07-15-2009, 10:15 PM
Hi,
In the late '90s, I revisited the field in conjunction with a satellite program we were developing. There had been some improvement of the models in the seven or so intervening years (and some of the code had actually been converted to C or C++ :rolleyes:) but the state of the art was not much different -- ad hoc models with enough adjustable parameters to predict an elephant if the modeler desired. Since then, my interest has been strictly that of an informed outsider. To paraphrase Will Rogers, all I knew is what I read in Science News. Furthermore, the past 4 years I've even let that slip quite a bit -- part of that time I did not have the ability to understand topics at that level, and much of the rest of that time I was too weak to care. It has only been in the past year, and especially the past four months, that I've had the mental and physical capacity to again take interest in scientific matters.
So, having thus established my (lack of) credentials, take what follows as you will.
I have not seen the most recent models. Nor have I seen any announcements of a major breakthrough in climatic models. Since there does exist an ongoing debate about the validity of the models used, I would expect any substantial improvements to be widely touted in support of the modeler's claims. That leads me to believe that the models are probably much the same as they were ten years ago -- which is to say, not that much different than they were thirty years ago when they were first able to 'predict' global warming.
Most of the existing models of ten years ago could do a sufficient job of postdicting any given historical period with an appropriate choice of parameters. The problem was that it would take a different set of parameters to postdict a different period. Thus, in essence, the models were little more than a curve fitting program. Each historic era was fit with one set of parameters, and then the next with another, and so on much like a cubic spline is used to fit experimental data. And just like a cubic spline does not 'explain' the data, neither did these models.
For a model to be valid, it does indeed need "to predict the entire global climate past". If it cannot, if it can only predict some eras but not others, then how are we to know that the coming era is one which it can predict? How can we know when we can use it and when we can't?
As long as there is debate over the models and parameters needed to postdict the existing data, so long will the models be useless to predict the future. From what I've seen published, that debate still runs hot. I'd love to go to a climatologist convention and see how many "my honorable colleague is a damned fool" arguments break out. Hell, I'd even buy the pizza and beer.:)
--Pete
Quote:Does hindcasting perform so poorly? My understanding, again self-admittedly inexpert, is that current models actually do quite well at predicting past temperature. Asking a single model to predict the entire global climate past is probably not possible or realistic, but doing things like predicting 20th century temperatures seems to yield solid results.My ex-spurt-eas is from the late '80s and early '90s when, because of the collapse of the Soviet Union, my weapons group was looking for other fields to conquer. Since our forte was non-linear dynamic models, there was a lot of overlap, and we were very familiar with the general tool set. At that time, the climatic models were very poor. As an example, there were instances of non-linear air mass flow phenomena modeled using Fourier analysis. For the non-mathematicians out there, linearity is one of the prerequisites for Fourier analysis to be applicable.
Is this based on your understanding from the late 1980s, plus having not heard of developments since? Or have you actually looked at the hindcasting skill of more current models? I'm afraid I'd be totally lost doing so, but my impression was obviously more positive than yours.
In the late '90s, I revisited the field in conjunction with a satellite program we were developing. There had been some improvement of the models in the seven or so intervening years (and some of the code had actually been converted to C or C++ :rolleyes:) but the state of the art was not much different -- ad hoc models with enough adjustable parameters to predict an elephant if the modeler desired. Since then, my interest has been strictly that of an informed outsider. To paraphrase Will Rogers, all I knew is what I read in Science News. Furthermore, the past 4 years I've even let that slip quite a bit -- part of that time I did not have the ability to understand topics at that level, and much of the rest of that time I was too weak to care. It has only been in the past year, and especially the past four months, that I've had the mental and physical capacity to again take interest in scientific matters.
So, having thus established my (lack of) credentials, take what follows as you will.
I have not seen the most recent models. Nor have I seen any announcements of a major breakthrough in climatic models. Since there does exist an ongoing debate about the validity of the models used, I would expect any substantial improvements to be widely touted in support of the modeler's claims. That leads me to believe that the models are probably much the same as they were ten years ago -- which is to say, not that much different than they were thirty years ago when they were first able to 'predict' global warming.
Most of the existing models of ten years ago could do a sufficient job of postdicting any given historical period with an appropriate choice of parameters. The problem was that it would take a different set of parameters to postdict a different period. Thus, in essence, the models were little more than a curve fitting program. Each historic era was fit with one set of parameters, and then the next with another, and so on much like a cubic spline is used to fit experimental data. And just like a cubic spline does not 'explain' the data, neither did these models.
For a model to be valid, it does indeed need "to predict the entire global climate past". If it cannot, if it can only predict some eras but not others, then how are we to know that the coming era is one which it can predict? How can we know when we can use it and when we can't?
As long as there is debate over the models and parameters needed to postdict the existing data, so long will the models be useless to predict the future. From what I've seen published, that debate still runs hot. I'd love to go to a climatologist convention and see how many "my honorable colleague is a damned fool" arguments break out. Hell, I'd even buy the pizza and beer.:)
--Pete
How big was the aquarium in Noah's ark?