What are the global temperature trends according to four different data sets?

[Andre]

Let's compare the four predominant data sets about the global temperature updated to include October 2008:

Shown in thin lines are the monthly averages while the thick lines represent 12 months running averages.

Red and orange are based on surface meteo station data as compiled by NASA (Hansen et al) and the British Met Office (HADCRU of Jones et al). Green and blue are two different products of the same satellite data series compiled by University of Alabama (Spencer et al) and Remote Sensor Systems (RSS).

See how NASA creeps up, whereas Jones et al of the UK met office holds the midgrounds between Hansen and the two satellite temperature sets. Although the latter show differences in monthly values, both have a robust fit of the 12 months running mean (bold black).

How would this compare to the predictions from the past?

sources

http://data.giss.nasa.gov/gistemp/tabledata/GLB.Ts+dSST.txt
http://hadobs.metoffice.com/hadcrut3/diagnostics/global/nh+sh/monthly
ftp://ftp.ssmi.com/msu/monthly_time_series/rss_monthly_msu_amsu_channel_tlt_anomalies_land_and_ocean_v03_2.txt[/URL]
[PLAIN]http://vortex.nsstc.uah.edu/public/msu/t2lt/tltglhmam_5.2 [/

 

[Xnn]

One of the obvious differances is that the University of Alabama series is not of the surface. It is measuring the lower troposphere, a shifting target.

Another observation is that ENSO adds considerable noise to the system and of course the oceans have much more heat capacity.

There is also the National Climate Data Center series. Here is a graph of their data:

http://http://2.bp.blogspot.com/_9LFTVlVyZZ4/SVldp8oz73I/AAAAAAAAADY/YEdnMzaQvEM/s1600-h/Monthly+Global+Temperatures_22735_image001.gif

 

[Ivan Seeking]

Let's compare the four predominant data sets about the global temperature updated to include October 2008:

...How would this compare to the predictions from the past?[/

Are you asking for amateur interpretations of data, which amounts to asking for personal theories, which are not allowed here?

What are the peer-reviewed interpretations of this data?

 

[Evo]

Are you asking for amateur interpretations of data, which amounts to asking for personal theories, which are not allowed here?

What are the peer-reviewed interpretations of this data?

If it's based on the data from the official sources posted, what is wrong with that? Are you saying members aren't allowed to comment on legitimate, official data?

 

[Ivan Seeking]

Most members aren't qualified to comment. The appropriate thing to do is to at least see what the real experts have to say, first. At least that way we know all of the variables that are considered.

It would be do-it-yourself [crackpot] science for amateurs to engage is an analysis. And if someone wants to take a respectable shot at this, we have the IR forum for that.

 

[Evo]

It's not an "analysis" of the data. It's looking at what these "official" sources predicted would happen as opposed to what actually happened. Like "it will rain Thursday" and on Friday you know it didn't rain Thursday.

 

[Monique]

Are you asking for amateur interpretations of data, which amounts to asking for personal theories, which are not allowed here?

What are the peer-reviewed interpretations of this data?

He's asking how the data compares to the predictions from the past. Of course the predictions (as well as the data) need to be from a credible source, otherwise it is not even worth discussing.

 

[Andre]

It's not an "analysis" of the data. It's looking at what these "official" sources predicted would happen as opposed to what actually happened. Like "it will rain Thursday" and on Friday you know it didn't rain Thursday.

That's the spirit of the scientific method, testing the predictions, and that's the intention of this post, to see if it rained on Thursday. No analyses, just comparing predictions with measured results.

This is the prediction that started the global warning alarm http://pubs.giss.nasa.gov/docs/1988/1988_Hansen_etal.pdf , centered around the model result, fig 3 (page 9347)

So what happens if we merge the actual results of the NASA and RSS (12 month running average) with the predictions:

Note that the vertical positions of the graphs depend on different definitions of the basic zero value. Therefore I have displaced the vertical plots of the both measured series to start at the average value between scenario A and B.

Also important is to note the presumptions for the about the three scenarios in appendix B page 9361 - 9362

"In Scenario A, CO2 increases as observe by Keeling for the interval 1958-1981 [Keeling et al 1982] and subsequently with 1.5% yr^-1 growth of the annual increment ...

B: "In Scenario B the growth of the annual increment of CO2 is reduced from 1.5% yr^-1 today to 1% yr^-1 in 1990, 0.5% yr^-1 in 2000 and 0 in 2010...

C: "In scenario C the CO2 growth is the same as in scenario A and B through 1985; between 1985 and 200 the annual CO2 increment is fixed at 1.5 ppmv yr^-1; after 2000 CO2 ceasaes to increase, its abundance remaining fixed at 368 ppmv...

So the next question is, which scenario is closest to reality.

 

[Astronuc]

Perhaps it's still too early to tell. The NASA 12 Mo RA and RSS 12 Mo RA oscillate quite a lot compared to the models, which seem smoother. Occasionally the measurements depart from the models. It's hard to tell A, B, C (but I assume C is the bottom one in the second plot). It would appear the measurements are dropping below C between 2006 and present.

 

[Monique]

How are standard deviations or confidence intervals normally taken into account with these types of models and data?

 

[Astronuc]

How are standard deviations or confidence intervals normally taken into account with these types of models and data?

Perhaps those details are buried in the papers by Hansen et al., e.g. the one cited by Andre in post #8.

I suppose they could use noise analysis. In some cases, I seen 5-year (rolling average) trending plots which smooth out variations. I'm not sure how the measured data are processed.

Hansen/GISS make the following comments:

Current Analysis Method
The current analysis uses surface air temperatures measurements from the following data sets: the unadjusted data of the Global Historical Climatology Network (Peterson and Vose, 1997 and 1998), United States Historical Climatology Network (USHCN) data, and SCAR (Scientific Committee on Antarctic Research) data from Antarctic stations. The basic analysis method is described by Hansen et al. (1999), with several modifications described by Hansen et al. (2001) also included.

Graphs and tables are updated around the 10th of every month using the current GHCN and SCAR files. The new files incorporate reports for the previous month and late reports and corrections for earlier months. NOAA updates the USHCN data at a slower, less regular frequency; we switch to a later version as soon as a new complete year is available.

The GHCN/USHCN/SCAR data are modified in two steps to obtain station data from which our tables, graphs, and maps are constructed. In step 1, if there are multiple records at a given location, these are combined into one record; in step 2, the urban and peri-urban (i.e., other than rural) stations are adjusted so that their long-term trend matches that of the mean of neighboring rural stations. Urban stations without nearby rural stations are dropped.

A global temperature index, as described by Hansen et al. (1996), is obtained by combining the meteorological station measurements with sea surface temperatures based in early years on ship measurements and in recent decades on satellite measurements. Uses of this data should credit the original sources, specifically the British HadISST group (Rayner and others) and the NOAA satellite analysis group (Reynolds, Smith and others). (See references.)

. . . .

Ref: http://data.giss.nasa.gov/gistemp/

And with respect to 2008, GISS offers the following:

The GISS analysis of global surface temperature, documented in the scientific literature [ref. 1], incorporates data from three data bases made available monthly: (1) the Global Historical Climatology Network (GHCN) of the National Climate Data Center [ref. 2], (2) the satellite analysis of global sea surface temperature of Reynolds et al. [ref. 3], and (3) Antarctic records of the Scientific Committee on Antarctic Research (SCAR) [ref. 4].

In the past our procedure has been to run the analysis program upon receipt of all three data sets and make the analysis publicly available immediately. This procedure worked very well from a scientific perspective, with the broad availability of the analysis helping reveal any problems with input data sets. However, because confusion was generated in the media after one of the October 2008 input data sets was found to contain significant flaws (some October station records inadvertently repeated September data in the October data slot), we have instituted a new procedure. The GISS analysis is first made available internally before it is released publicly. If any suspect data are detected, they will be reported back to the data providers for resolution. This process may introduce significant delays. We apologize for any inconvenience due to this delay, but it should reduce the likelihood of instances of future confusion and misinformation.

Finally, we note that we provide the rank of global temperature for individual years because there is a high demand for it from journalists and the public. The rank has scientific significance in some cases, e.g., when a new record is established. However, otherwise rank has limited value and can be misleading. Note that, given our estimated error bar in Figure 1, we can only say that 2008 probably ranks as somewhere between the 7th and 12th warmest year. As opposed to the rank, Figure 3 provides much more information about how the 2008 temperature compares with previous years, and why it was a bit cooler (note the change in the Pacific Ocean region).
. . . .
References
1. Hansen, J., R. Ruedy, J. Glascoe, and Mki. Sato, 1999: GISS analysis of surface temperature change. J. Geophys. Res., 104, 30997-31022, doi:10.1029/1999JD900835.

2. Peterson, T.C., and R.S. Vose, 1997: An overview of the Global Historical Climatology Network temperature database. Bull. Amer. Meteorol. Soc. 78, 2837-2849.

3. Reynolds, R.W., and T.M. Smith, 1994: Improved global sea surface temperature analyses. J. Climate 7, 929-948.

4. Scientific Committee on Antarctic Research (SCAR), www.scar.org.

A press release from Met Office Hadley Centre and the Climatic Research Unit (CRU) at University of East Anglia
http://www.metoffice.gov.uk/corporate/pressoffice/2008/pr20081216.html

. . .
La Niña events typically coincide with cooler global temperatures, and 2008 is slightly cooler than the norm under current climate conditions. Professor Phil Jones at the CRU said: "The most important component of year-to-year variability in global average temperatures is the phase and amplitude of equatorial sea-surface temperatures in the Pacific that lead to La Niña and El Niño events".

The ten warmest years on record have occurred since 1997. Global temperatures for 2000-2008 now stand almost 0.2 °C warmer than the average for the decade 1990–1999.

Dr Peter Stott of the Met Office says our actions are making the difference: "Human influence, particularly emission of greenhouse gases, has greatly increased the chance of having such warm years. Comparing observations with the expected response to man-made and natural drivers of climate change it is shown that global temperature is now over 0.7 °C warmer than if humans were not altering the climate."
. . . .

Plot of temp anomaly by rank - http://www.metoffice.gov.uk/corporate/pressoffice/2008/images/latest_rankings_jan_to_nov.gif

2008 ranks as tenth warmest in the set for the period considered.

Variability (transient processes) on the order of ~.2-0.4°C seems to be expected.


Hadley Center paper on uncertainties - http://hadobs.metoffice.com/hadcrut3/HadCRUT3_accepted.pdf


MSU/AMSU atmospheric temperature products.
http://www.remss.com/data/msu/Changes_from_Version%202_1_to_3_0.pdf (pdf - use 'save target as')

and

Construction of the Remote Sensing Systems V3.2 atmospheric temperature records from the MSU and AMSU microwave sounders.
http://www.ssmi.com/data/msu/support/Mears_and_Wentz_TMT_TTS_TLS_submitted.pdf

 

[Xnn]

Hansen papers assumes in scenario A and B that CO2 emissions will grow 1.5% annually. Actual atmospheric CO2 concentrations have been growing only about 0.4% annually since 1980 for a total of 13.6%. Hansen also assumes a climate sensitivity of 4.2C per doubling of CO2.

Since 1980, according to the NCDC data base, global 5 year average temperatures have risen about 0.51C, whereas NASA GISS finds 0.53C. Considering the actual increases in CO2, Hansen’s sensitivity assumption works out to 0.57C of warming. So, actual climate sensitivity to CO2 doubling looks to be less than 3.8C since CH4 has also played a role (at least up to the 1990's).

Hansen should have assigned an uncertainty band for his CO2 sensitivity or anticipated solar irradiance falling as much as it has. On the otherhand, perhaps he seriously underestimated climate sensitivity to CH4. Go figure.

 

[Gokul43201]

Why am I not surprised? Another thread that seems to be based on misinformation and fallacious arguments.

Let's compare the four predominant data sets about the global temperature updated to include October 2008:

...

See how NASA creeps up, whereas Jones et al of the UK met office holds the midgrounds between Hansen and the two satellite temperature sets.

Umm... NASA (GISTEMP) does NOT "creep up", nor does the UK Met (HADCRUT3) hold a middle ground between GISTEMP and the LT dataset.

In fact, if you take the littlest trouble of adjusting for the baselines, one will find that 3 of the 4 datasets match fairly closely. The outlier is the UAH set, (not GISTEMP, as anyone reading any number of the threads in this forum - including this one - would have come to believe).

Here's what you'd get for the means and trends of the 4 datasets (from a linear least squares fit to 12-month running averages from last 30 years of data) after correcting for the baselines by using the 1979-1998 mean values (which RSS and UAH use) for all four sets:

GISTEMP: mean anomaly = 0.082C, trend = 0.16C/dec
HADCRUT3: mean anomaly = 0.080C, trend = 0.16C/dec
RSS: mean anomaly = 0.081C, trend = 0.16C/dec
UAH: mean anomaly = 0.065C, trend = 0.13C/dec

If anything, it looks like UAH is the outlier, not GISTEMP (NASA)!

Please check these numbers for yourself and let us know if you get something different.

And oh yes...

Although the latter show differences in monthly values, both have a robust fit of the 12 months running mean (bold black).

That's just flat out ridiculous!

UAH and RSS do not share the same indistinguishable 12-month running average. In 1980, for instance, the difference in the 12-month average is nearly 0.1C (after matching almost exactly in 1979). Whoever made the plot in the OP will need to make that "bold black" line about 20 times thicker in order to pull off the story that UAH and RSS share the same running mean for every month of the last 360 months!

As I've said above, they do not even posses closely matching trends.

 

[Andre]

Indeed I made a mistake, using the same UAH data twice for the 12 month running average in the OP giving a RA slightly above UAH. I should have more cautious because it's meaningless and has nothing to do with misinformation. UAH is indeed the outlier with the lowest trend, however NASA is the only of the four not having 1998 as the warmest year which results in an optical outlier.

The second graph in my last post does not suffer from that error because it has been generated differently, calculating the running average from the correct data manually

for scrutiny:
http://rapidshare.com/files/178377031/the_big_tempfile-3.xls

 

[Gokul43201]

Well Evo, I guess this answers your question: "If it's based on the data from the official sources posted, what is wrong with that?" The plot in the OP was NOT a true representation of the data from the official sources! I guess that leaves Ivan's question unanswered.

UAH is indeed the outlier with the lowest trend, however NASA is the only of the four not having 1998 as the warmest year which results in an optical outlier.

UAH is the only of the four that has 1980 as the warmest year in the 1979-1987 period. Does that make UAH an optical outlier?

What is an "optical outlier"?

Is a judgment of the quality of a dataset based on one single "optically" chosen year of any real scientific value?

Can we please stop scapegoating NASA (or anyone else, for that matter) with hand-waving "optical" arguments?

 

[Evo]

Well Evo, I guess this answers your question: "If it's based on the data from the official sources posted, what is wrong with that?" The plot in the OP was NOT a true representation of the data from the official sources! I guess that leaves Ivan's question unanswered.

But the links to the official sources are valid, so what I said stands.

 

[Gokul43201]

But the links to the official sources are valid, so what I said stands.

Then can we please delete all the plots that are not copied from published papers (such as the one in the OP, which we know is wrong), as well as any description of such unverified plots?

 

[Evo]

Then can we please delete all the plots that are not copied from published papers (such as the one in the OP, which we know is wrong), as well as any description of such unverified plots?

Yes, that's a good idea.

 

[mheslep]

Hansen papers assumes in scenario A and B that CO2 emissions will grow 1.5% annually. Actual atmospheric CO2 concentrations have been growing only about 0.4% annually since 1980 for a total of 13.6%.

Note that while that may be be correct for concentration, CO2 emissions did increase as Hansen assumed and more: http://http//www.mnp.nl/en/dossiers/Climatechange/TrendGHGemissions1990-2004.html" :

...Apparently the rate of uptake by CO2 sinks, either the ocean,
or, more likely, forests and soils, has increased.

 

[aaroman]

about the global temperature

There is no such thing. The temperature is defined for a system in (cvsi)equilibrium.
That's it. The "global temperature" the statisticians talk about is something else than a "temperature" (that is, a system parameter, a physical property, blablabla). It's not something that has a physical meaning. It should be named bull...rature, to avoid equivocation.

 

[Skyhunter]

The MSU data is contaminated by stratospheric influence on channel 2.
"[URL
Here[/URL] is the University of Washington's reanalysis of the data where they eliminate the stratospheric influence on the upper troposphere.

From 1979 to 2001, temperatures observed globally by the midtropospheric channel of the satellite-borne Microwave Sounding Unit (MSU channel 2), as well as the inferred temperatures in the lower troposphere, show only small warming trends of less than 0.1K per decade (refs 1–3). Surface temperatures based on in situ observations however, exhibit a larger warming of ,0.17K per decade (refs 4, 5), and global climate models forced by combined anthropogenic and natural factors project an increase in tropospheric temperatures that is somewhat larger than the surface temperature increase6–8. Here we show that trends in MSU channel 2 temperatures are weak because the instrument partly records stratospheric temperatures whose large cooling trend9 offsets the contributions of tropospheric warming. We quantify the stratospheric contribution to MSU channel 2 temperatures using MSU channel 4, which records only stratospheric temperatures. The resulting trend of reconstructed tropospheric temperatures from satellite data is physically consistent with the observed surface temperature trend. For the tropics, the tropospheric warming is ,1.6 times the surface warming, as expected for a moist adiabatic lapse rate.

 

[vanesch]

The problem I see with the paper in nature posted by skyhunter, is that one uses a subtraction technique which is a black box fit on balloon data, it is not an instrument analysis.

If I may play the devil's advocate:
We have a tropospheric heating, represented by T1, and we have a stratospheric cooling, represented by T2. We are of the opinion that T1 doesn't rise "enough" and hence define a corrected T1, which goes as T1' = T1 - alpha T2 where alpha is the "black box correction". In doing so, one can of course obtain just any trend, by picking the right alpha: in other words, you've now defined a variable (T1') which you can give any slope you want.
You then fit this (by fitting alpha with a least squares estimator) on other data (the balloon data), and lo and behold, you get out the right slope. But what we've done is now to *fit* the slope of the balloon data. We've lost the independence of the satellite measurement in doing so. By forcing those measurements onto a calibration with other data, we've correlated these measurements with the former data.

If one were to analyse the physics of the measurement, and derive a correction based upon that, we would still have independent data. But by the above procedure, we've lost that.

Or not ?

 

[Skyhunter]

The problem I see with the paper in nature posted by skyhunter, is that one uses a subtraction technique which is a black box fit on balloon data, it is not an instrument analysis.

If I may play the devil's advocate:
We have a tropospheric heating, represented by T1, and we have a stratospheric cooling, represented by T2. We are of the opinion that T1 doesn't rise "enough" and hence define a corrected T1, which goes as T1' = T1 - alpha T2 where alpha is the "black box correction". In doing so, one can of course obtain just any trend, by picking the right alpha: in other words, you've now defined a variable (T1') which you can give any slope you want.
You then fit this (by fitting alpha with a least squares estimator) on other data (the balloon data), and lo and behold, you get out the right slope. But what we've done is now to *fit* the slope of the balloon data. We've lost the independence of the satellite measurement in doing so. By forcing those measurements onto a calibration with other data, we've correlated these measurements with the former data.

If one were to analyse the physics of the measurement, and derive a correction based upon that, we would still have independent data. But by the above procedure, we've lost that.

Or not ?

I am not sure what you mean by "analyze the physics of the measurement." Channel 2 has been analyzed and found to be contaminated by stratospheric cooling. Maybe I am missing something, but it seems to me that if we know there is a bias, the value of that bias can be determined since we have the radiosonde and Channel 4 data.

If their statistical method was flawed, Ross McKitterick would be all over it.

 

[vanesch]

I am not sure what you mean by "analyze the physics of the measurement." Channel 2 has been analyzed and found to be contaminated by stratospheric cooling. Maybe I am missing something, but it seems to me that if we know there is a bias, the value of that bias can be determined since we have the radiosonde and Channel 4 data.

I'm not claiming an error. What I mean is that I thought that the "sensitivity functions" over the depth over the atmosphere for the different channels were known (each physical channel has an attached weighting function as a function of atmospherical depth). As these functions are not square block functions, it is obvious that each channel has influences from different layers, so there is a "mixing matrix" which mixes the influences of the different layers into each physical measurement channel. In order to try to extract the layer temperatures from this, one should then of course apply the inverse matrix of this mixing matrix. I thought that was already done.

But what's done in the paper is to replace this inverse matrix with free parameters, which are then fitted to a set of calibration data. There's nothing intrinsically wrong with this, except that it is disappointing that one had to resort to this, and couldn't determine the original mixing matrix "from first principles". In doing so (and again, it is not wrong to do so), one looses an independent check, because one has now, through this calibration, correlated these measurements with former measurements. If we go and look for small trends, then in doing so, we have foregone the possibility of an independent check, because of this.
If by any chance, the trend in the calibration data was wrong, we will now "recover" artificially this same trend (because we forced this by using the free parameters in the fit).

 

[Skyhunter]

Thank you for the explanation.

The satellites were designed to monitor weather, not the temperature of the atmosphere. Which actually they do not measure. RSS and UAH both us the same data with different methodologies arriving at different results. My guess is that this is about as good as it gets until an instrument is deployed that can detect and distinguish between troposphere and tropopause.

 

[Gokul43201]

We have a tropospheric heating, represented by T1, and we have a stratospheric cooling, represented by T2.

Here it looks like you are using T1 and T2 to represent trends (something like <dT/dt>) rather than actual temperatures.

We are of the opinion that T1 doesn't rise "enough" and hence define a corrected T1, which goes as T1' = T1 - alpha T2 where alpha is the "black box correction".

Here, I'm not sure, but it looks like you are saying that T1 is actually a temperature, rather than a slope.

In doing so, one can of course obtain just any trend, by picking the right alpha: in other words, you've now defined a variable (T1') which you can give any slope you want.

In the paper, alpha is determined by a least squares fit to the actual temperatures, not by a least squares fit to the overall trends. There is a big difference.

When you fit to the temperatures, the size of the error in the fit of alpha is important. Even if the error is large, the fit will generate the right slope but it's nevertheless meaningless, for the reason you've given above. However, T1 and T2 are not single temperatures (or single trends), but are in fact, large ordered sets of several points. To exactly generate a third ordered set, T'(balloon), from the sets T1 and T2, you actually need an ordered set alpha = {alpha_i | i in 1 to N}, such that T(balloon)_i = T1_i +(alpha_i*T2_i), for all i in 1 to N. That you can closely match T(balloon), and not just the slope of T(balloon), with the ordered set T' given by T'_i = T1_i +(alpha_0*T2_i) is what makes the set T' an independent substitute for the T(balloon) set.

In other words, you can always find an alpha_0 which will make the trend in T' match the trend in T(balloon) to any arbitrary degree of precision, but there is no a priori reason that you should be able to find an alpha_0 that will make the sets T(balloon) and T' have an arbitrarily small RMS variation.

i.e., you can always make $\sum_i (T(b)_i - T'_i) = 0$, but you can't necessarily make $(1/N)\sum_i (T(b)_i^2 - T'_i^2)$ arbitrarily small, by the choice of a single alpha_0. That the authors were able to find that the alpha_0 which closely matches the two trends also keeps this RMS error small is what I think is significant.

 

[Skyhunter]

UAH is definitely the outlier here.

The higher GISS average is due to the extrapolation of Arctic temperatures from satelllite data as opposed to the HadCRUT3 method of trending the Arctic to zero in grids with no surface station data.