Guest Post by Willis Eschenbach
Well, either it’s a genetic defect or I’m just a glutton for punishment, but I’m going to delve some more into the ice ages. This is a followup to my previous post, Into and Out Of The Icebox. Let me start by looking at the cycles in the insolation and the cycles in the geological temperature. I’ll use the same temperature proxy dataset used in the discussion by Science of Doom here and here, which is the Huybers ∂18O dataset . For the insolation, I’m using the same Berger dataset that I used in my last post. Figure 1 shows the cycles in the two datasets:
This graph demonstrates extremely clearly what is called the “100,000 year problem”. As you can see, the length of the ice ages has a very strong 100,000 year cycle, with a cycle amplitude greater than 40% of the swing of the data.
But in total contradiction to that, the June insolation at 65°N, which is the insolation that is supposed to cause the interruptions of the ice ages, has virtually no cycle strength in the 100,000 year (100 Kyr) range. The insolation has its greatest cycle strength between 19 and 24 Kya, and a smaller peak at 41 Kyr, but there is almost no power at all in the 100 Kya range.
It is worth noting that both the temperature and the insolation do show power in the ~ 23 Kyr and the ~ 41 Kyr range … but only the temperature has power in the 100 Kyr range.
Now, back in 2006 Gerald Roe wrote a paper called “In Defense of Milankovich”. In that paper, he said that the reason there was little relationship between the Northern Hemisphere insolation and the ice ages was that people were looking at the wrong thing. His point was that when the sun increases, the ice doesn’t immediately disappear. Instead, what changes is the rate of melting of the ice. This is also called the “first difference” of the ice volume. Roe used an earlier version of the same Huybers temperature proxy dataset I’m using to demonstrate his hypothesis, reasoning that the ice volume is a function of the global temperature.
So let’s start by looking at the effect of taking the first differences on the underlying cycles. Figure 2 is the same as Figure 1, except that I’m using first differences instead of using the raw Huybers temperature proxy data.
Now, that is an interesting result. As you might imagine, it hasn’t introduced any new frequencies into the mix. However, it has greatly decreased the size of the 100 Kyr cycle, slightly increased the size of the 23 Kyr cycle, and slightly decreased the size of the 41 Kyr cycle.
And what would be the result of those changes? Well, the correlation will indeed be better, as Roe observed … but for the wrong reasons. The correlation will be greater because in the temperature data (blue) the ~ 20 Kyr cycle and 41 Kyr cycles are now about the same size as the 100 Kyr cycle. So those cycles will fit better … but we still have no explanation for the 100 Kyr cycle.
In any case, here’s the match between the June insolation at 65°N and the first difference of the temperature proxy:
Figure 3. A comparison of the June insolation at 65°N (red) and the first difference of the ∂18O temperature proxy. I am using the negative of the ∂18O data, so that increasing values show increasing temperatures.
Looks good, doesn’t it … but it’s not. Unfortunately, this is merely a wonderful example of the human propensity for seeing patterns. If you look at parts of this, it looks like a perfect match. The problem is, humans are shaped and bred by millions of years of evolution to find visual patterns … and as a result we find patterns even where no such patterns exist. The best example I can give you is that virtually every culture has found constellations in the stars. We identify Orion and Gemini and a host of others … and despite that, the stars contain no such patterns, just a random scatter.
And when we look closely at Figure 3, we can see that in many of the cases, the blue lines are in between the red lines … in all, they seem to be aligned at around 600 Kyr BP and also around the present, but badly out of alignment in between.
In order to keep ourselves from making such mistakes in pattern identification (among other reasons), we’ve invented an entire branch of mathematics called statistics. It allows us to do things like measure just how much of one variable is explained by another variable. The measure of this is called “R^2”. It varies from 0.0 (no relationship) to 1.0 (one variable totally explains the other).
And the R^2 value for the two variables above? How much of the first difference of the temperature variation is explained by the variation in northern insolation?
Well, the R^2 of the two is a mere 0.05 … that is to say, the June insolation at 65°N only explains about 5% of the variations in the first difference in temperature. Color me unimpressed.
Now, it’s possible that there is some lag in the data. To check that, we can run a “cross correlation”. This looks at the correlation, not just at the same time, but at a variety of time lags. Here is the cross correlation of the two variables:
Figure 4. Cross correlation of insolation and first difference of temperature. Positive lags show temperature changes lagging insolation changes. Blue lines show the level where the p-vaule is 0.05, which must be exceeded to achieve statistical significance.
So … there you have it. The relationship just barely achieves statistical significance. Is it true that looking at the first difference of the temperature improves the correlation? Yes, it is … but for the wrong reason. Taking the first difference of the temperature proxy reduces the amplitude of the 100 Kyr signal and increases the amplitude of the ~20 Kyr signal. Since the ~ 20 Kyr signal is the largest signal in the insolation, as a result the overall correlation increases … but this still doesn’t help us at all with the “100,000 year problem”. Not only that, but at the end of the day, the relationship is so weak as to scarcely achieve statistical significance.
Me, I’d say that Roe certainly didn’t solve the 100,000 year problem … although as always, YMMV …
Best wishes to everyone,
My usual request—if you disagree with someone, please QUOTE THEIR EXACT WORDS THAT YOU DISAGREE WITH. This is the only way for everyone to be clear as to the exact ideas that you are objecting to.