Guest Post by Willis Eschenbach
I wrote before of my investigations into the surface air temperature records of the TAO/TRITON buoys in the Pacific Ocean. To refresh your memory, here are the locations of the TAO/TRITON buoys.
Figure 1. Locations of the TAO/TRITON buoys (pink squares). Each buoy is equipped with a sensor array measuring air and sea temperatures and other meteorological variables.
I have hypothesized that there is a thermostatic mechanism involving clouds and thunderstorms that maintains tropical temperature within a certain range. To investigate this mechanism, I decided to look at what happens at a given buoy on days when dawn temperatures are warmer than average, versus what happens at the same buoy on days when dawn temperatures are cooler than average.
My speculation was that when it was warmer at dawn, there would be more cloud and thunderstorm activity during that day. This would tend to drive the temperature down. On the other hand, when it was cooler at dawn, there would be less or no clouds or thunderstorms during that day. As a result, this would tend to drive the temperature upwards. And while I did find this, I was still surprised by the exact patterns.
To begin with, I compared the overall average of all days for each station with the overall average of the warmer days for each station, and the overall average of the cooler days for each station. Here are those results:
Figure 2. Average of all buoy records (heavy black line), as well as averages of the same data divided into days when dawn is warmer than average (heavy red line), and days when dawn is cooler than average (heavy blue line) for each buoy. Light lines show the difference between the previous and the following 1:00 AM temperatures.
First, the black line, showing the average day’s cycle. The onset of cumulus is complete by about 10:00. The afternoon is warmer than the morning. As you would expect with an average, the 1 AM temperatures are equal (thin black line).
The days when the dawn is warmer (red line) show a different pattern. There is less cooling from 1AM to dawn. Cumulus development is stronger when it occurs, driving the temperature down further than on average. In addition, afternoon thunderstorms not only keep the afternoon temperatures down, they also drive evening and night cooling. As a result, when the day is warmer at dawn, the following morning is cooler.
In general, the reverse occurs on the cooler days (blue line). Cooling from 1 AM until dawn is strong. Warming is equally strong. Morning cumulus formation is weak, as is the afternoon thunderstorm foundation. As a result, when the dawn is cooler, temperatures continue to climb during the day, and the following 1AM is warmer than the preceding 1 AM.
So this is the result that we would expect with a thermostat operating on a daily basis. If the dawn is warm, clouds and thunderstorms ensure that the following day starts out cooler. And when the dawn is cool, extra sun and few clouds and thunderstorms warm the day up, with the warmth lasting into the night.
Now … is this just a statistical oddity? One way to determine if we’re looking at a real phenomenon is the “dosage effect”. That is to say, the response should be proportional to the dosage. In this case, the “dosage” is the overall average temperature for that particular buoy. My hypothesis says that the effect seen above in Figure 2 should be greater in those buoys where the average air temperature is warmer, and less in those buoys where the air temperature is lower. And indeed, that proved to be the case, as is shown in Figure 3. This shows the buoys divided into four quarters (quartiles) on the basis of annual average temperature.
Figure 3. Differences between warm days (red line) and cool days (blue line) for the TAO/TRITON buoys divided into quartiles by temperature. Black line is average for all days.
Note that the response systematically grows larger and more exaggerated as we go from the first quartile (the coolest quarter of the buoys) sequentially to the fourth, warmest quarter of the buoys.
I hold these results out as strong support for my hypothesis that the temperature of the tropics is regulated by the combined action of clouds and thunderstorms. The difference in the temperature response of the warm and cool days shows the homeostatic mechanism in action, with warm mornings having cooler afternoons, and vice versa. All of this shows the clouds and thunderstorms at work.
I will ask again that if you disagree with something I’ve said, please quote it so that we both know what we’re discussing.
All the best,
w.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Rosco: Didn’t we have our global snowball during the times when the land masses were concentrated around the equator?
This is excellent work Willis. The tropics are the heat source for heat pump Earth, the poles are the radiators dumping excess heat. The chaotic unplumbed system in between gives us weather.
The last few rampant sun cycles gave us a little extra heat so the poles after lag dumped more heat as shown by a reduction in ice.
Indeed you have found the control that tries to maintain a constant heat input into the heat pump.
This is scary stuff for the AGW mob as radiative forcings become irrelevant in the grand scheme of things. You are very naughty dropping this on them as they are trying to digest the cloud stuff.
You will probably also find that the temperature in the tropics does not vary much with solar changes, the rampant sun gives less cloud cover to the oceans outside the tropics, thus more heat.
It would seem that the CERN experiment shows another thermostat, yours is for constance and theirs for the variability following the sun cycles, it will be found that there are others, and CO2 is not one off them.
Thank you Willis
Willis forget these “reversions to the mean” arguments. They are irrelevant.
Dynamical, causal processes never “revert” to any mean for the simple reason that they “produce” the mean as they go.
A dynamical causal process never knows what the FUTURE mean will be so it can’t revert to it. It may know about the PAST mean but there is no reason to “revert” to past mean.
This kind of argument is just confusing a dynamical causal process with a random process with constant mean and variance (a stationary process).
Indeed if the weather was a stationary random process then the temperatures would just be independent random variables distributed along an invariant frequency curve and would behave like a die throw.
As the proportion of events above and below the mean must stay invariant (not necessarily equal) then one has necessarily more hot->cold and cold->hot sequences than hot->hot and cold->cold sequences.
But of course we all know that the weather is neither a die throw nor a random stationary process.
Averages and variances are not constant etc.
That’s why the behaviour of dynamical variables like temperature must have a cause and what you show is that your causal hypothesis – “clouds are the cause of the temperature evolution and act like a negative feedback” is supported by the data. The negative feedback “looks” like a “reverting” to the mean but has nothing to do with it.
Of course if you had cloudiness data on the same chart, the hint would transform in a proof.
Very interesting and well presented results – also noce to hear that there is a lot more data available from this source to allow further investigation. I’d certainly say however that as a first test of your hypothesis, this looks to be confirming rather than countering it.
The quartiles data are particularly interesting – the coolest quartile suggests that below a certain temperature threshold the effect is insignificant, but that as the temperature rises the differing behaviour intensifies.
As has already been suggested, is there sufficient data on cloudiness to improve the analysis? There are a couple of questions that this data would help with:
1 – Are hotter nights related to increased cloud between sunset and sunrise (i.e. slower loss of heat overnight), or are these clear nights where the previous day has ended warmer?
2 – From the above, is there a marked change in the rate of cumulus formation in the mornings depending on the cloudiness of the previous night?
Those quartiles sure show cycles.
mindbuilder wrote: This looks to me like just regression to the mean.
Except that there is no regression toward the mean; instead there is what might be called “overshoot” or some such, producing persistent negative correlations from one day to the next.
Willis, this is nice. It would be good if you could access actual cloud data, and relate your results to the Lottka-Volterra (predator prey) modeling done earlier this year and cited above Note that the Lottka-Volterra models are “negative feedback” models and not “regime change” models (apropos an earlier interchange between us.) It is possible, or at least conjecturable, that increased concentrations of CO2 will produce increased magnitude of the day-to-day oscillation, without changing the time-averaged mean temperature. Can you see a change in magnitude of the oscillation with change in measured atmospheric CO2? The record is probably too short to to show much, but it might be worth looking at now, and following in subsequent decades.
Brian H wrote: yes, “reversion to the mean” is just a matter of any current deviation from the true average/mean being swamped and reduced to insignificance by the piling up of future instances which will, by definition, average out to the mean. Since that’s what mean means.
That is incorrect. When paired observations are positively correlated (Galton’s pioneering study collected heights of fathers and their sons), and you select a group that is extreme on one measure (tall fathers), the values of the other measure (sons’ heights), while still extreme (tall fathers have tall sons), are less extreme (the sons’ mean height is intermediate between the selected fathers’ mean and the population mean.)
If you select days that are way below average at morning, they are way above average at evening. That is not “regression to the mean”. Regression to the mean would produce evenings that are not as below average as the mornings.
I don’t mean to heckel, but I have one more suggestion which will, onfortunately require more work, but maybe not too much.
Willis said in comments:
“Well … absolutely not. Remember that despite the fact that it has a mean, it is a drunkards walk. At every point, the odds of warming and cooling are equal, 50/50”
You can settle this issue by using a control data set. Try feeding data generated from a semi-random number generator into your functions. I believe there are plenty of R codes already written which simulate the drunkard’s walk of temperature. It shouldn’t be hard to produce some test data. If the random data doesn’t show the increasing trend in your quartile analysis, then that should settle it. Or, if the random data shows the same thing, then you’ll know it’s a statistical artefact.
Gary Swift says:
August 26, 2011 at 11:06 am
Thanks, Gary, but I’ll pass. I’ve dealt with too many random numbers sets, and the appearance of the increasing patterns with the increasing temperatures in the quartile datasets is plenty of proof for me that it is not a random occurrence. I fear that a Monte Carlo analysis (using random datasets) won’t establish anything.
w.
Septic Matthew says:
August 26, 2011 at 10:04 am
Thanks, Matthew, for clarifying the issue. As I showed above, a random walk dataset has a mean, but shows no “regression to the mean”. What we are looking at in the TAO/TRITON records is something different. It is a mechanism which actively cools the warm days and warms the cool days … in other words, a thermostatic mechanism.
w.
Willis wrote: in other words, a thermostatic mechanism.
If you have not already studied the Lottka-Volterra equations, among the many systems of equations whose solutions are oscillations, you might do so — in your “spare time” of course. Probably you already have studied the Lorentz equations, another well known example.
matthew;
You introduced a qualification that renders the issue tautological: “when paired observations are positively correlated”. I was speaking of the general case, in which a “spike” or run in a random sequence temporarily seems to introduce a trend or new mean. 6 “heads” in a row up front looks very significant; add 1000 more flips that show a fair and accurate representation of the mean and you get 506 heads vs 500 tails, not significant. So the apparent “pattern” or trend “regressed to the mean” by expansion of the denominator.
The observed facts that the temperature recordings throughout the tropics are usually 10 degrees less for environments close to the ocean than those in the middle of a land mass of substantial size ought to convey that water controls the climate on the planet not carbon dioxide.
Enormous quantities of energy are absorbed by evaporating water molecules that otherwise would heat land surfaces. The ocean’s heat capacity is greater than the atmosphere and evaporation involves large input of energy with no temperature increase.
This so called positive feedback effect of water vapour is obviously wrong.
If it were right Singapore, almost on the equator with high humidity year round would be far hotter than Baghdad during summer with low humidity.
But observation show the opposite – Singapore almost never exceeds 32 C while Baghdad regularly exceeds 42 C.
I won’t call that a negative feedback because that term doesn’t sit well with me.
It simply shows how water controls the heat –
Near water – cooler during day, warmer at night
Away from water – hotter during day, cooler at night.
I would have thought water as the thermostat was so obvious scientists would have to be brain dead to ignore the observable results of at least decades of meterological data.
Brian H. wrote: You introduced a qualification that renders the issue tautological: “when paired observations are positively correlated”. I was speaking of the general case, in which a “spike” or run in a random sequence temporarily seems to introduce a trend or new mean. 6 “heads” in a row up front looks very significant; add 1000 more flips that show a fair and accurate representation of the mean and you get 506 heads vs 500 tails, not significant. So the apparent “pattern” or trend “regressed to the mean” by expansion of the denominator.
So, first off you misused the phrase “regression to the mean” which technically applies as I described it.
Second, your example is based on independent events, but Willis E. has demonstrated that the day-night pairs are negatively correlated, not independent.
This may seem like a bit of a nit picky question but your posting does say: “”””” Figure 1. Locations of the TAO/TRITON buoys (pink squares). Each buoy is equipped with a sensor array C and other meteorological variables. “””””
Now the interesting wording there to me, was this part: “”””” measuring air and sea temperatures “””””
Note that both air Temperatures and sea surface Temperatures are measured.
BUT your report purportedly graphs ALL Buoy records; presumably including both sea surface and air Temperatures. So why would you add them all in together ?
And is it not just these Buoy records, that John Christy et al reported on in Jan 2001, that proved that sea surface Temperatures, and air Temperatures (Lower Troposphere) are not even correlated (why would they be ?)
Seemingly a minor point, but it proved that the previous 150 years of global Temperature data records, for about 73% of the earth surface are pure garbage, (sea surface temperatures) and since they aren’t correlated, the corresponding Lower Troposphere air Temperatures are forever unrecoverable.
Also because oceanic currents, like rivers, meander, you can return to the very same GPS co-ordinates, and be in totally different waters from where you were last year; so even the sea surface Temperatures, are not reliable. The air and the surface, are seldom in contact for long enough to equlibrate, so they should never be the same; or correlated. Air over the Sargasso Sea a week ago, will be over Martha’s Vineyard pretty soon, thanks to Irene.
George E. Smith says:
August 26, 2011 at 5:47 pm
Sorry for the confusion. My findings regard solely the surface air temperature. I did not use the sea temperature at all.
w.
Great stuff Willis. This sort of reasoning gets the cart properly behind the horse – so to speak. Could you do a ‘nightlight’ style analysis over the buoy locations using cloud cover imagery to verify that it is actually cloudier when it’s supposed to be?
It occurred to me that Anthony’s banner pic is very appropriate to this topic with that bright cumulonimbus tower casting its long shadow 🙂
Dixon says:
August 27, 2011 at 3:41 am
Sure I could … if I didn’t have a day job and a host of other interesting projects and the outdoors always beckoning.
Seriously, the TAO/TRITON data is there and it’s not that hard to download. I encourage any and everyone to do the followup project of their choice.
Indeed. The shadow size of cumulus clouds is an under-appreciated cooling mechanism.
w.
Septic Matthew says:
August 26, 2011 at 4:38 pm
Matthew, thanks for your comments. Unfortunately, you’re reversing the causation. You can’t explain the results as being merely a “regression to the mean’ without answering the question of why the day-night pairs are negatively correlated. As you point out quite accurately, the “regression to the mean” is a consequence of the negative correlation, and thus by your own admission cannot be the cause of the negative correlation. I say the cause of the negative correlation is the combined effects of the regime changes I have discussed in this post and elsewhere.
w.
Willis: You can’t explain the results as being merely a “regression to the mean’ without answering the question of why the day-night pairs are negatively correlated.
Reread my comments in order. You agree with me, and I specifically deny that “regression to the mean” applies to your result.
I agree Matthew .
The “regression to the mean” argument can be falsified by just one sentence :
The process described by Willis is not a stationary random process.
Being non stationary , it has no constant mean which it could “regress to” etc etc .