Naked Science Forum
Life Sciences => The Environment => Topic started by: neilep on 31/01/2008 02:05:41
-
Wocha Peeps !
Nice day for it eh ?....I was out this evening seeing a client...it's late Jan ..the temperature was 4 degrees celsius and I kid you not ......a bloomin ice cream van came round the corner...this was at 6:30pm and all completely dark !!...the thing is...he actually had a customer...then off he went...and I giggled as he reappeared at the other end of the neighborhood 5 minutes later!..I couldn't believe it..I have never seen an Ice Cream van operating in the winter !!......he must use a climate model !!
So, how does a climate model get created ?...do they just use the last few hundred years of weather records and data and average it all out or is it a tad more complicated than that ?...and how accurate are they ?
........ Presumably these things are 'live ' as they are constantly receiving data and therefore are always updating yes ?....could I make one(if I knew how which I don't) with my PC ?....surely it doesn't require more computational power than a ZX81 ?
I am serious about the ice cream van !!
*hmmm..does this belong in Technology section ?*
-
Perhaps he wasn't selling Ice cream in the dark middle of winter eh??????
-
I think this is okay for Environment
Basically, you make a weather simulator (if windspeed=x, direction=y, then Z happens - but about a millionfold more complicated) and then plug real data into it. If the outcome of your 'simulator' consistently matches the observed changes in climate over a given timescale, you have a working climate model. If it fails to predict what you already know to have happened, then you need to tweak your model.
When your simulator can accurately predict past changes based on past data, you can assume that it may accurately predict future events based on current data.
-
When your simulator can accurately predict past changes based on past data, you can assume that it may accurately predict future events based on current data.
Only if the future environment is sufficiently similar to the past environment you were looking at.
-
The first thing you need to do is buy 6 of the fastest, most capable Cray computers ever made. Then get an army of programmers, assign all of the programmers a single variable, have an army of meteorologist supply answers as needed and about twenty (20) years to get it all put together, with revisions to the software and hardware, and you will get a first approximation. A second approximation may take another 20 years.
-
Depends what you want to do with it. For some purposes a zero dimensional energy balance model may be suitable and then you may, with a lot of work, be able to produce one that doesn't take months to run. The radiative effects of greenhose gases are well known and there isn't all that much more to it. If you want to look at impacts then you need a high resolution model and that army of programmers and meterologists for 20 years.
Climate models generally aren't live. You will find many using 1990 or 2000 conditions as their starting point as the scientists want to improve comparibility with other modeling efforts.
If the results were sensitive enough to initial conditions that data from a year later caused significantly different climates then you might as well give up on climate models. Fortunately climate unlike chaotic weather is resonably stable making climate modeling efforts possible.
How accurate are they?
Again it depends what you want to know. For global average temperature, they tend to do pretty well. For other variables and regional impacts - hmm bit dodgy different models can predict different things and then which model do you decide to trust? May improve with higher resolution, bit more moore law....
If you want to know about weather events forget it.
The IPCC range of 1.4-5.8C warming this century from IPCC 3rd assesment report may seem a bit wide but about half of that variation depends on the scenario - we don't yet know what greenhouse emisions and other climate forcing agents will do and the other half is uncertainty because the models are not perfect and you don't know which model to trust.
Whether you regard this accuracy of getting the model uncertainty down to the same level as the scenario uncertainty as good or bad probably depends more on whether you want to say the models are good or the models are just about useless.
A lot of modeling effort was done to prepare for the 4th assessment report. And was all this modeling useful to reduce the 1.4-5.8C range down to 2-5.8C? Err well no all the models were too similar and the scientists didn't believe the models were that good. So in the end the revised range came from a simple energy balance model that runs on a laptop in a few seconds rather than the models that take months on supercomputers.
-
So, how does a climate model get created ?...do they just use the last few hundred years of weather records and data and average it all out or is it a tad more complicated than that ?...and how accurate are they ?
Well, surely you've heard of 'a storm in a teacup'?!
-
Climate models, much like other models, plug a lot of historical data into a learning algorithm of some kind, and out pops a model.
I think the details of most climate models are too complex, but in general they use the same modeling algorithms that are used for most other machine learning applications. Search for "machine learning," if you are interested.
As to how accurate they are, I have no idea. They probably say in the scientific literature. The most common way to test the accuracy of your models is to train your model on one set historical data and then test the model to see how well it predicts another set of historical data. However, there's a problem with most modeling when it comes to climate change.
Man-made carbon emissions is a relatively recent development, and if global warming is dynamic and is reaching a tipping point or more broadly, if global warming is occurring at a rate which has not been seen in past climate data, then you can't be that sure how well your climate models work. You are modeling something which you have not seen data for. So, you're really left with a best educated guess on how accurate your climate model is.
-----
http://climatepact.org
-
A climate model uses data to predict future events but as they have not as of yet been able to even remotely predict anything correctly the answer to your question is no one knows. [:)]