glenncz said:
>considering the information you are giving me, you sound more like a climate scientist "hobbyist", because a $real$ climate (rush limbaugh cough cough) $scientist$ can't possibly think that way and get paid by anyone, not in this day and $age$. must be retired.
I did retire in January, but not from climate science. Even though I have a Masters in Operations Research and Statistics, I spent most of my career trying to get computers to do more than they could. It seems crazy that computers are now over a billion times faster than when I got interested in computer modeling, but they still are not fast enough.
It works like this. Faster computers (or better algorithms, usually my part of the picture) means that you can do more modeling at a lower price. As the price drops demand rises. However, people like me (with my statistic hat on) keep finding "interesting anomalies or features" in the data. So run the model a few more times, probably at higher resolution.
When modeling a car crash or bending an aircraft wing, someone has the job of telling me to put my concerns in a memo, and consider the job finished. (I've worn that hat too, just don't try to wear both at the same time--you get ulcers.) For well behaved models, that process ends pretty quickly. It is very unusual to have issues that can't be nailed down by tweaking the model and running it again.
As for climate modeling, and atmospheric modeling in general, the big problem can be stated succinctly. You want to model four dimensions, latitude, longitude, height above sea level, and time. It is possible to tile a full Earth model so points don't get close together at the poles and far apart at the equator. But that saves you less than a factor of two. Let's start with modeling every fifteen minutes of latitude and longitude, and twenty points between the surface and the stratosphere. About as rough as you can expect any results at all from. So 360x4x180x4x20, just over 20 million points per iteration. How about 5 minutes for the time step? You might be able to do it today on a desktop computer--OpenCL and DirectCompute allow you to use your high-end video card to crunch the numbers. (But you will need help from me or some other expert to flow the data through the computer in the best order to keep it crunching data not moving it around.)
In any case, you might be able to model 5 minutes of real time in say two minutes. NOAA has worked with models like that where to predict three days forward from today, you started the model running the day before yesterday.
Go to a model about five times finer in every dimension, and you should get some really good climate data. (Assuming the math of your model is right--that is the easy part.) But we just made the model run 3125 times slower, and it would be nice to model a decade or three to get useful data. Maybe in another ten years, but not happening today.
So why are the current models not good enough? Unfortunately the big 64 dollar question is how do clouds and CO2 interact? Heat transport from the surface to the stratosphere occurs through radiation and convection. If you bias the models in a non-realistic manner, say no radiative transport, convection increases to take up most of the slack. But the models also do strange things like moving the snow cover line way south. So when you try to relate the models back to reality, you need "fudge" factors to reflect what you really learned. You also try to put brackets on the model results rather than looking at one "perfect" run.
As a result anyone who argues that the net effect of CO2 is to trap heat either fell asleep during the discussion, or has a bridge they are trying to sell you. In any decent model I have seen it traps heat on Tuesday but not on Wednesday...
