the Air Vent

Because the world needs another opinion

Bifurcated Temperature Trend

Posted by Jeff Id on January 23, 2009

I have been studying the satellite vs ground temperature measurements lately trying to understand why they are so different. There is an unresolved dichotomy in the data. Thirty year longer term trend in the RSS and UAH data is less than GISS data while the short term 2 – 5 year temperature variation in RSS and UAH is between 10 and 25 percent greater than GISS. What makes this worse is that the models actually predict that RSS and UAH should have a 20 percent greater variation than ground measurement on both the short term and the longer term scales. This model predicted difference occurs becasue the measurements cover different sections of the troposphere. Satellite data measures a thick layer of lower atmosphere whereas ground measurments reflect temperatures immediately next to the surface.

The three datasets are plotted below. The GISS anomaly graph has been offset to lay on top of the others for easier comparison.


The slopes of the data over the last 30 years are:

GISS 30yr = 0.183 C/Decade

RSS 30yr = 0.157 C/Decade

UAH 30yr = 0.127 C/Decade

After I did a homogenization analysis using ground data I contacted Dr. John Christy and had a short email conversation about my results. I am now convinced the best long term trend for sat data is the UAH trend. Not that it is perfect, it is just that after we remove a step in the data at 1992 related to satellite changeover the two series match almost perfectly. This places sat trend about 0.127 C/Decade.

Satellite Temp. Homoginization Using GISS

Well from the graph above it is difficult to make out the difference in either short or long term trends. To improve the visibility of the difference I removed the trend from the three graphs and overlaid them below.


This plot shows GISS overlaid on UAH and RSS with a two year filter. If you look close you can see the GISS black line is slightly closer to zero at each peak of the dataset. The standard deviation confirms the short term variance effect.

For disclosure purposes, the higher frequency of noise level in GISS data creates a more matched SD value. Unfiltered data is a poor comparison when the frequency of the datasets doesn’t match. The gaussian filter removes the highest frequency element so we can see 1.5 year and longer variance more clearly.

GISS SD = 0.107

RSS SD = 0.128

UAH SD = 0.132

SD RSS/GISS = 1.20

SD UAH/GISS = 1.23

RSS and UAH both have higher variation around the trend than GISS alone by a factor of 1.2. This is caused because satellite profiles are controlled by the microwave emissions from the atmosphere resulting in complex pressure/altitude sensitivity of the temperature measurement. A plot of the various profiles and their sensitivity is below, T’LT is what this post deals with.


If the SD isn’t convincing enough for you, I did a covariance plot of the two year filtered data.


The slopes of the covariance have a lesser ratio than the standard deviation above. For GISS it is 1.14 and for RSS it was 1.08. From my post at the link above, the primary difference between RSS and UAH is the single step at 1992 so I trust the UAH variance more. Either way with a covariance slope greater than 1 and a 30 year slope ratio substantially less than 1. Ratio (.127 C/Decade UAH) / (0.183 C/decade GISS) = 0.694, Something is amiss. Yes there is an expected high degree of serial correlation in the filtered data and I didn’t present an r value simply because I don’t see any use for them here.

From a discussion on climate audit, Dr. Christy recommended that when comparing GISS to satellite data the models demonstrate a 1.2 times multiplier for surface temps to satellite lower troposphere temps . After several discussions with Dr. Christy looking for an explanation, he gave me permission to use this paragraph.

The global-mean short term tropospheric amplification factor of 1.2 (it’s 1.3 in the tropics) indicates (a) that the ocean’s thermal inertia (sfc datasets use SSTs) works against large shorter-term changes while the atmosphere is much less massive and can respond to a greater extent and (b) there is a lapse-rate feedback process where the lapse rate tends to move toward the moist adiabat when thermally forced from below. Why we don’t see this amplification factor in the trend metric (which models show also occurs for the trend) likely deals with the feedbacks of the climate system – there appear to be negative feedbacks on longer time scales that models don’t capture. This is a hypothesis we want to test.

John C.

Ok, so if your like me you need to read this a few times. It has interesting implications for sure. Models agree with the 1.2 times more variance for the satellite data as shown in the SD ratios. Where the models don’t agree is that they also predict this 1.2 times trend in the long term data for the TLT channel. The conclusions about the potential missing feedback in the models relate to a more serious issue.

Six months ago when I started the Air Vent, it was an entertainment project where I was going to honestly explore climate science, read a bunch of papers, make a conclusion and get on with life. Well you can’t always get what you want, now I have an even bigger problem. There are three possibilities as I see it and all, two or only one of them may be correct. I beleive the former is the case but here’s what we have.

1. GISS trend is exaggerated from reality. Corrections don’t remove enough UHI effect, especially in foreign countries so the slope is too high.

2. Models are missing a negative feedback or several which would mitigate the long term trend but not the shorter term.

3. Satellite data long term trend is reduced from actual and may actually be 1.2 times GISS or 1.83*1.2 = .22 Deg C/Decade.

I always say, I am a skeptic not a denier but one thing I am a ‘denier’ on is that we can project temperature out a hundred years with any degree of accuracy. Santer’s latest work inadvertently demonstrated that we can’t even do it on a 30 year timescale. We’re still trying to figure out how to rectify the differences in our temperature measurements, the concept that we could project temps for a hundred years is almost an insanity. Still the attempt to model the climate isn’t in vain, I believe it will eventually work. The task isn’t insurmountable yet it may be far away. When it does the temp curves, model curves and projections will all be in agreement.

Why we don’t wait until then to change our way of life?

11 Responses to “Bifurcated Temperature Trend”

  1. Layman Lurker said

    So negative climate feedbacks impart negative tropospheric amplification? I don’t understand how a feedback mechanism would cause negative amplification on a decadal scale but not a shorter term scale. Any insights from Dr. Christy on that Jeff?

  2. Jeff Id said

    Not really, just that something must be there. I will ask because your question is good.

  3. “there appear to be negative feedbacks on longer time scales”
    So, delayed negative feedback? I.e., the basic cause (practically definition) of oscillation?
    Delayed negative feedback on longer timeslace than models => cyclic oscillations with period longer than models => models will project linear trends from things that are really cyclic.

    No wonder the models suck so hard, if there are oscillations too long for them to handle. Maybe one day we’ll have enough historic temp data to take a useful fourier transform, then we’ll find all the cycles! ‘Till then, let’s not go taxing carbon, eh?

  4. Jeff Id said

    I just keep asking myself how can the 30 year trend be less than GISS and yet the variance even over 5 years is more. Just how long does this feedback take…

  5. Layman Lurker said

    #3 and #4

    I think that there would have to be different mechanisms at play to cause both negative and positive tropospheric amplification on differing time scales.

  6. John F. Pittman said

    #5 Or it is like the explanation given for CO2 during past glacial cycles…an unknown starts the temperature increase, releasing CO2, CO2 takes over until it reaches it level of diminishing return, and then the unknown reverses itself and brings the temperature down. From this analogy it is similar that the unknown has a greater effect than CO2, and the sum of the influence of CO2 and the unknown after peak must be negative. The CO2 positive effect is also then of a shorter duration than the unknown. In this case the unknown is assumed to be cyclical.

    One of the problems I have with what I deem as alarmist contentions is that CO2 is so powerful. Geologically, the record indicates the opposite. CO2 can be defeated and has always been defeated by the natural order. Otherwise, the world would be like Florida from pole to pole. Note that there is an apparent discrepancy between the approach for CO2 caused warming in the past (geologic) periods and the modern period in at least the absolute effect of CO2. The estimates of the temperature are not much higher than present when CO2 was much, much higher than present, and in far exceedence of the claims of causing decalcification of coral. Yet Coral grew and prospered. And if the temperature was much warmer, the leeching and decalcification would be much worse.

    I think if Dr. Christy can make headway with his work, it will be a most interesting and enlightening read.

  7. Layman Lurker said

    I think that Dr. Christy’s explanation gives a very plausible explanation of positive tropospheric amplification with his points on ocean inertia and lapse rate feedback. I think he implies that this should be manifested in the linear trends as well, but that this affect is overwhelmed by the hypothesized negative feedback (with inverted amplification). Dr. Christy does not say whether his test will include a proposed mechanism or just a demonstration that the negative feedback exists. Can anyone demonstrate by example how such feedback / inverted amplification might work? Dr. Spencer’s low cloud feedback proposal maybe?

  8. Jeff Id said

    I have a calc which I believe demonstrates Dr. Christy was right about this. R is giving me fits right now though.

  9. Layman Lurker said

    “I have a calc which I believe demonstrates Dr. Christy was right about this. R is giving me fits right now though.”

    Definitely looking forward to that. What a great series of posts this has been Jeff.

    I think a possible source of negative amplification could come as a reversal of the postive lapse rate feedback which Dr. Christy alluded to. The positive feedback could arise as lapse rate is affected by increased specific humidity. Something (like trends in low cloud formation) may cause the reverse of this process to occur. Is such a process workable?

  10. wait a minute: maybe this is all an artifact produced by the GISS ‘correction’ process. Are you using raw giss data or are you using the ‘homogenised’ series? Since the pasteurised GISS is always ‘corrected’ towards the trend it’s supposed to have (and the filtering is I gather ‘inductive’ ie. it resists changes to the first derivative), it’s going to have less (short-term) variance but whatever long-term trend Hansen & co. want. So maybe the problem is that you’re comparing inductively-filtered GISS to (presumably) capacitively-filtered RSS/UAH – and any further filters you apply such as the two year Gaussian can’t mask the effect of an inductive filter with effectively infinite duration such as is applied to pasteurised GISS data.

  11. Jeff Id said

    #10 You have found the hidden message in all of this. The problem can exist in one or more of GISS, UAH or the models. But we know that the models cannot explain the difference between the two. By hypothesizing a hidden feedback in the models Dr. Christy suggests that the modelers look for an explanation. If there is a major correction required to models, it would mean adjustment to reduce the LT warmup above ground without adjustment to ground trend. Since he works on UAH data how does it look if he just singles out GISS. Keep in mind that the models are the only means for prediction of warming long term.

    My latest post examines what would happen if the primary UAH – GISS difference was only a linear trend. It isn’t, indicating to me that he may be right. There could be a missing feedback on the ten to 15 year scale in the models among other things.

Leave a Reply to Soundandfury Cancel reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: