Sunday, March 02, 2008

Stick That In Your Climate Model...

...and fine-tune it.

Apparently, it's been know for awhile now that airborne bacterial streams could prompt precipitation in the form of rain or snow. The single-celled bodies which make up these streams serve as "nucleators" that water vapor can coalesce and freeze around. However, how ubiquitous these streams were, and how important they might be to cloud formation, has only recently become apparent:

"Atmospheric scientists haven't previously recognized that these particles are so widely distributed," [Louisiana State University microbiologist Brent Christner] said.

The findings raise the question of how climate change and human activities will affect bacterial balances in the sky. More immediately, they're a starting point for research on bacterial contributions to cloud formation and precipitation.

Since the impact of feedback loops involving clouds on global weather patterns is the largest source of uncertainty in current predictions of climate change, the new research should eventually allow for greater precision in these forecasts.

More cool cloud stuff here.

27 comments:

Ti-Guy said...

From your link:

"...and Christner's study found that bacteria are the most common warm-temperature nucleators of all."

I didn't know that. I wonder what percentage of nucleators they comprise?

Dante said...

So I suppose an injection of acid rain should cool things right off by killing the bacteria. Better add that variable too.

I am surprised that you haven't caught on to the fact that complex systems are not "closed systems". Classical statistical error is assuming a variable change will not materially change the statistical population.

Keep adding variables and you may end up screwing the model more...not reducing the error of certainty.

A model based on probabilities is doomed to fail if one can not even demonstrate historical impirical evidence that proves correlation instead of causation. This means explaining the times that rising CO2 levels lagged rising temperatures. I written many statistical models and I know what I'm talking about.

bigcitylib said...
This comment has been removed by the author.
Anonymous said...

Dante,

Shhhhh, church is in session. They believe because they're told to believe, not because it makes sense.

bigcitylib said...

Dante, you talk Science the way I talk Norwegian.

Anonymous said...

Not only do the AGW computer models invlove a significant amount of hypothesizing (unproven guesswork) about the magnitude and even direction of the variables,

they admittedly fail to employ many variables which should effect the model, but they simply do not know much about.

What's really funny is that not only do the "believers" not appreciate the inherent unreliability of models which involve the number and complexity of variables like the AGW model,

they actually think it is immutable.

There's a reason why computer models can't predict earthquakes, tornadoes, or even what the weather will be like in any location within a week's time.

But again, thinking this through would require rational thought, and as we all know religions are based on faith, not rational thought.

Dante said...

Science=study

Science <> constructing reality.

BCL. Try working in a quantitative field in the Private sector. You might get past mean.

BTW. How many predictive models have you written and how far past "Proc report" have you reached in SAS ?(please let me know if I am now speaking Chinese to you)

bigcitylib said...

Dante,

I once predicted that the Leafs would win the Stanley cup. But I didn't write it, I just said it in a bar. People said I was Mad.

Ti-Guy said...

I am surprised that you haven't caught on to the fact that complex systems are not "closed systems".

How did you...scientifically...come to make this conclusion?

Classical statistical error is assuming a variable change will not materially change the statistical population.

Where exactly is this error being manifested?

Keep adding variables and you may end up screwing the model more...not reducing the error of certainty.

Whaa? If the data being plugged into the variables is based on rigorous and reliable measurement, all things being equal, the predictive model becomes more reliable, not less.

I think this is where you slipped into Norwegian, which I understand with only great difficulty.

BCL. Try working in a quantitative field in the Private sector. You might get past mean.

What's a "quantitative field in the private sector?" And (just to throw this out) if you know anything about statistical science, quantitative and qualitative analysis are both perfectly legitimate and scientifically-rigorous treatments of data we gather to measure the world.

Admit it, Dante...you overreached with your comments here. Common behaviour these days, it seems.

Dante said...

---Where exactly is this error being manifested?----

It assumes no feedback loop that effects other variables. As an example look at a relatively simple system such as Bank credit losses where there are clear variables associated with the probability of default. If the bank decided to limit future business to the top decile, it will find that the outcome would not match what rigorous multi-variable regressions had predicted. Why? Because in doing so, it neglected to understand that the subsequently low approval rates, changed the behaviour of the people in the branches who then decided it would be better to focus on selling a different product. This behaviour change effectively changes the population and that population change changes the outcome.
So lets now model in behaviour of branch people. Don't forget to factor in education levels, demographic areas, working hours etc. Each may represent an incremental improvement in the probability for a sample population but increases the point of failure when one variable changes significantly and screws up the whole thing. All variables would then need to be recalibrated because the model is no longer predictive.



---What's a "quantitative field in the private sector?"---

How about credit risk management, treasury management and data mining for a few of examples.

---And (just to throw this out) if you know anything about statistical science, quantitative and qualitative analysis are both perfectly legitimate and scientifically-rigorous treatments of data we gather to measure the world.---

So if I understand you, climate models produces qualitative output.
I guess the real news is that climate scientists are now using artificial intelligence systems to run their "models".

Have you a clue what you are talking about?

Holly Stick said...

So dante thinks that if he writes enough dumb posts denying AGW that the laws of physics will change their behaviour and decide to produce a different product and everything will be alright. It works for banking, doesn't it?

"...This means explaining the times that rising CO2 levels lagged rising temperatures..." dante, real scientitst know aboutCO2 lagging rising temperatures, and they have explained how this happened. Get your Mommy to help you with the big words here:

http://www.realclimate.org/index.php/archives/2004/12/co2-in-ice-cores/

"...I written many statistical models and I know what I'm talking about." I think not.

bigcitylib said...

Dante,

It assumes all sorts of feedback loops.

But that aside, would you then argue that eonomic models such as the one you talk about can never be predictive?

Dante said...

First of all Holly, I won't begin to explain to you the difference between correlation and causation. If you are interested, real stats 101.

BCL. I wasn't talking about an economic model. I was talking about a statistical model. In relatively simple systems, statistics work quite well though they do need calibration every few years and the variable often don't exceed a dozen or so variables. Complex systems often called chaotic systems are extremely complex. Weather is one of the most complex systems that I can imagine.

Holly Stick said...

Dante, start by learning the difference between weather and climate.

Ti-Guy said...

It assumes no feedback loop that effects other variables. As an example look at a relatively simple system such as Bank credit losses...

I stopped there. Where, in the physical phenomenon of climate change...in the introduction of this variable (the role of biological nucleators in cloud formation)...is the manifestation of the error in assuming a variable change will not materially change the statistical population?

Don't start talking about finances and banking...the statistical models of which rely on some arguable assumptions to begin with and have to account for the decisions of intelligent agents.

So if I understand you, climate models produces qualitative output.

No...you were throwing around the word "quantitative" in a way that suggested you don't know what you're talking about. Qualitative analysis uses math and statistics; it just is based on qualitative measurement, which is the only way to measure certain things.

...although I have heard of economists working on quantifying "love" recently, so who knows what's going on with some people these days.

Ti-Guy said...

Weather is one of the most complex systems that I can imagine.

Oh, I see the problem here. Yes, it is complex, but it's based on physical phenomena that can be more reliably measured than other types of phenomena that other statistical models have to account for...like economics.

That's why economists get everything wrong all the time. They think what they're doing is science, like physics and biology.

Poor dears...

Dante said...

Holly...move on. You are embarrassing yourself. If you want to argue semantics, try your highschool classmates.

Ti-Guy
What it means is that introducing a variable with no understanding of it inter-relation with other variables could skew results significantly. The variable likely does not operate in isolation. It's mere introduction makes any model much more complex. it's movement will likely change other variables, requiring further calibration.
My point is that this argument will not be advanced with a "simple addition" of one variable. Let the scientists try though. If the science is settled, it shouldn't take too long.

Ti-Guy said...

So dante thinks that if he writes enough dumb posts denying AGW that the laws of physics will change their behaviour and decide to produce a different product and everything will be alright. It works for banking, doesn't it?

Just goes to show what's driving a lot of this discussion...the fear of economic re-structuring/re-ordering and the psychological dimensions of capitalist economics.

It's a legitimate concern for the financial economy, but it shouldn't be used to confuse or distract from the science underpinning climate and climate change.

And Dante...trying to make people appear stupid because they're not familiar with your narrow expertise and jargon is really weak.

Ti-Guy said...

My point is that this argument will not be advanced with a "simple addition" of one variable. Let the scientists try though. If the science is settled, it shouldn't take too long.

That's all we should expect.

Holly Stick said...

Dante, the fact that you came up with that tired old argument about CO2 lagging rising temperatures shows that either you are too ignorant and too lazy to find out what the climate scientists are saying, or you think that we are dumb enough to believe any old shit.

I neither know nor care about statistical wankery, but I recognize bullshit when a rightwinger produces it. And I see you did not bother to read the link I provided.

If you honestly want to know how climate modelling developed and how it works, read this:

http://www.aip.org/history/climate/GCM.htm

As for your comment: "...What it means is that introducing a variable with no understanding of it inter-relation with other variables could skew results significantly..."

As Ti-guy pointed out, we are talking about science here, not people's economic behaviour. Scientists know how atmospheric CO2 works because they do experiments and find out how it works. They know that water vapour amplifies the greenhouse effect of CO2, because they've done experiments on that too.

Dante said...

---Scientists know how atmospheric CO2 works because they do experiments and find out how it works. They know that water vapour amplifies the greenhouse effect of CO2, because they've done experiments on that too.---

I can post links too Holly like:

http://www.dailymail.co.uk/pages/live/articles/technology/technology.html?in_article_id=440049&in_page_id=1965

Perhaps you are too lazy to research alternative opinions.

In any event, you have nothing to contribute to this argument that any idiot hasn't read yet.

As for "statistical wankery"...the only other thing that these models could be built on is horseshit but i will at least give the builder the benefit of the doubt.

Holly Stick said...

If you think that's a good reference you are an idiot. Sorry, I thought you were a liar, but you must be just stupid.

John Mashey said...

Complex models are not needed to know that the average temperature of the Earth *as a whole* is getting warmer, and will continue to do so.

If you are slowly filling a bathtub, and are trying to measure its depth, it's harder when there's a kid in there splashing around.

But the overall bathtub is still filling up.

All one needs to know is:
1. Incoming solar radiation is higher than outgoing radiation, which by the First Law of Thermodynamics, means the total heat energy of the Earth is rising.

2. So, where does the heat go?
A: into the oceans, mostly, since the ocean's heat capacity is about 1000X higher than that of the atmosphere.

3. Over the long-term, energy flows from hot to cold [Second Law of Thermodynamics], but in the short term, there are lots of oscillations, like El Nino, where the ocean gives up a lot of energy to the atmosphere, and the *surface* temperature spikes. All this temperature jiggling that bothers people is because the surface temperature only measures a small slice of the Earth. It happens to be an important slice, but over time, and on the average, it goes up along with the ocean, which is rather more than the sea-surface.
====
Do scientists understand this? OF COURSE.

IPCC AR4 Chapter 5 is good start.

OR
http://pubs.giss.nasa.gov/docs/2005/2005_Hansen_etal_1.pdf

Dante makes the common error of thinking physics-based models work like statistical analysis and prediction of things having no physics, like economics or stock markets.

Dante said...

John.
The fact that this model runs on a super computer indicates some some sort of neural net employing Bayesian statistics. Don't begin to tell everyone that statistics operate differently in physics models. It is a simply not true.

Anonymous said...

Wait a minute - just yesterday you were defending climate science as all-knowing and settled, that the climate models were perfect. NOW you're telling us something NEW has come up?

Anonymous said...

Ice core records have shown that around the time of the Younger Dryas, temperatures rose as much as 7ÂșC in about 20 years or so. No known reason.

Stick that in your climate model as a natural variable.

John Mashey said...

I wrote an earlier post, but it seems to have disappeared, so I'll try again.

Dante:
I'm sorry, but you are either suffering from Dunning-Kruger Effect, http://en.wikipedia.org/wiki/Dunning-Kruger_effect, or else you are purposefully trying to confuse the folks here.

You can look me up, via Google & Google Scholar.

I used to be Chief Scientist at Silicon graphics, I've helped architect a bunch of high-performance chips, software and supercomputers [SGI Origin 3000, Altix, some work on Origin 2000 and earlier SGI SMPs.]

I also helped sell them to senior scientists and engineers worldwide, which meant spending a lot of time with them to understand their problems, algorithms, scaleability issues, and technology trend matches.

I worked with:

-climate scientists at NCAR, NASA, GFDL, and various universities.

- seismic & reservoir sim people at many of the big oil companies worldwide

- fluid dynamics folks at NASA, Boeing, the car companies, Mattell [Barbie dolls], Unilever [chocolate ice cream bars], and many others.

- structural dynamics folks at the car companies, i.e., for crash codes

- Wall Street risk management simulation folks

- simulation folks for molecular modeling, earthquakes, various other sorts of physics, at US national Research Labs, universities, and most of the big pharma cos.

- Relevant top software companies like Landmark Graphics, MacNeil Schwendler, and one you mentioned, SAS. These folks all do very different things.

- Although half-retired, I still give invited talks on statistical analysis methods at places like Stanford, Princeton, Cambridge, etc.

- And of course, besides the PhD in Computer Science, and BS in Math, I was one course of a second BS in Physics, took courses in statistics and operations research, stochastic processes, and later worked at Bell Labs, which had people like John Tukey to learn from.

"The fact that this model runs on a super computer indicates some some sort of neural net employing Bayesian statistics. Don't begin to tell everyone that statistics operate differently in physics models. It is a simply not true."

This is not only wrong, but doesn't even make any sense.

The fact that a model runs on a supercomputer means absolutely nothing about what sort of model it is, and in any case, most supercomputer apps aren't that kind.

Anyone who can deliver utter nonsense with total certainty ... Dunning-Kruger all the way, but they did find that people could learn, so there is hope.

We had a similar discussion over in RealClimate a while back, with a poster who was certain that climate sims couldn't be useful, because he was used to the difficulties of protein-folding (yet another different kind of sim).
See http://www.realclimate.org/index.php?p=498#comment-68087
for the comment that finally fixed this, in passing describing the wide variety of sorts of simulations.