I am currently investigating how to find errors in trend lines for autocorrelated data (and temperature data is often autocorrelated). A few months ago, I had never even heard the term 'autocorrelated' ...
David,
Here's a link that might help in the discussion of 95% confidence intervals on an autocorrelation "correlogram":
I do not know what you mean when you say that the atmosphere above the oceans is not warming. It is. Look at the graphs.
As to natural variation, over the short term natural variation overwhelms increase in temp due to CO2. This is why you can get flat decades or booming decades. I have never denied this - indeed, my admittedly very simple models show this clearly.
As to the sun, yes, I agree: when solar output is low, solar output is low. The question is: what is the magnitude of the effect on global temperatures? All studies show that, when measured over a decade or so, this is smaller than the effect of CO2 increase.
__________________ 'Avast, ye scurvy dogs! Rotate the rubber baby buggy bumpers!'
Because if the physics of the measurement is false, one can have a perfectly normal set of statistics that won't give you the correct answer. I had that happen to me in one experiment I did. Everything looked normal statistically but we screwed up and didn't actually measure correctly what we thought we were measuring. Happens all the time.
But the key to this portion of the debate is that you, Glenn, seem to be trying to get us to tell you how the tails of any given distribution can possibly make physical sense.
The answer all along is that they probably don't. That is kind of what is meant by error.
In the journal articles I've been reading they discuss this very point:
Calculation and reporting errors can be large (changing the sign of a number and scaling it by a factor of 10 are both typical transcription
errors; as are reporting errors of 10C (e.g. putting 29.1 for 19.1)) but almost all such errors will be found during quality control of the data. Those errors that remain after quality control will be small, and because they are also uncorrelated both in time and in space their effect on any large scale average will be negligible.(Brohan et al.)
And I'm not even sure they have caught all or most of the bad errors in the data.
As we've seen a lot of bad data is in the RAW records. There's almost no way I can imagine for two stations 20 miles apart in Oklahoma to have one day in which the difference between means was 60 degrees. So I assume that is a bad transcription/QC error.
But the point, indeed the POWER of statistics is that it allows us to still use data that has problems as long as we have some appreciation for those problems and indeed as long as we attempt to overcome them.
The Brohan paper goes on to say in another area:
To interpolate the station data to a regular grid the methods of [Jones & Moberg, 2003] are followed. Each grid-box value is the mean of all available station anomaly values, except that station outliers in excess of five standard deviations are omitted.(ibid)
The end result of all this QC and averaging and error terms is that the data we wind up with should be less impacted by strange, completely bizarre, probably horribly erroneous data.
If you want to measure the temperature of the atmosphere and you put your thermometers next to furnaces. You would have a statistically gaussian distribution of error. But you wouldn't have measured the temperature of the atmosphere--only the temperature of foundaries.
BUT only if the majority of your temperature measuring equipment had that bias. So far no proof has been presented that the majority of equipment (be it sea surface temp measurement or land-surface temp measurement or satellites) have that same systematic bias.
In every collection of gauges there are bound to be bad gauges.
If I'm reading this correctly it should allow you to find the number of lags after which you cannot reject the null hypothesis (Ho: no autocorrelation).
But again, I'm not sure on that. (Thanks for bringing this up, this is interesting!)
The paper that you linked to seems to give me exactly what I need: a way to calculate the autocorrelation coefficient. I had seen a similar formula, but it was presented in a way that I did not understand. This seems much simpler.
__________________ 'Avast, ye scurvy dogs! Rotate the rubber baby buggy bumpers!'
"The world’s ocean surface temperature was the warmest for any August on record, and the warmest on record averaged for any June-August (Northern Hemisphere summer/Southern Hemisphere winter) season ..."
You all think I am wrong to believe in natural variation as a major cause of the climate change over the past 40 years. New Scientist this week has an interesting article on a speech given by Mojib Latif, one of the IPCC authors. Note how he has to be sure to tell everyone he isn't a skeptic. I know why he does that--if people think you doubt the almighty God of global warming, they will quit listening to you.
"We could be about to enter one or even two decades of cooler temperatures, according to one of the world's top climate modellers."
"'People will say this is global warming disappearing,' Mojib Latif told more than 1500 climate scientists gathered at the UN's World Climate Conference in Geneva, Switzerland, last week. "I am not one of the sceptics. However, we have to ask the nasty questions ourselves or other people will do it.'" Fred Pearce, "World will 'cool for the next decade'" New Scientist, Sept 12, 2009, p. 10.
...
OH my GOSH!!!!! He is saying something very close to what I have been saying!!!!!!
I am not so sure of that. I am currently in the process of going through the actual presentation and listening to the audio. It sounds to me like Dr. Latif is discussing the variability in the measurements on shorter time scales (ie multidecadal), in order to help explain how this functions in relation to the larger overall trend.
Actually if you listen to the audio of Dr. Latif's presentation he clearly states early on that you can see the long term warming trend in the graph of data and "we all believe this long term warming trend is anthropogenic in nature". (Audio available HERE)
From what I've been able to gather Latif is not saying anything that isn't already known in the world of global climate change.
The key is that variability on multidecadal scales and shorter will overwhelm the overall trend on short time scales:
Originally Posted by IPCC
Difficulties remain in reliably simulating and attributing observed temperature changes at smaller scales. On these scales, natural climate variability is relatively larger making it harder to distinguish changes expected due to external forcings.(SOURCE)
The key point is that no climatologists believe that global warming is "monotonic" (ie simply increasing linearly without any ups and downs). Decadal variability is still a big player. From what I can gather that is the key point of Latif's paper.
Latif's powerpoint presentation can be downloaded here
The New Scientist article states:
Originally Posted by NewScientist
Yet many now agree that the short-term prognosis for climate change is less certain than once thought.(SOURCE)
And I think that is telling. Indeed short term variability makes it harder to see longer term overall trends, which appears to be the crux of Latif's presentation.
A transcript from the audio of Dr. Latif's presentation (according to this site) reads:
It may well happen that you enter a decade, or maybe even two- you know- when the temperature cools- alright- relative to the present level- alright?
And then- you know- I know what’s going to happen -you know? I will get- you know- millions of phone calls- you know:
“Eh, what’s going on? So, is global warming disappearing?” You know? “Have you lied on [sic] us?”
So- you kn0w- and therefore this is the reason why we need to address this decadal prediction issue.
So it seems that Latif is acknowledging the science so that when people misinterpret the science for whatever agenda it will already be out there and discussed.
Indeed Dr. Latif does discuss the relative short-term variability versus the long-term variability and noted that there is still some discussion about how that impacts the long-term trend estimation. And indeed he also stresses the importance of better models. I don't think anyone is disagreeing that the detailed understanding of any system can always be improved upon. Especially in a discussion of such important matters.
Last edited by thaumaturgy; 17th September 2009 at 09:57 AM.
Linear Mode
