The Most Extra New Super Global Warming Thread

[bkelly]

Originally Posted by xTYBALTx

The black line...*snip*...I presume that it would be subject to the criticisms Lucky posted earlier...

Indeed it does as the HadCRTU2v series uses data from those monitors. I plan to discuss that later.

Originally Posted by xTYBALTx

The grey line...*snip*...is... strange to say the least.

The fat grey line is a borehole temperature reconstruction done by Pollack and Smerdon in 2004 (http://www.ldeo.columbia.edu/~jsmerd...on_Journal.pdf). I believe its smoothness is due to surface temperature signals being smeared as the surface is buried so the available information has a coarser resolution.

Originally Posted by xTYBALTx

...timelines and baselines are probably difficult to determine for any given data source, and you get discrepancies.

Absolutely.

Originally Posted by xTYBALTx

The question, then, becomes one of which sources are the best sources, which sources are the worst sources...

Excellent question. One probably is better than the others and you're right that people have been working to decide. Unfortunately, the answer today is we don't know and probably never will.

Originally Posted by xTYBALTx

Or do we just average out all of the sources with a trend line? [I hate that last option. "Accuracy through mob data" is a bit like "Science by consensus."]

Agreed. "Science by consensus" is a terrible option, but until you come up with something better...


Click on the image for a larger version

Finally, just a style note, but this graph almost seems like it was designed specifically to make past temperature fluctuations appear as noise and current temperature fluctuations as remarkable.

Well, I believe the current fluctuations are remarkable.

A great post Tybalt. I was glad to see it.

bkelly

[xTYBALTx]

Well it's not very convincing when your support is derived using a tool which we both agree is pretty bad. The onus is not on me to come up with something better, it's on you.

And what's up with the land monitor based temperature data? The problems regarding reliability in temperature collection, placement of collection stations, and so on?

Thanks for the link on the gray line. As I suspected, it is indeed comprised of a regression derived from several hundred sources. This doesn't make it wrong, but again the "mob data" technique is not very rigorous.

[GlobalWarmin]

Originally Posted by xTYBALTx

And what's up with the land monitor based temperature data? The problems regarding reliability in temperature collection, placement of collection stations, and so on?

I don't believe there is any controversy about satellite temperature data, though.

[xTYBALTx]

Correct.

But, two things:

1) The sat data was largely adjusted to match the land data.

2) The temp records Kelly posted didn't include sat data - the alarming part the graph, the black line, is land data.

[bkelly]

Originally Posted by xTYBALTx

Well it's not very convincing when your support is derived using a tool which we both agree is pretty bad.

Who said I'm trying to convince you of global warming?

Let me rephrase something in my last post. '"Science by consensus" is a terrible option when searching for "the right answer".' If you wonder why I keep bringing up babies and bathwater, this is it. We have twelve sources giving twelve different answers to the same question. In addition Tybalt, you're right that there are discrepancies in all of them which is why they're published with an estimated error range. So not only do none of the answers agree, each of them probably misses the actual temperature curves of the time. If you wanted to throw out the IPCC's report because of potential political influence, I can only imagine your thinking here.

With the discrepancies, error range and missing the temperature curves, I agree that any one of these studies isn't worth much, however all twelve being in the same neighborhood makes for a very strong case. For example, suppose each of these studies is totally independent and only has a 51% chance of being accurate. If all twelve agree about a specific thing, then there is a 99.98% chance that it is correct.

The graph I presented in my last response is a special one. It's not just the temperature lines of the twelve recreations but an overlay of their margins of error as well. So, these twelve studies, flawed as they may be, tell us with 80% confidence that the times around 1450 and 1500 were about .35 degrees centigrade colder than the baseline for each study. We're also 60% confident that the years just before the 1100s was about .2 degrees colder. I find that last stat amazing because a few studies stopped at the 1400s and the harder it is for consensus the further you get from the year that each study was centered.

So, I had three points with this graph:

1) If you believe the temperatures from the HadCRTU2v network, we're almost certainly the hottest we've been in 600 years and likely the hottest since 700 A.D, although we're not as sure because few studies have gone that far back. If you don't accept HadCRTU2v temperatures, we're still 80% likely to be the hottest in the last 600 years.

2) Flawed data, disagreeing results and the possibility of taint do not necessarily invalidate evidence. Certainly look at it with a skeptical eye and give it less weight in your considerations but don't assume it's all bad just because it's not perfect.

3) I believe anyone who speaks of global warming in absolutes is overstating their case. These recreations are good evidence, but it's possible they are totally wrong. At the moment, there is no concrete proof that global warming is or is not happening and people who say one or the other have either closed their minds or have an agenda.

Originally Posted by xTYBALTx

And what's up with the land monitor based temperature data?

*smiles* Their readings.

bkelly

[xTYBALTx]

I'm not saying all evidence is invalidated.

Originally Posted by bkelly

With the discrepancies, error range and missing the temperature curves, I agree that any one of these studies isn't worth much, however all twelve being in the same neighborhood makes for a very strong case. For example, suppose each of these studies is totally independent and only has a 51% chance of being accurate. If all twelve agree about a specific thing, then there is a 99.98% chance that it is correct.

Except that not all data are in the same neighborhood. Some of the data sources clearly suggest current warming to be unremarkable. Others clearly suggest that the current warming is indeed quite remarkable. Some show numerous large spikes and valleys over very short periods of time, others show few deviations beyond .2C+ even over long periods. This doesn't mean that they're all wrong. It does make me wonder which one(s) I should believe.

Ultimately though, we agree:

Originally Posted by bkelly

...At the moment, there is no concrete proof that global warming is or is not happening and people who say one or the other have either closed their minds or have an agenda.

You certainly won't see me disagreeing with you on that point. I don't believe the AGW theory has been disproved - it may very well turn out to be true and the potential consequences are frightening; and I also think it's important to note that "concrete proof" isn't really required for action on this issue, as concrete proof doesn't really exist. But you must understand why it's so queer to us that not only the IPCC but apparently many researchers and academics do indeed state the AGW theory with such force and certitude, when the theory doesn't seem very robust even on cursory inspection.

It seems like our disagreement, aside from differing "interpretations" of data as outlined above and elsewhere, is that you think the AGW theory is "mostly" supported by evidence whilst I think the AGW theory has a ways to go before it gets there.

[Kerostasis]

Originally Posted by bkelly

With the discrepancies, error range and missing the temperature curves, I agree that any one of these studies isn't worth much, however all twelve being in the same neighborhood makes for a very strong case. For example, suppose each of these studies is totally independent and only has a 51% chance of being accurate. If all twelve agree about a specific thing, then there is a 99.98% chance that it is correct.

Thats...uh...not quite true. One moment while I pull out my calculator.

*calculates*

Your number is about right, but your interpretation of the meaning of that number is totally off base. It would be more correct to say that, if that reading was false, there would only be a .02% chance that all 12 studies would say it was true anyway. That seems impressive, until you notice that even if that number was true, there would still only be a .03% chance that all 12 studies would say it was true. The other 99.95% of the time, at least one of the 12 studies would disagree with the other 11. Studies with a 51% chance of being accurate really aren't very helpful no matter how many of them you have.

What you need instead is a way to show that the studies are more reliable than 51%, rather than simply relying on the fact that there's lots of them.

[bkelly]

Originally Posted by xTYBALTx

Except that not all data are in the same neighborhood. *snip* Some show numerous large spikes and valleys over very short periods of time, others show few deviations beyond .2C+ even over long periods.

I agree, but we have two different definitions of "neighborhood".

I think the data shows remarkable consistency in terms of the degree of the temperature anomaly throughout the time scale. In English, they generally trend up and down together. I suspect almost all the data points are within three standard deviations of each other so an average is a quality estimate. This allows us to say with some confidence that it's warmer now than it was then.

However, what you are saying is right on- some of the studies have huge spikes while others are calm. That's the volatility of each study and some are very volatile and some are not. An averaging destroys volatility so the flatness of the graph above means nothing. If we want to know how unusual the recent jump in temperature is with these studies we'll need to break them down in a different way. Perhaps I will do that later, but right now, I agree, these graphs say nothing about the rate of the recent temperature increase.

Originally Posted by xTYBALTx

This doesn't mean that they're all wrong. It does make me wonder which one(s) I should believe.

Until one is proved wrong or another proved more right, I think you need to believe them all a little bit.

Originally Posted by xTYBALTx

But you must understand why it's so queer to us that not only the IPCC but apparently many researchers and academics do indeed state the AGW theory with such force and certitude, when the theory doesn't seem very robust even on cursory inspection.

Each piece of evidence isn't "robust", but there are so many pieces that the evidence is often called "overwhelming". The 2007 IPCC report said human caused global warming is "very likely" which is defined as a >90% probability. Many pushed for the stronger phrase "extremely likely" which would mean >95% probability.

I made sure to write "at the moment" in my third point regarding concrete proof because I believe more evidence will come in the next 5 to 10 years and somewhere between 99% and 99.9% probability scientists are justified in calling it a fact.

bkelly

[bkelly]

Originally Posted by Kerostasis

It would be more correct to say that, if that reading was false, there would only be a .02% chance that all 12 studies would say it was true anyway. That seems impressive, until you notice that even if that number was true, there would still only be a .03% chance that all 12 studies would say it was true. The other 99.95% of the time, at least one of the 12 studies would disagree with the other 11.

I don't think I'm getting your point. It just sounds like to me your repeating the same stat from four different perspectives.

1) Odds of a result being correct if 12 independent 51% studies agree upon it: 1 - .49^12 = 99.98%

2) Odds of a result being wrong if 12 independent 51% studies agree upon it: .49^12 = 0.02%

3) Odds of 12 independent 51% studies agreeing on a single correct point of data: .51 ^ 12 = 0.03%

4) Odds of at least 1 of 12 independent 51% studies missing a single correct point of data: 1 - .51 ^ 12 = 99.97%

It seems to me that all these points of view say that when 12 independent 51% studies agree, you better pay attention!

Originally Posted by Kerostasis

Studies with a 51% chance of being accurate really aren't very helpful no matter how many of them you have.

No! My point was the complete opposite. Even with a very slim advantage you can see which side is more likely to win with enough trials. A Las Vegas craps table is a terrific example of this.

bkelly

[Kerostasis]

Let me explain. Keeping the same example, Suppose we have 12 trials of 51% accuracy. Before we even look at the results, we can predict that 99.95% of the time, the studies will disagree. Of the small number of cases where they all agree, we have .03% where they are all right and .02% where they are all wrong.

Now we look at the data and discover that they do, in fact, all agree. We know that today, all of those 99.95% cases didn't occur. That leaves the .03% of all correct and .02% of all incorrect. But thats no longer .02% out of 100% total -- thats .02% out of .05% total. So the end result % is 40% chance that they all agree incorrectly, and 60% chance they all agree correctly. That tends in favor of the data being correct, but not very far...60% is a pretty weak level of confidence, considering we had 50% confidence before we even took any measurements at all.

[bkelly]

Originally Posted by bkelly

This allows us to say with some confidence that it's warmer now than it was then.

I agree, these graphs say nothing about the rate of the recent temperature increase.

I've been thinking about this and what I said isn't quite right. Let me explain.

I still believe that the way we've looked at the information in the graphs says a lot about general temperature trends but nothing about it's volatility. My mistake was using the word "now" and "then" and suggesting we can compare temperatures in pinpoint moments in time. Since we only know trends, not volatility, that's an incorrect statement.

What I should have said was "This allows us to say with some confidence that the recent decade was warmer than any decade in the past 1000 years."

bkelly

[bkelly]

Originally Posted by Kerostasis

...we can predict that 99.95% of the time, the studies will disagree.

How did you calculate the 99.95% probability?

[xTYBALTx]

Originally Posted by bkelly

I think the data shows remarkable consistency in terms of the degree of the temperature anomaly throughout the time scale. In English, they generally trend up and down together.

You must be looking at a different graph, because the one you posted shows nothing of the sort. The one I see looks more like the one you describe in the following passage:

Originally Posted by bkelly

However, what you are saying is right on- some of the studies have huge spikes while others are calm. That's the volatility of each study and some are very volatile and some are not. An averaging destroys volatility so the flatness of the graph above means nothing. If we want to know how unusual the recent jump in temperature is with these studies we'll need to break them down in a different way. Perhaps I will do that later, but right now, I agree, these graphs say nothing about the rate of the recent temperature increase.

Originally Posted by bkelly

Each piece of evidence isn't "robust", but there are so many pieces that the evidence is often called "overwhelming". The 2007 IPCC report said human caused global warming is "very likely" which is defined as a >90% probability. Many pushed for the stronger phrase "extremely likely" which would mean >95% probability.

Seems like you might have some internal contradictions somewhere. Once second you're saying the regression is unacceptable and the next you're saying that it's okay. One second you're saying that the data may make recent data appear unremarkable, in the next you're saying that there is "overwhelming" evidence for AGW.

[Kerostasis]

Originally Posted by bkelly

How did you calculate the 99.95% probability [that at least 1 study would disagree]?

The science of statistics looks at things backwards from the way most people are accustomed to looking at data. Rather than asking, "how likely is it that this data point is correct?", we ask, "given a true data point of [insert number here], how likely is it that our study would still have produced the number we see?" To get a meaningful answer, you usally have to ask the question several times using a range of different [insert number here] values.

To keep things simple, I assumed that we had only 2 different possible values for our data point: TRUE and FALSE. This is of course an oversimplification, but its the same simplification you seemed to be using so I ran with it.

So our statistics questions come out to "Given a true data point of TRUE, how likely is it that all 12 studies would produce TRUE?", and "Given a true data point of FALSE, how likely is it that all 12 studies would produce TRUE?" The math to answer these 2 questions is pretty simple: the first is 51% ^ 12, while the second is 49% ^ 12.

Now to your question: If the studies didnt all agree on TRUE, and they didnt all agree on FALSE, then at least one of them must have disagreed. So the probability of disagreement is 1 minus the probability of agreement, or "1 - .51^12 - .49^12", which comes out to 99.95%.

--EDIT--
I made a slight mistake there and blurred two scenarios, although it doesn't end up changing the math any. For reaching the 99.95%, we ask "given TRUE, how likely will all studies say TRUE", and "given TRUE, how likely will all studies say FALSE", and those 2 also have a combined likelihood of .05%, leaving 99.95% chance of disagreement assuming the real datapoint is TRUE. If the real datapoint is FALSE, the numbers are reversed but still come out to the same total probability. The questions I listed above are still useful for a slightly different purpose: Once we see that the 12 studies did, in fact, agree, all other scenarios except those 2 can be thrown out, and we only have to compare the probabilities of those 2 scenarios. We can then see that though they are both extremely unlikely, one is only slightly more unlikely than the other, so the revised probability using our new data is about 60/40.

[bkelly]

Originally Posted by xTYBALTx

You must be looking at a different graph, because the one you posted shows nothing of the sort.

No? You don't see each trial showing rising temperatures from 700 to 1000 and then a drop-off from 1000 to 1250? 1250 to 1500 is chaotic but 1500 to 1850 shows a slight cooling trend while 1850 to the present is quick warming.

You don't see those trends in most of the data points in the first graph?

Originally Posted by xTYBALTx

Once second you're saying the regression is unacceptable and the next you're saying that it's okay.

I don't believe I've said either but I understand my point is confusing.

I believe the data is as simple as it can get in the first graph without butchering the available information. Each study needs to be considered as an equal attempt to describe reality. Representing all twelve studies by a one line average is a "terrible option" because it destroys so much data and implies things that aren't true. We have twelve different answers and we shouldn't forget that.

What an average of the twelve studies offers is a better estimate of the average of the actual temperature. For example, we are more confident that the average temperature around 1000 A.D. was .2 degrees below the baseline than the .1 above indicated by the ECS2002 study because all the studies combined give that average temperature at that point a higher probability.

Now, it's possible the climate was almost never .2 degrees below as it spent most of it's time 5 degrees above or below that temperature. An average won't tell us that, but calculating the volatility will yield a better estimate than any study alone. One way is to determine the variance/standard deviation (http://en.wikipedia.org/wiki/Variance) of each study and average those to get an idea of how much reality may depart from the average calculated above.

A mean average is also a good place to start if we want to find potential trends. For example, the second graph does show the trends I described in my first paragraph of this post but also shows a possible larger warming period from 1625 to 1775 inside the 1500-1850 slight cooling I mentioned before. I didn't see that just looking at the raw data, so averaging them was helpful.

So, in our case an average makes a terrible summary of the data but a good tool in understanding what it all means.

Originally Posted by xTYBALTx

One second you're saying that the data may make recent data appear unremarkable, in the next you're saying that there is "overwhelming" evidence for AGW.

Right, I was addressing your frustration of people's certainty when the pieces don't seem "very robust". One piece of unimpressive evidence is... unimpressive, but a hundred pieces together is very strong evidence. It's kinda like a train squashing one penny on the rails, but being unable to start when a penny is put in front of each wheel.

bkelly