If You Can't Explain It, You Can't Model It

Tamino thinks that the models forecast rapid warming since 2002.
http://tamino.files.wordpress.com/2008/03/rahmstorf2.jpg

sod (14:33:18) :
http://www.realclimate.org/images/Hansen06_fig2.jpg
but you knew that, of course?!?
But scenario’s A & B are the ones promoted. Perhaps we should dismiss those scenarios for scenario C, seeing as they are currently, note, currently, closest to reality.

Well, call me simple minded or whatever, but when the amount of experience, time, money and people involved in near-time weather forecasting cannot accurately predict what the conditions are going to be 5 days from now, why should I trust our Climate experts to be accurate 10 – 20 – 50 – 100 years from now? I know a lot of the experts claim that Weather forecasting and Climate Forecasting are different things entirely. Well, Math and Physics are also separate, but if you can’t do the math accurately you aren’t going to answer many physics questions.

Question for Anthony possibley concerning models. I believe that it was almost a year ago that you published a letter regarding the Aqua Satellite results. As I recall, they were not being released pending further peer review since the findings were at odds with current thinking [models?]. Any update on the status?

JohnG, In my experience giant code is often not “just bloat” — it’s hacking. Like for instance “if the surface of the ocean at the tropics freezes force fix the data…” That’s the kind of thing I’d be looking for if I audited the code. I agree that the code really *should* be simple to have any hope of validity.

I have always had a few questions regarding the modeling processes used by the IPCC, and although these questions are argumentative in that they reveal my opinion about the usefuness of climate models, perhaps someone here is in a position to address the factual bases for these questions or explain why the questions might be misplaced.
Based on a very cursory reading of the latest IPCC report, it is my understanding that the conclusion of a causal effect between GHGs and warming is premised in part on a finding that exising climate models cannot simultate current climate trends without a substantial CO2 warming effect. Due to the chaotic nature of the climate system, and the unknown nature of what initial conditions to use in the models, a model is “run” many times with different intial conditions to produce a range of climate trends, and the reasoning is that, because none of these model runs show the existing warming trend without the modeled CO2 sensitivity, then it is reasonable to conclude that the existing warming trend in the real world is caused by CO2.
My first question is whether I have fairly summarized both the modeling process of the IPCC, and the reasoning behind it’s finding of a causal effect between CO2 and warming; because if so, this logic seems flawed. All this exercise demonstrates is that the models used by the IPCC require a substantial CO2 contribution to temperature in order to reasonably fit known climate trends. Nothing more. It certainly does not demonstrate that NO climate model could be constructed that both reasonably fits known climate data and uses an insignificant sensitivity to CO2. In other words the premise that models A, B, and C cannot simulate climate trends without relationship 1 does not logically support the conclusion that relationship 1 holds true in the real world system that the models A, B, and C attempt to simulate. Models A, B, and C may be simply wrongly constructed in one or more respects.
This leads into the second question. Has anyone actually tried to construct a model that has little or no CO2 effect, and then spend as much effort tweaking it to best fit past and current climate, as has been spent tweaking climate models that DO assume substantial warming.
My last two questions are a little more general. What possibile utility is there in modeling a relationship that cannot be physically measured, either directly or indirectly, e.g. the temperature response of the Earth as a function of CO2 concentration? If you don’t have the ability to measure a physical system at a level of detail that you need to perform a task, why would someone think that this deficiency could be overcome by simply modeling the system at the desired level of detail and testing the model against the things you can measure?
Finally, given that a computer does nothing more than what the programmer instructed it to do, how can merely running the computer model tell you something that you did not already know or assume about the system modeled? I can accept that a computer model can be highly useful as a shortcut in performing tasks too complex to be reliably done manually, once it has first been established that the computer model is accurate with respect to the specific task at hand. I have a much harder time believing that a computer model can TEACH you anything about the system modeled, particularly with respect to modeled relationships that can’t be independently verified apart from the simulations.

John Philip,
Given the seven year drop in temperatures, I take it that you are predicting a rapid rise over the next year or two to get back on the scenario A curve?

“… it would not get much better until they had supercomputers 1,000 times more powerful than at present.”
The “we just need more computing power” line is both tired and false. Increased computational power does not convey increased understanding.

I wonder what the systemic error of the model is given the very large elements they use and what is the systemic error due to feedback of the errors?
My guess is that its very, very large and the feedback completely dwarfs any meaningful analysis.
I know the current numerical models used by NWS and others suffer from tremendous feedback issues.

“when . . . people involved in near-time weather forecasting cannot accurately predict what the conditions are going to be 5 days from now, why should I trust our Climate experts to be accurate 10 – 20 – 50 – 100 years from now? I know a lot of the experts claim that Weather forecasting and Climate Forecasting are different things entirely.”
Lets say that I have a very weak, repetitive, audio signal that I’m sending across the country. On it’s path, that audio signal is subject to quite a bit of random noise. When received, over short intervals the signal is going to represent mostly the random spurrious noise, and reconstructing the repeating pattern will be almost impossible. With sufficient time, however, the original signal can be reconstructed because it’s pattern holds over time while the random noise cancels itself out. Once reconstructed, you can acurately predict the up and down trends of the signal so long as they are avereraged over a sufficient interval, but accurately predicting any instantanous point will be impossible.
Notice in this example, however, that the source of the signal and the source of the random noise are mathematically uncorrelated. This is what permits you to separate the short term random noise from the long term repeating signal. The same physical phenonena that produces weather however, produces climate. In fact, weather is nothing more than the instantaneous manifestation of climate. Those who argue that long term climate is predictable even though short term weather is not, are essentially making an unstated assumption that the chaotic nature of the earth’s climate system fades out over frequencies longer than four decades or so.
Which side of the argument you take depends on how reasonable you think this assumption is. I think it’s a lousy assumption. For example, when I look at long-term climate reconstructions, I don’t see any pattern in the various climate optimums and minimums. The standard deviation for the length of ice ages/interglacial periods, from eyeballing these reconstructions, also seems a lot longer than four or five decades. All of this seems to indicate to me that there is a lot of randomness in the Earth’s climate system that is not removed by taking averages over a period of decades.

The less you know about something the easier it is to model.
That’s very encouraging. Now that my solar system orbital simulation model is working tolerably well, and I have a planet Earth in it spinning once every sidereal day, and orbiting just about once every year, I’m thinking that I may extend it to include the terrestrial atmosphere, which I’ll divide up into layers in some sort of grid. Knowing the thermal conductivity and heat capacity of air, sea, and land, and the incident solar radiation at any point on the planet surface, and the heat losss from the planet interior, I’ll be able to work out the conductive heat flow vertically to outer space at absolute zero, and horizontally through the atmosphere and ground.
I’m not sure at the moment what to do about air flows using this grid system. I’m wondering if, instead of having a fixed rectangular grid, I might have mobile air masses which would ‘jostle’ with each other as they expanded and contracted, no doubt forming warm and cold “fronts” surging to and fro, and spinning clockwise and anticlockwise under the influence of coriolis forces, and very likely northern and southern “jet streams” as well.
That should all be working by next Thursday. Then I’ll add in clouds and precipitation. I’ve heard this is difficult to do, but I can’t see any big problem there. As my mobile air masses jostle with each other, the warmer lighter humid ones will float to the top of the atmosphere, and will turn white as the water in them condenses into clouds and rain. I expect I’ll have cirrus and stratus and cumulus clouds appearing shortly after that. And I’ll be able to make weather predictions for the next few days, although obviously not quite as accurate as the Met Office’s (or perhaps more so?).
It should be early April by the time that’s working. Looking at climate over a long period of time should be straightforward. I’ll just have to leave the laptop running all night under my bad where Twinkle can’t find it and started treading on the keyboard like she enjoys doing when I’m working on it.
Mid-April now. And then there’ll be the final tweak to get the CO2 and global warming working, maybe just by decreasing air conductivity a bit, or boosting solar insolation by the same fraction.
Finally, in early May, I’ll release the results of my GCM to the world in a series of increasngly panic-stricken press releases that will describe how all my mobile air masses have been jostling and bumping into each other to produce the entirely new and hitherto unforeseen phenomenon of Global Swelling, which is what happens when warm sticky air masses form a “traffic jam” around the equator, and pile up higher and higher on top of each other, before rolling downhill towards the poles again, very likely destroying entire civilizations in the process – unless governments Do Something About It Before It’s Too Late – like buy me a newer and faster laptop with WiFi and BlueTooth and all the other latest fancy trimmings.

My first question is whether I have fairly summarized both the modeling process of the IPCC, and the reasoning behind it’s finding of a causal effect between CO2 and warming;
You have.
This leads into the second question. Has anyone actually tried to construct a model that has little or no CO2 effect, and then spend as much effort tweaking it to best fit past and current climate, as has been spent tweaking climate models that DO assume substantial warming.
All attempts to produce numerical predictions of future climate are models. I think what you are asking is, are there Global Climate (Circulation) Models that have little or no CO2 effect?
The simple answer is we don’t know, because models are proprietary and only selected model runs are published.
However, with reference to your question 1, this means the IPCC’s case is built on an unverified and unscientific assumption. Namely, the published runs are representative of all possible runs.
What possibile utility is there in modeling a relationship that cannot be physically measured, either directly or indirectly, e.g. the temperature response of the Earth as a function of CO2 concentration? If you don’t have the ability to measure a physical system at a level of detail that you need to perform a task, why would someone think that this deficiency could be overcome by simply modeling the system at the desired level of detail and testing the model against the things you can measure?
The IPCC used to be upfront about the fact the models are wildly wrong on almost every climate metric except temperature. They then bizzarely concluded we can trust the model’s temperature predictions.
Anyone can see that the reason they get temperature right (when hindcasting known data) and all other values wrong, is the models are tuned to the temperature data.
Finally, given that a computer does nothing more than what the programmer instructed it to do, how can merely running the computer model tell you something that you did not already know or assume about the system modeled?
You are correct, a computer model doesn’t tell you anything you don’t already know or had assumed.
How the computer models work, is they iteratively process datasets with the output of one iteration becoming the input of the next iteration. Each iteration being a time period.
What computer models allow you to do in theory is to produce predictions of much greater accuracy.
There is no empirical evidence that climate models do in fact produce more accurate predictions than for example, drawing a line on sheet of graph paper or the computer equivalent.
The IPCC’s belief in the temperature predictions of the climate models is purely faith without any scientific basis.

I am curious as to whether one of the modelers has recalibrated his model to the now 10 years of stagnation in warming.
If this has been done, I’ve not heard about it.
If not, then why not? (Other than, of course, it would mean the sensitivity of CO2 would need to be cranked way down!)