I don’t have a lot of time today, but I found this interesting. Commenter “Basil” has offered this for discussion. So I’m putting this up without comment on my part. See also the decadal trends table below. Have at it folks.
Click for full sized image
1979:01-1992:12
---------------------------------------------
GISS 0.000783764** (0.094C/decade)
HadCRUT 0.000460122** (0.055C/decade)
RSS_MSU 0.000498964 (0.060C/decade)
UAH_MSU 1.71035E-05 (0.002C/decade)
1993:01-2001:12
---------------------------------------------
GISS 0.00174741** (0.210C/decade)
HadCRUT 0.00147990** (0.178C/decade)
RSS_MSU 0.00221135** (0.265C/decade)
UAH_MSU 0.00217023** (0.260C/decade)
2002:01-2008:1
---------------------------------------------
GISS -0.00091450 (-0.110C/decade)
HadCRUT -0.00270338** (-0.324C/decade)
RSS_MSU -0.00208111 (-0.250C/decade)
UAH_MSU -0.00130882 (-0.157C/decade)
Well, I will take a swing at it. This chart highlights a problem I had in comparing the UAH satellite data set to the GISS. The GISS uses an average of 1950-1979 (or something like that) to set the “zero” anomaly. I am not sure what UAH uses, because I cannot find it documented. However, to compare the two data sets, we really should have their zeros set to the same basis for comparison.
One way is to just set the 1979’s equal to each other. In effect, that uses a one year period to set the anomaly. when that is done, we get a lot of “divergence” between the two sets. Or, one could use averages of longer periods of time. I found that the longer period of time one used to set this average, the less difference there was between the two data sets. This explains why. The data sets are very similar, except for the first several years after 1979. I am not sure why. But this is why a good argument can be made that there is less divergence than often portrayed between the data sets.
For years I have said that the satellite approach is a better one, and I still support that. It has full coverage and is not subject to location biases. I think a lot of the debate in the past has divided the catastrophists and the skeptics over this point has been colored by the fact that each had a data set that they liked the answer from (ie warmer or cooler). This chart points to a conclusion I have reached – the differences may be overstated, so we maybe can finally focus on choosing the method that makes the most sense, not the method that gives the answer we want.
How much is the final section of the chart distorted by the January lows?
Seems like there is something wrong if RSS and UAH are so far apart from each other.
Hi Bishop Hill
I made 2 excel diagram from RSS data a while ago where you can see how the trend is with and without january added.
It is monthly unsmoothed values.
jan 2002-dec 2007 can be found here:
http://farm3.static.flickr.com/2194/2240553280_32a3424107_o.jpg
jan 2002-jan 2008 can be found here:
http://farm3.static.flickr.com/2048/2242920570_156a4125ef_o.jpg
As you can see a single really low month has some impact on the trend. I estimate it increases the downward trend with ~0.03C during those 6 years.
A couple of quick responses, and then I’ll have a bit more shortly.
Coyote, I’m not quite sure I follow (but it may be that I’m just reading too quickly). What I’ve done (and some others since Anthony got us going on this) was to normalize or standardize all four metrics on their mean for the entire period we are looking at (1979:01 to 2008:01). As to which approach is better, I think both have their pros and cons. Right now, I’m working on a composite trend. You can see the result here:
http://i27.tinypic.com/2r74myd.jpg
This shows all four series, with a composite trend. I’ll post up the numbers here, in the format that I used earlier for the individual series, in a little bit.
Bishop Hill,
There is no “distortion” in the final section of the chart. If you look at the new chart, which I just linked above, which is the composite trend, you can see how the January low dips far below the trend line.
More soon.
Bishop Hill,
Have a look at the detail figure Anthony posted in the last article
http://wattsupwiththat.files.wordpress.com/2008/02/giss-had-uah-rss_global_anomaly_zoomed_1979-2008.png
The decline started around September or October of last year. It isn’t just a single data point influencing those trends, but sustained cooling over four months.
Here’s an update, that goes along with the chart linked to in my previous post:
1979:01-1992:12
---------------------------------------------
GISS 0.000783764** (0.094C/decade)
HadCRUT 0.000460122** (0.055C/decade)
RSS_MSU 0.000498964 (0.060C/decade)
UAH_MSU 1.71035E-05 (0.002C/decade)
Composite 0.000508623*** (0.061C/decade)
1993:01-2001:12
---------------------------------------------
GISS 0.00174741** (0.210C/decade)
HadCRUT 0.00147990** (0.178C/decade)
RSS_MSU 0.00221135** (0.265C/decade)
UAH_MSU 0.00217023** (0.260C/decade)
Composite 0.00179077*** (0.215C/decade)
2002:01-2008:1
---------------------------------------------
GISS -0.00091450 (-0.110C/decade)
HadCRUT -0.00270338** (-0.324C/decade)
RSS_MSU -0.00208111 (-0.250C/decade)
UAH_MSU -0.00130882 (-0.157C/decade)
Composite -0.00183429** (-0.220C/decade)
All I’ve done here, compared to what Anthony copied into the blog posting, is to add the “Composite” lines. These are weighted averages of the 4 metrics, with weights inversely proportional to the variance of the estimates. Generally, the variance is higher for the satellite series than for the sea-land series, so this has the effect of giving the sea-land series greater weight. I’m not opining at all on which do the better job of measuring “global temperature.” But this is a defensible way of using all four, for what they are worth, in coming up with a single, “combined,” metric.
I’ve been looking at the data this way for some time, as opposited to the simple singular trend line often presented in charts like this one:
http://www.remss.com/data/msu/graphics/plots/sc_Rss_compare_TS_channel_tlt.png
which shows a 0.181K/decade trend. It is clear to me that the record is too complex for a single number like this to be meaningful. Looking at the composite numbers presented above, the trend was more on the order of 0.061C/decade from 1979 through 1992, 0.215C/decade from 1993 through 2001, and a -0.220C/decade since 2001. One of the advantages of having multiple metrics like this is that even though one or more of them might not be “signficantly different than zero,” when we combine them we can obtain much greater statistical significance and conclude that there has indeed been a strong downward trend in global temperatures since 2001 at a rate that is, perhaps coincidentally, almost the exact opposite of the trend in the preceding period.
As for whether a trend is “significantly different than zero,” given the variability in the data, that’s not necessarily a very meaningful way of looking at the data. I’m not finished with my calculations, but I’m reasonably certain that none of the individual data series can be used to falsify the composite trend. On the other hand, we can probably reject soundly the notion that the current trend is, say, the 0.181K/decade figure shown on the RSS web site.
More later.
I decided to average the 4 series, since they seem well correlated, and I have little interest in the arguments about which are best. I drew my own trend lines. Clearly the selection of end dates makes a big difference to the trend line:
79-91 +0.1/decade
91-96 +0.2/decade
96-01 +0.13/decade
01-07 -0
07-08 – too short to trend
Most interesting to me was that my 79-91 trend line looks close to picking up where the 07-08 trend might leave us. Some warming forcing 91-01, decaying over the last few years, and temp is now back on the longer term trend. Re-plotting with this long term trend removed just emphasises that without any specific forcings to relate the trends to this exersise is probably pointless.
I see no relation to sunspot cycles…
Basil,
You cannot average together satellite and land measurements because the measure different things. According to Dr Christy (see my post in the thread with the historgrams) you must divide the satellite measurements by 1.2 before comparing them with the land measurements. This will significantly reduce the trends you reported.
This 1.2 factor comes from the fact that tropospheric temperatures must warm faster than surface temperatures if one assumes that the rise is due to GHGs.
Have a safe trip Rev. Knock ’em dead.
Raven: That was upper tropospher.
GISS tries to add in artic and antartic the others are limited in their coverage no. and so.
Here is Dr Christy’s comment:
“Now, I have one misrepresentation to point out on Steve M.’s charts. The temperature comparisons shown are not apples to apples. All climate models indicate the global tropospheric temperature should warm at a rate of 1.2 times that of the surface (1.4 times that of the surface for the tropics – see CCSP SAP 1.1. or Douglass et al. 2007). So, to put surface temperature projections from models on a chart with observed tropospheric temperatures, one must reduce the tropospheric temperature trend by a factor of 1.2 for the comparison to be legitimate. I think the result would be of interest to the readers, and it is entirely defensible as shown in numerous publications”
The satellite data is lower troposphere – not surface so it is not comparible directly to surface measurements.
The fact that GISS includes extremely unreliable and sparse data from the poles is one of the reasons why it is not the best dataset to use for analysis. That said, the poles only make up small percentage of the earth’s area and one of the poles is cooling so I don’t see how the pole data can explain the differences.
The GISS uses an average of 1950-1979 (or something like that) to set the “zero” anomaly.
14°C on the nose for GISS.
All climate models indicate the global tropospheric temperature should warm at a rate of 1.2 times that of the surface
Whoah. Wait I’ve got this all wrong.
It’s not the OFFSET tha’s 1.2 x surface. It’s the dang TREND. The RATE. Oh, hell. That makes it complicated!
I must assume “trend” applies to both warming and cooling. So adapting straight sat. data to surface is a TOTALLY apples/oranges conversion. Ugh! (I just hate it when multiplication drinks Mr. Hyde juice and turns into calculus.)
Raven,
Fair point about satellite and land-sea temperatures not being measures of the same thing. However, I don’t think we have to adjust the data by 1.2x. Actually, isn’t this roughly what is shown for the period when warming was the greatest, 1993 through 2001? Look at the trends I posted. The average for the two land-sea sets is 0.194C/decade, and for the two satellite series is 0.263C/decade. That works out to the troposphere warming at 1.36x the rate of the surface.
When I have time — not before next week — I’ll break the composite into two, one each for the two land-sea series, and the satellite series.
BTW, is the 1.2x from the CGM’s the result of CO2 forcing, or does it hold for any source of warming at the surface, and how it plays out in the troposphere? And what about cooling, like we’ve had since 2001? Does the troposphere cool at 1.2x the rate of surface cooling?
Thanks for the observation.
Basil,
I believe the adjustment is only necessary if one wishes to use the data assert that GHGs caused the warming. In that case, the models tell us that the satellite temperatures trends must be higher than the surface trends.
If something other than GHGs cause the warming/cooling (i.e. the sun) then a different correction would need to be applied. I am speculating on the last point. Dr. Christy did not elborate on his comment on CA.
However, I don’t think we have to adjust the data by 1.2x.
No, no, it’s worse, Basil, it’s so much worse if I am reading this correctly. I think Raven is right:
Christy isn’t saying tha DATA is adjusted (divided) by 1.2 That would be easy.
He says the TREND of the data has to be adjusted by that factor. The slope.
I wouldn’t even know how to begin to adjust for that! (My usual method is to stand well back and radio in for brain support.)
But if sat. trends are lower than surface trends, Houston, we got a problem. And except for the 1993-2001 slice which has a lot of up-down action, we’re seeing a lot of that.
Evan Jones,
I understand that it is the trend that would have to be adjusted by 1.2x. That’s why I called attention to the 1.36x trend difference for the period of greatest warming, 1993 through 2001. It is already there, no need to adjust the data. Since 2001, all the metrics show cooling, but the numbers are all over the place, and it doesn’t look like a similar simple comparison between land-sea and satellite metrics holds. This could be because the period is shorter and the data otherwise too uncertain or variable to yield any precise inferences. Do note that in the 1993-2001 period, all the trends are statistically significant, while that’s not the case for the period since 2001.
I’m not convinced that the 1.2x difference that should be there is going to be discernible. There is too much variation as it is, with the effects of ocean circulations, La Nina’s and El Nino’s, and such. It would probably take a century of steady warming to see a 1.2x difference that is statistically discernible. So far, we’ve only 28 years of satellite data. Let’s look at this again in 2079. 😉
What I really think is useful about all this analysis is in showing that there isn’t a discernible difference. I.e., the data are too noisy to meaningfully test a significant AGW/GCM hypothesis. And hypotheses that cannot be tested are not scientific. To make my point clearer, are their any GCM models which can explain why temperatures in the 1993-2001 period increased at more than 3x the rate of increase from 1979 through 1992, and then simultaneously explain a 3x cooling since then? Oh, and don’t anyone blame the 1993-2001 warming on the 1998 El Nino. I haven’t reported the statistics on my ’98 El Nino variables, but the trend for 1993-2001 excludes the effect of the ’98 El Nino. You can see that in the trend lines, though.
Basil, Your composite trend shows a DECLINE in temperatures at the rate of 0.22 C per decade over the latest six-year period, whereas the IPCC’s 2007 report said that “Six additional years of observations since the TAR … show that temperatures ARE CONTINUING TO WARM near the surface of the planet” (WGI, Chapter 9, p. 683, EMPHASIS added).
In my opinion, your calculations contradict the latter statement. It is true that the noisiness of the data is such that the downward trend in observed temperature between Jan 2002 and Jan 2008 may prove to be a blip in a long-term upward trend – but that remains to be seen.
Trends during recent 11-14year solar cycles?
Ian,
I think the IPCC in AR4 only goes through 2005, so the “last six years” they would be talking about are not the last six years shown in my chart. And in 2005, it might have been harder to discern a downward trend from 2001 than it is today. However, I would venture the possibility that had they looked at the data for 1999 to 2005 carefully, they would have been forced to acknowledge that there was no trend, i.e. that any “trend” for those years was not significantly different than zero. You can almost see this just looking at the other chart I posted:
http://i27.tinypic.com/2r74myd.jpg
From 2001 to 2005, it is certainly hard to visualize any discernible trend, and given the monthly variation in the data it would be highly unlikely to find a “statistically significant” trend of any kind in the data.
While three of the four trend variables since 2001 are significantly different than zero, I wouldn’t be surprised to find them significantly different than the trend value for the preceding period. When I have time this week, I’ll check it out.
Basil
you say
“To make my point clearer, are their any GCM models which can explain why temperatures in the 1993-2001 period increased at more than 3x the rate of increase from 1979 through 1992….”
Pinatubo probably explains the difference. Pinatubo erupted in June 1991 and caused considerable world-wide cooling in 1992, 1992 and possibly 1994 also. This will have the effect of exaggerating the 1993-2001 warming trend while reducing the 1979-1992 trend. Over short time periods the effect of major volcanos is significant.
I’m not sure if you can remove the Pinatubo effect but, as a rough guide, I think both 1991 and 1992 should (without Pinatubo) have been at least as warm as 1990 since there was a strong-ish El Nino in effect during 1991/92. There were also generally warmer ocean temps in the following years up to around late 1995, so 1993 and 1994 should also have recorded similar – if perhaps slightly lower – temperature anomalies.
You also may want to see what things look like if you take the spike of 1998 out. Makes sense to do that anyway since the quick heating then cooling don’t fit the GHG scenario of slowly rising lows.
John Finn,
I can look into how to factor in (or out) the Pinatubo eruption. Now that you have brought it to my attention, it probably dampens the growth rate for my first period, and amplifies it for my second period. So if I somehow control for it, and take it out, it will tend to increase the trend in the first period, and reduce the trend in the second. Given the timing of Pinatubo, if I do try to control for it, I’ll probably end up adjusting the time periods a bit.
Aaron, I’ve already controlled for the 98 El Nino. The trend you see before and after the data spike up in 98 is what is left after controlling for the El Nino.
Ahh, I seen now, sorry. Very interesting.
Ahh yes, now I see. Sorry. Very interesting.
[…] Hat tip to Basil at Watts Up With That for mentioning the technique by name so I could look it […]