Models or Observations?

Epistemology Matters

The title of this piece comes from a paper by Hugh Ellsaesser in Atmospheric Environment [Ellsaesser, H.W. (1982). Should we trust models or observations? Atm. Env. 16(2):197-205].

The compelling need for me to do this piece arose from a comment made at a recent conference on integrated regional models. The comment went like this: "The role of the biosphere on the atmosphere was not recognized until Shukla published his paper."

Shukla is a latecomer to this topic. Shukla's paper was a General Circulation Model (GCM) paper. There was not a lick of observational data in it. How is it that we have come to believe models and not the dozens of scientific papers using data that come to the same conclusion?

CED readers, readers-in-the-know that you are, know that vegetation impacts on the dynamics of the atmosphere are very real. A robust literature is available and it has a very long history. Now, for a decade some of us have been pushing GCM modelers to recognize the role of sulfates in cloudiness and climate change. Paper after published paper implicated sulfate as a cloud maker. Then, Jim Hansen published a paper using the GISS GCM output statistics that included planetary albedo adjustments for the trend in atmospheric sulfate and found it largely offset CO2 warming. The IPCC report included it as fact.

We need a new piece of jargon "model-fact." With time the hyphen in model-fact will disappear and it will appear in polite discourse as “modelfact.” It will then join metadata as a term we just can't do without. So here we are, if you can get a model to project it, somehow, it becomes larger than life.

Chris Folland, an IPCC author, when questioned about this strange inversion of science noted that "The data don't matter." He went on to note that "global warming is model-driven not data driven. The data just don't matter."

Now I have been engaged in modeling of one sort or another for 15 years and will continue to participate. However, the standard long held in the atmospheric sciences is that the models must be proven to have skill! It must out-perform chance, climatology and simpler models before you offer it as an icon for belief.

For example, the GFDL GCM transient run with 1% CO2 increase per year puts out a global field of temperatures for each year. If the model begins with background CO2 circa 1820 and runs forward in time, one could ask the question how well has the model accounted for the variance in global temperature patterns that was actually observed.

Such tests, here at the University of Virginia, show that the model explains 4% of the variance. Clearly, you would not want to use a model to make "predictions" about future conditions if you were only explaining 4% of the variance which is equivalent to a correlation coefficient of 0.2! GCMs are wonderful tools of experimentation [change one thing and hold all else constant] and hypothesis generation, but have yet to prove their stuff as prediction tools!

It wasn’t so many years thereafter that the Hockey Stick shaped “global” temperature curve hit the journals. Big time-change (warm) after 1850 AD and not so much for a thousand years leading up to 1850. All of a sudden the “data” did matter and we began a new phase of climate discourse.

Fortunately, we have at hand a major archive of radiocarbon data published in the journal Radiocarbon. Bryson and Podoch used radiocarbon data indicating for times of environmental (biotic, cultural and geophysical change) and statistically determined the most probable dates. 17 such date of change and 18 climate episodes were found. You can find Reid Bryson and Christine Padoch’s paper in Climate and History: Studies in Interdisciplinary History Edited by Robert I. Rotverg and Theodore K Rabb. Princeton University Press.


Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer