Indiana Jones Science .
Thanks and a hat tip to blogger Curious George
Indiana Pi Bill, NCAR Common Atmosphere Model 5.0, and Indiana Jones science.
In 1897, the Indiana Assembly attempted to legislate the value of Pi to be 3.2 – a value 1.9% too high. It created so much ridicule that the bill never became a law.
The Common Atmosphere Model 5.0 from National Center for Atmospheric Research ( NCAR ) decreed latent heat of water vaporization to be a constant, independent of temperature. They chose a value that is 2.5% too high for tropical seas (where most of surface water evaporation on our planet happens). NCAR CAM 5 is considered “science”.
Is a 2.5% error ridiculous? Would it be nice to have a 97.5% reliable climate forecast? Yes, but .. it would only be a one hour forecast. CAM 5 is an iterative model. It takes a state of the atmosphere and in a computational step it estimates the next state of the atmosphere – usually about one hour later. Should this estimated state contain a 2.5% error, the following step starting from it would add another 2.5% error of its own. The errors tend to accumulate. After 40 steps you may get a 100% error – your result is influenced by accumulated errors as much as by the underlying physical system. After 80 steps the accumulated errors totally dominate.
In this worst case scenario, predictions of the model become completely unreliable after 40 hours. Let’s hope CAM 5 behaves better – for example, an error in a coefficient does not imply the same error in a result. I asked NCAR.
Begin email Excerpts ( first, from me ):
*** 5/18/12 From: (me)
In your “Description of the NCAR Community Atmosphere Model (CAM 3.0)”, NCAR/TN – 464+STR, Table 6.1, “List of Physical Constants”, there is a Symbol Lv, Latent Heat of Vaporization, 2.501*10**6 J/kg. This is a correct value for 0 degrees Centigrade. Do you apply any corrections for other temperatures? If so, the correction coefficients should also be listed in Table 6.1. If not, could you please point me to a paper showing that no such correction is necessary.
*** 5/24/12 From: ucar.edu
As far as I know we only use the value for 0 deg and don’t any temperature corrections.
*** 6/1/12 From: (me)
Thank you very much for researching this matter. Since the latent heat of vaporization is
2.575 for -30C (in units of 10**6 J/kg)
2.501 for 0C
2.430 for +30C
the treatment that you describe introduces the error of 3% for this temperature range. I believe atmospheric temperatures may easily reach - 50C to + 50C ( that would introduce a 5% error at the extremes ). I repeat my question: is there a paper showing that such an error does not negatively impact the predictive power of your model?
*** 6/1/12 From: ucar.edu
Not that I know of ( off ) the top of my head.
*** 6/1/12 From: (me)
Thank you for your frank answer. Until I looked at your webpage, I had no idea that there were also CAM4 and CAM5. Is the latent heat treated differently there?
*** 6/19/12 From: ucar.edu
The latent heat of vaporization is a quantity that is identical in all components of the model (ice, land, ocean, sea ice) and it is a major task to change this since it is used to ensure energy conservation within the whole model system. As CESM is a community model I would encourage you to try and quantify the error with the current approximation and if significant try and correct it across the model. We will discuss this potential problem in our next model development meeting next week and discuss it’s potential priority.
*** (end of email excerpts) ***
I like the lighthearted approach NCAR takes to the accuracy of their flagship model. A total disregard for a temperature dependence of a basic thermodynamic property of water, with unknown consequences, is labeled a “potential problem”. Even better, the model has now been “scientifically validated”, whatever it means.
Please don’t get me wrong; I am not against climate models. I am not even against bad climate models – we have to start somewhere. Just reading the CAM 5 description will show you an incredible amount of work that went into the model. What a pity that authors then take an undocumented shortcut which renders any results worthless.
I know that today’s computing technology is barely adequate to model a severely simplified climate. What I am against is to take that extremely simplified model seriously enough to worry about emperor penguins. NCAR should worry about a quality of their models and an accuracy of model results. But homeless penguins or drowning polar bears invoke much stronger emotions. Media love them.
I don’t know if a NCAR model development meeting mentioned in a 6/19/2012 email discussed this “potential problem”. A year later there is no trace of it. There is a new list of Notable Improvements to the model, which does not contain a word “latent” at all. However it hints that the model has been scientifically validated. Scientific validation apparently goes the way of peer review and dinosaurs.
There is also an updated CAM 5 description, dated November 2012. It lists many approximations used in the model, but I could not find anything related to the temperature dependence of the latent heat. A nice display of scientific chastity: hide anything that disagrees with your “science”. Why worry about potential problems ?
Indiana Jones never does.