See the CRU data between 1925 and 1955 for this site.