Wednesday
Apr172013
by 
Bishop Hill
Bishop Hill Models vs observations: the troposphere
Apr 17, 2013 Climate: Models Climate: Surface Roy Spencer has posted up a very useful comparison of lower tropospheric temperatures against model predictions so that we can assess if the models are doing any better in the upper atmosphere.
They aren't.
Reader Comments (12)
Not sure which year(s) model predictions (as opposed to hindcasts) start.
First rule of climate science , if the models and reality differ in value its reality that is wrong .
Roy doesn't mind my comments!
Obviously the model predictions lag the observations by about 800 years or more.
It's been a long day I'll switch the light off now.
lol
As bad as they appear, the comparisons still might deceive the unwary into thinking that the models had some predictive skill, before they left the rails. The lie to this is given by noting that the model (hindcasts) all seem in excellent agreement with reality at the point where they shouldn't be: After the 1990 eruption of Pinatubo. Funny, that.
Half-time score:
Reality 44 - 0 Models
Would be interesting to know which part is "hindcast" and which is actual prediction. Does anyone know?
I think Roy Spencer suggests that the codes that best agree with empirical results employed lower values of climate sensitivity. Obviously doesn't verify actual values of the parameter, but endorses the suggestion that the IPCC's preferred value is too high. It certainly supports the suggestion that a Royal Commission on the subject should be set up.
Why are the implications of this not blindingly obvious to policy makers?
Maybe they haven't been told?
Over to you, Richard:-)
If models diverge from observation the MODELS ARE WRONG. This is basic science though it would appear that climate ''scientists'' failed that part of the course.
Yes, the models are wrong, but as has been stated elsewhere, some are useful. Turbulence models are used in aircraft design software. They contain parameters which have to be adjusted to match accurately executed wind tunnel experiments. If better real life data become available, the parameters must be altered to fit. This is legitimate and necessary for good design. It has been going on for a long time, since long before the computer era. Aircraft no longer fall out of the sky because the aerodynamicists were ignorant.
Turbulence is complicated enough but climatology is surely much more difficult to do correctly. Not only are there more fundamental phenomena to account for but there are probably too few data to guide theory and theory seems to be inadequate at present. The assertion that weather and climate prediction are separate is not acceptable. Weather predictions beyond a week are very poor.
Looking at Spencer's diagram, the modellers were probably quite happy up to 1997. The ad hoc parameter adjustments evidently worked.
From about 1997 to present (16 years), global average temperature as measured by satellite is constant to +- 0.25 deg C. Over the same period and extended into the future, the GCMs taken as a whole predict the globally averaged temperature to increase relentlessly, roughly linearly, by 0.3 deg C per decade up to 2025 at least. The spread of modelling predictions is +- 0.5 deg C.
Formally, they could claim collective agreement (invoking all models simultaneously) if the mean of the predictions plus the scatter (uncertainty) reliably overlapped the observed behaviour. But it doesn't. The measurements are right on the lower boundary of the swarm of computations and the two sets are clearly about to separate completely. When that happens, that the models really are wrong can hardly be denied.
As this cynic is writing these words though, are they waiting impatiently for yet more tuning runs to finish? Will they then be able to ‘explain’ why the old projections indeed suffered from premature understanding and to claim that the new projections now cover the observed temperatures?
It would be interesting to see exactly what the parameters are that have to be fitted to the data and how their ‘best’ values have varied over the past 20 years or so for each GCM.
Mark Well writes
"the modellers were probably quite happy up to 1997"
That's a pretty generous conclusion considering only results after 2004 - 2006 are actual forecasts, prior to that they are hindcasts.