Models and Their Discontents

Here are some bits out of a posting about climate models by Roy W. Spencer on his blog:

The strongest piece of evidence the IPCC has for connecting anthropogenic greenhouse gas emissions to global warming (er, I mean climate change) is the computerized climate model. Over 20 climate models tracked by the IPCC now predict anywhere from moderate to dramatic levels of warming for our future in response to increasing levels of atmospheric carbon dioxide. In many peoples’ minds this constitutes some sort of “proof” that global warming is manmade.

Yet, if we stick to science rather than hyperbole, we might remember that science cannot “prove” a hypothesis….but sometimes it can disprove one. The advancement of scientific knowledge comes through new hypotheses for how things work which replace old hypotheses that are either not as good at explaining nature, or which are simply proved to be wrong.

Each climate model represents a hypothesis for how the climate system works. I must disagree with my good friend Dick Lindzen’s recent point he made during his keynote speech at the 4th ICCC meeting in Chicago, in which he asserted that the IPCC’s global warming hypothesis is not even plausible. I think it is plausible.

And from months of comparing climate model output to satellite observations of the Earth’s radiative budget, I am increasingly convinced that climate models can not be disproved. Sure, there are many details of today’s climate system they get wrong, but that does not disprove their projections of long-term global warming.

Where the IPCC has departed from science is that they have become advocates for one particular set of hypotheses, and have become militant fighters against all others.

They could have made their case much stronger if, in addition to all their models that produce lots of warming, they would have put just as much work into model formulations that predicted very little warming. If those models could not be made to act as realistically as those that do produce a lot of warming, then their arguments would carry more weight.

Unfortunately, each modeling group (or the head of each group) already has an idea stuck in their head regarding how much warming looks “about right”.

...

One of the first conferences I attended as a graduate student in meteorology was an AMS conference on hurricanes and tropical meteorology, as I recall in the early 1980’s. Computer models of hurricane formation were all the rage back then. A steady stream of presentations at the conference showed how each modeling group’s model could turn any tropical disturbance into a hurricane. Pretty cool.

Then, a tall lanky tropical expert named William Gray stood up and said something to the effect of, “Most tropical disturbances do NOT turn into hurricanes, yet your models seem to turn anything into a hurricane! I think you might be missing something important in your models.”

I still think about that exchange today in regard to climate modeling. Where are the model experiments that don’t produce much global warming? Are those models any less realistic in their mimicking of today’s climate system than the ones that do?

If you tell me that such experiments would not be able to produce the past warming of the 20th Century, then I must ask, What makes you think that warming was mostly due to mankind? As readers here are well aware, a 1% or 2% change in cloud cover could have caused all of the climate change we saw during the 20th Century, and such a small change would have been impossible to detect.

...

It seems to me that all the current crop of models do is reinforce the modelers’ preconceived notions. Dick Lindzen has correctly pointed out that the use of the term “model validation”, rather than “model testing”, belies a bias toward a belief in models over all else.

It is time to return to the scientific method before those who pay us to do science — the public — lose all trust of scientists.

I spent a number of years building models of computer system performance as part of engineering complex systems. Back in the early 1980s I had some confidence that my models were correct. But by the time the year 2000 rolled around, I realized that most of my models were wishful thinking. The underlying systems had gotten much more complicated. At the same time, the computer vendors were supplying less and less engineering data to support model building. The bottom line: it is extremely hard to model complex systems. Physicists are lucky in that their systems at root are simple, it is the interactions that are complex. But in most real world systems you have complex systems where the details of the elements at the bottom are murky at best. Models can't predict but they can show sensitivity and point out potential problems. They don't represent "reality".