Guest Post by Willis Eschenbach
I wrote my first computer program in 1963. It was an implementation of the Sieve of Erastosthenes, used to find prime numbers. I haven’t stopped programming since then. So I am intimately acquainted with the innards of computers, computer programs, and computer models, both iterative and otherwise.
And those who read my writing know that I don’t have much use for the suite of IPCC computer climate models as a way to predict the future evolution of the global climate. I think they are Tinkertoy™ exercises in parameter tuning.
So it may come as a surprise that there is a model out there that I wouldn’t say I trusted, that would be far too strong. But I would say that it certainly bears watching, because it’s the best of the models. It bears scant similarity to any of the IPCC models. In its current incarnation, it has the lovely name of GATOR-GCMOM.
It came up recently in a Discover magazine article entitled “White Roofs May Actually Add to Global Warming”. Go figure, huh? Figure 1 shows what painting roofs white was supposed to do for the climate.
Figure 1. How it was supposed to work … but didn’t. Ah, well. So much for the White Roof Project.
When I read the Discover story, I wasn’t surprised to find that the model that produced such a counterintuitive result was the same GATOR-GCMOM, whose development I’ve been following and speaking favorably of for over a decade. It’s the work of a brilliant man named Mark Jacobson at Stanford University. It started small, as a local or regional model to trace the paths of pollutants around point sources. In its current form it includes literally dozens and dozens of chemical, atmospheric, and oceanic processes which are not represented in any other climate model on the planet. It uses a variety of ingenious ways to do things to reduce computational overhead. A full list of the differences from IPCC models and a discussion of the development of the GATOR model is here (PDF).
So why does the GATOR-GCMOM model say that painting the roofs white will heat the planet?
The Discover article says:
The model found that more white roofs means less surface heat in cities (which is obvious enough to anyone who’s sat in a car with a black interior in the sun). Lower local temperature means less water evaporates and rises up to eventually form clouds, says lead author and Stanford University researcher Mark Jacobson. The decrease in clouds allows more sunlight to reach the Earth’s surface, leading to higher temperatures overall.
So, clouds once again affect the climate in an unexpected way. I’m shocked. The article also states:
The model also predicts that much of the light reflected by rooftops will eventually be absorbed by dark carbon soot and particulates that are especially prevalent in the air above urban areas. This could limit local cooling and cause warming elsewhere as the particles drift away.
This shows an unexpected (but reasonable) interaction between two factors, reflected sunlight and black carbon particles in the air.
Why would I think that Jacobson’s model might be showing something near reality in this question, when I am generally scornful of the IPCC models? Several reasons:
1. The time frame of the analysis is short, he’s not futzing around with 100 year fantasy forecasts.
2. Both outcomes, once examined, make sense. Changes in clouds, and in atmospheric heating from sunlight hitting black carbon, certainly would affect the outcome, the physics is well established.
3. The GATOR model started small, modeling local conditions, many years ago and built gradually outwards from there. From the start, it was frequently compared to reality and tested and refined. It wasn’t conceived of as a global model like many climate models. So it was continually being tested on how accurately it could represent the temporal evolution a host of local conditions around cities and bays, studying pollution plumes and their changes over time, comparing them to observations … a host of real-world testing unlike anything that any of the IPCC models have undergone. Then, over about twenty years, it has been slowly expanded to be a global model.
4. I may be wrong, but I cannot find any indication of tunable parameters anywhere. Seems like there must be some somewhere, but for the most part it’s truly physics and chemistry based, unlike IPCC models.
5. It uses a nesting grid scheme which allows for a variety of grid-size resolutions as needed. This lets some areas be intensively sampled (say around a city) while a larger area of the ocean might need far fewer samples.
6. It handles chemistry at a very detailed level, involving hundreds of chemical compounds in both the ocean and the atmosphere. Other climate models don’t even touch chemistry except perhaps in the simplest ways.
7. The result was counterintuitive, but still demonstrable. A model that only shows us what we already know is not that useful. This one showed us something we didn’t know.
Anyhow, for me the takeaway message is CLIMATE ISN’T LINEAR. The IPCC paradigm is, change the forcing and the temperature has to change proportionally.
But in this case, not only is the temperature response not proportional. It’s not even in the right direction. Kinda deals the whole “temperature change equals forcing change times climate sensitivity” idea a body blow …
So that’s all the reasons why I find this result quite plausible. It’s the best model on the planet, and it is uniquely qualified to look at this particular question. If Jacobson were to start using the model for hundred year runs looking for trends, that would be a big question mark for me, I don’t think any model can do that.
But for this kind of analysis? It does what the best of models can do—it points at things in front of our eyes which we might not have noticed. Doesn’t prove anything, the output of a computer model is never evidence … but it certainly teaches us something, which is much more valuable. It teaches us that in a complex system like the climate, a simple, totally obvious cause and effect relationship may not work out anything like that. As in this case, where something that obviously, logically, and unquestionably will cool the earth … may just end up warming it.
Anyhow, that’s the latest news from the land of Settled Science, where all temperatures are unshakably tied to forcings …
Regards to everyone,
w.
The model found that more white roofs means less surface heat in cities (which is obvious enough to anyone who’s sat in a car with a black interior in the sun). Lower local temperature means less water evaporates and rises up to eventually form clouds, says lead author and Stanford University researcher Mark Jacobson. The decrease in clouds allows more sunlight to reach the Earth’s surface, leading to higher temperatures overall.
This seems really counter-intuitive. Looked at the other way around, the paved surfaces in cities that are currently absorbing sunlight should be raising the evaporation rates and creating more clouds. This would lead to a net COOLING according to this logic. So the “Urban Heat Island” effect would be a net COOLING of the rest of the world as much as it is a warming of the cities.
4. I may be wrong, but I cannot find any indication of tunable parameters anywhere. Seems like there must be some somewhere, but for the most part it’s truly physics and chemistry based, unlike IPCC models.
The is the number one key factor in making a model that might actually produce meaningful results.
Any model that has a “tunable parameter” is subject to the “experimenter expectancy effect” or “clever Hans effect” When the model doesn’t fit what the experimenter expects, they change the parameters so that the model fits their expectations.
This has the effect of introducing errors into the model, because not all inputs to the model are 100% accurate. So, in effect by tuning the parameters, you are making it impossible to isolate input errors from model errors. Over time this allows errors to build up in the model, so that its value for accurate prediction is reduced. In effect the model ends up predicting exactly what the experimenter expects it to predict, not what the model’s internal physics says it should predict.
Whether climate science recognizes the problem or not, parametrized models are machine learning programs. Their design is such that they are intended to learn and deliver more accurate results over time. Back-casting is a machine learning technique. You are training the machine using past experience in the hope that this will improve its future performance. As is done when training animals, humans and students. In the case of back-casting the effect can be even more subtle, as the computer typically adjusts its own parameters (weights) to improve the fit. Any errors in the inputs thus train the model to make errors in the outputs.
As such, computer models suffer from much the same problem that occur in animal learning studies. Unless climate models are designed using double blind or similar techniques, they are unlikely to deliver meaningful results.
Here is what Wikipedia has to say:
” Recognition of this striking phenomenon has had a large effect on experimental design and methodology for all experiments whatsoever involving sentient subjects (including humans).”
http://en.wikipedia.org/wiki/Clever_Hans#The_Clever_Hans_effect
http://en.wikipedia.org/wiki/Observer-expectancy_effect
RE: richardjamestelford October 21, 2011 at 1:30 am
I have to agree with you. I don’t know about the sky over Stanford, but most of the US cities I’ve visited are not so sooty that absorption of reflected sunlight could ever be a significant factor.
Likewise, the difference in evaportation would have to hit just the right sweet spot to make a difference. It could be the case that there is enough evaporation to form clouds even if the roofs were white, so the white roofs would have no warming effect in this case. And if having dark roofs didn’t form a cloud then the additional water vapor as well as the additional IR emissions would increase local greenhouse warming, along with the heat directly transferred into the air from contact with the hotter roof.
I’m a big fan of Willis’ writing, but think his trust in the GATOR-GCMOM model is misplaced, especially in light of this result.
“an implementation of the Sieve of Erastosthenes”
That’s always the first or one of the first programming assignments I give when teaching programming. A great way to see if their code is tight – or not.
So, I’m thinkin’ Steven Chu is going to be crushed with this news?
richardjamestelford says:
October 21, 2011 at 1:30 am
Easy for you to say now. Where were you when the “White Roof Project” was getting all of the press? How come you didn’t let Secretary Chu know?
If you can break out a statement of yours from before this study saying the same thing, that white roofs would not cool the earth but would warm it, that would be impressive.
Right now, it just sounds like my kid saying “I knew that already, Dad” …
w.
Harold Ambler says:
October 21, 2011 at 3:20 am
I know that … but damn, the boy can build a model. I don’t think it (or any) model can solve the basic questions about CO2 as people are always trying to do, using 100 year runs.
I do think the model can be useful, though, for studies like this one.
w.
Russ R. says:
October 21, 2011 at 7:23 am
I’ve never heard of it doing so, if it does.
w.
Great work, Willis. Your description of the model strikes just the right balance between its use as an Analytic Tool to reveal things that we do not yet know, the net warming effect of white roofs, and its incorporation of an unusually complete Description of the Natural Processes underlying the phenomena investigated. To be clear, the natural processes include the chemistry not found in other models. All of this is not yet science, as you recognize. The science comes in creating and testing physical hypotheses which provide a more complete and detailed description of the natural processes that make up climate. Limited parameters means limited rejiggering and limited fudging.
Thanks for getting out the message about this model and its author.
This whole thread is scientific navel gazing on a par with the worst of realcimate (and that for me is the benchmark).
The science being discussed here is totally irrelevant to the problems that face the human race (the planet is fine by the way).
The issue is about climate; is it changing, how is it changing and why is it changing? Any and all computer models including the gator variant are totally and utterly useless. Any and all currently understood science is totally and utterly useless.
If you dont understand everything about how and why the climate of this planet changes then you dont know anything at all in terms of predicting the future.
Scientists need to show a little humility and accept that.
If you dont understand everything then controlling the climate is a pipe dream. It is quite possible that in a thousand years when (highly unlikely) we do understand absolutely everything, we will still not be able to control our climate.
History is telling us a great deal about how climate changed in the past, highly intelligent and well educated humans should concentrate on making sure we can survive (as a society and as a species) the events of the past. That is the best we can do.
The fact that some scientists and apparently all politicians are crippling the world economies, depriving their citizens of hard earned freedoms, comfort and in some cases lives in order to prevent one possible contributor to climate change is the most disgusting insanity I have ever witnessed.
TomB says:
October 21, 2011 at 8:12 am
In 1963, nobody’s code was tight, not by modern standards. Plus we were using ALCOM (IIRC), an ugly inefficient language to start with.
These days I program in R, and I can’t recommend it enough. Steve McIntyre was the one who got me into using it, and it has been much more than worth it.
w.
Quick hint for someone wanting to become the next Steve Jobs or Bill Gates. Develop a material which can be changed from black to white and back using a trickle of electricity (something like liquid crystals, but not necessarily the same). Then hook it up to a computer and computer program (say using panels one meter square as the unit). Feed the program the layout of the house, the Solar insolation, and the weather forecast. Let it determine which panels should be white and which black at a given time. Should reduce heating and air-conditioning costs much more than the cost to run the system. Caveats and details left as an exercise for the entrepreneur.
This sounds like an unstable feedback effect. White roofs reflect sunlight from the surface causing the average surface temperature to be cooler so less clouds form allowing more sunlight to reach the surface and thus making the average surface temperature warmer than it would have been if the sunlight had not been reflected from the roofs in the first place.
Perhaps there is just a simple negative feedback effect that results in about the same overall surface temperature no matter what color the roofs are.
That’s what I was trying to say above. You are describing the Thermostat Hypothesis™ by Willis, in action.
It sounds intuitively correct to me, although I think Willis was describing the tropics in particular. It may be time for him to update this theory and expand it to elsewhere.
Dave – heat flow is proportional to the DIFFERENCE in temperature between the warm and cold sides T1 & T2. If the black roof raises the temperature of the outside of the insulation by 2 degrees, (T1 is inside the house, T2 is inside the attic with the insulation being on the floor of the attic) – that’s a 2/R reduction of heat flow.
With regard to cooling at night, I note that (having worked on selective surfaces in the 70s when this was a big flap) emissivity in the infrared can be can be quite different than in the visible. Thus dark-skinned people (cleverly) appear white in UV light and are less susceptible to skin damage near the equator. So to get a net solar gain in winter you need a paint black in visible light and white in the infrared. In summer you need the opposite to reduce AC load. Both are hard to get as the on and off work on selective surfaces continuing for some 40 years shows.
But … but … but …
Gavin has spent decades … decades … decades … refining his models …
Surely anything else is inferior …
So what happens when we move away from coal fired electrical generation and to mainly electric or H2 cars. Will we get rid of enough atmospheric black carbon to reverse this effect?
So lets put water pools on our roofs and cool the planet down and maybe even get rain in Texas.
richardjamestelford: The feedbacks you get so excited about here are the sort you would dismiss as third order effects if they were in the other direction.
That is correct. My question to Willis is, do you still believe that third-order effects are too small to care about?
One way or another, doubling CO2 in the next 70 years will produce a third-order effect, but it probably makes a great difference to us whether the net effect is to raise or lower global mean surface temperature from what it is now, or have no effect at all.
Two comments:
First – I really wonder whether much vaunted Nobel Prize winner Sec.Steven Chu is going to be fired or admonished by Obama for peddling the “paint roofs white” concept to combat “global warming”. Solyndra then this added to much else – I don’t think we have ever had a worse science adviser for a President.
Second – I looked twice at the PDF describing GATOR and could find no references to incoming solar radiation or other than ‘variation in earth-solar distance’ … Since it is completely clear to me that at least half (and likely more than 80%) of temperature change on Earth is due to variation in Sol, GATOR&MOM is incapable of being used to explain temperature changes here.
If anyone here has noticed indication somewhere that GATOR actually does include solar variation/ irradiance variation/ comic ray, etc., in some sneaky less defined way, I would appreciate them flagging me as to the page of the PDF that is described.
AFPhys
blockquote>Gary Pearse says:
So lets put water pools on our roofs and cool the planet down and maybe even get rain in Texas.
I have a better way.
First, you need nuclear power, and plenty of it.
Yes, nuclear power, safe and cheap.
Yes, that same nuclear power that the press is constantly saying is sooo dangerous.
Yes, that same nuclear power that, despite being sooo dangerous, and despite being available in great quantities in Japan, where they had a huge earthquake and a tsunami, has still killed or even injured exactly zero people.
Just keep telling yourself, number of people killed in Japan by radiation, still zero…
You can ignore the recent newspaper article that spoke of radiation “hot spots” found in some Japanese city, since it admitted that the radiation was so small that it was harmless to anyone, which means that the spots were not “hot”, and which thus showed that the entire article was nothing more than a bald faced lie. I guess “cold spots” just doesn’t have the same punch to it.
Now that you have all that safe, cheap power (also useful for air conditioners…), you can build a desalination plant or two or a dozen. Now that massive sucking sound you hear from the cities drinking up all the water the farmers need goes away. Now not only do you not need to ship water to the cities, which the farmers can use, you can even pipe the water to the farmers directly. And now all that water means the farmer farm, and all that electricity means the factories produce, and the local economy booms.
To sum up:
You live in Texas.
Texas is a desert.
Droughts are frequent in Texas.
Texas is next to an ocean.
Nuclear power and desalination plants exist.
Nuclear power is cheap.
Japan has shown that nuclear power, even from old power plants, is safe even under the most extreme conditions.
Put the two together and your water problems go away.
Conclusion, you need to take you local politicians, and your local mainstream press, and run them out of town on a rail. Make sure you bring plenty of tar and feathers.
Uh, make them white feathers.
About white roofs –
Good in the summer to keep attic temperatures down.
Good in the winter, too, as opposed to black roofs. Black is a good absorber of sunlight in the winter, but it is also a good radiator of heat for the great majority of the winter day when the sun isn’t up. White, on the other hand, impedes the radiative transfer of heat, summer and winter, definitely a color for all seasons.
Painting roofs white would provide lots of green jobs, tho they would likely be taken by the illegals who are already doing most of the re-roofing work in America. It would likely also void the manufacturer’s warranty on your shingles.
Septic Matthew says:
October 21, 2011 at 12:11 pm
The effects in the white roof case are enough to swing the effect of the white roofs from a strong net cooling (see what was expected in Figure 1) to a weak net warming. That is a very large swing.
Richard and Matthew, how on earth you can mistake that for a third order effect is beyond me. It’s a large enough effect to put the White Roof Project entirely out of business … good luck convincing them it’s only a third order effect …
Nobody, and I mean nobody, predicted this outcome or anything like it … and now you want to pooh-pooh it by calling it a third order effect?
That spells A-G-E-N-D-A to me, it let’s me know you’ve got one.
w.
AFPhys says:
October 21, 2011 at 1:17 pm (Edit)
I doubt it. GATOR, as far as I know, is used for short-time analysis (some years or a decade) of problems like the one above. Since the future unraveling of the sun’s output is not known, it’s not possible to include it (other than in the most general way).
So for the problems GATOR gets used for, I don’t think solar variations are meaningful, and for predictions of the future (that Mark Jacobson seems to steer away from) it can’t be predicted.
w.