This is something I never expected to see in print. Climate modeler Dr. Gavin Schmidt of NASA GISS comments on the failure of models to match real world observations.
While the discussion was about social models, it is also germane to climate modeling since they too don’t match real world observations. Below is an example of climate models -vs- the real world; something’s clearly not right.
Graph source: IPCC AR5 draft
Is it maths or assumptions (or both) that cause the divergence?
UPDATE: In comments, I had a discussion with reader “jfk” which I think is worth sharing. He made some good points, and it helped hone my own thinking on the issue:
jfk says: Submitted on 2013/06/01 at 8:40 am
Well, I still think it’s a bit unfair to Gavin (and I am no fan of his). But hey, it’s Anthony’s site.
For a good review of the many failures of statistical modeling in social sciences (and one or two successes) see the book “Statistical Models: Theory and Practice” by David Freedman. Whether or not climate modeling has devolved to the point where it is social science rather than physics, well, I hope it’s not quite that bad…
REPLY: And I think it is more than a bit unfair to us, that if he believes what he tweets, he should re-examine his own assumptions about climate modeling. We have economies, taxes, livelihood, etc. hinging (or perhaps failing) on the success of these models to predict the climate in the future. The models aren’t working, and Dr. Schmidt knows this. Unfortunately his job is tied to the idea that they do in fact work. I feel no regrets at making this comparison front and center. – Anthony
UPDATE2: RussR in comments, provides this graph below showing Hansen’s modeled scenarios against real world observations. He writes:
Here’s an excel spreadsheet comparing observed temperatures vs. model projection from: Hansen (1988), IPCC FAR (1990), IPCC SAR (1995) and IPCC TAR (2001), in pretty charts.
It can be updated as more observations are added.
UPDATE3: Dr. Roger Pielke Sr. adds this in comments.
Climate models are engineering code with quite a few tunable parameters, and fitting functions in their parameterization of clouds, precipitation, land-atmospheric interfacial fluxes, long- and short-wave radiative flux divergences, etc. Only a part of these models are basic physics representations – the pressure gradient force, advection, the Coriolis effect.
The tunable parameters and fitting functions are developed by adjustment from real world data and a higher resolution models (which themselves are engineering code), but only for a quite small subset of real world conditions.
I discuss this issue in depth in my book
Pielke Sr, R.A., 2013: Mesoscale meteorological modeling. 3rd Edition, Academic Press, in press. http://www.amazon.com/Mesoscale-Meteorological-Modeling-International-Geophysics/dp/0123852374/ref=sr_1_2?ie=UTF8&qid=1370191013&sr=8-2&keywords=mesoscale+meteorological+modeling
The multi-decadal global climate model projections, when run in a hindcast mode for the last several decades are showing very substantial errors, as I summarize in the article