Controversy over comparing models with observations

by Judith Curry
My draft talk elicited an interesting conversation on twitter, that deserves some wider discussion.

The figure in question is from John Christy’s recent congressional testimony

Gavin Schmidt tweeted:  TMT (trop + glob) comparisons to CMIP5 models on a reasonable baseline, and provided these figures

My reaction was that these plots look nothing like Christy’s plot, and its not just a baseline issue.
Fernando Leanme tweets: @ClimateOfGavin @curryja I’m happy because you use CMIP5 RCP4.5
Gavin followed up with this tweet:  @curryja use of Christy’s misleading graph instead is the sign of partisan not a scientist. YMMV.
gavin tweets: @curryja Hey, if you think it’s fine to hide uncertainties, error bars & exaggerate differences to make political points, go right ahead.
JC responds: @ClimateOfGavin says the king of overconfidence. What political point? A true partisan sees politics in all that disagrees with him
Gavin tweets:  @curryja The only place Christy has ‘published’ his figure is in his testimony to Congress. You think that isn’t political?
JC responds:  @ClimateOfGavin I’ve decided to leave out that figure, I haven’t seen a figure that I am confident of using at this point
Chip Knappenberger tweets: @ClimateOfGavin @curryja Should also include a comparison of the trends (to avoid accusations). Something like this:
Chip Knappenberger tweets:  @ClimateOfGavin @curryja Is there a “right” answer when it comes to what baseline to use? [refers to blog post by Ed Hawkins] http://www.climate-lab-book.ac.uk/2015/connecting-projections-real-world/
JC tweets @PCKnappenberger @ClimateOfGavin baseline is a subjective decision, related to the scientific point you want to make.
JC tweets: @PCKnappenberger @ClimateOfGavin The key point is the difference in trends. How to explain this in brief public presentation is a challenge
Gavin tweets: @curryja @PCKnappenberger If you want to show trends, just show trends – incl ens spread & uncertainties in data

JC responds:  @ClimateOfGavin @PCKnappenberger sorry, such a graph is incomprehensible to a non technical audience with 1 minute per slide
Gavin responds: @curryja If you want a talk for non-technical audiences you should start over and get rid of all the graphs. @PCKnappenberger
JC responds: @ClimateOfGavin @PCKnappenberger they can understand graphs, they like to see data, understand correlations & visual diff btwn models/obs
Gavin responds: @curryja Then they can read a histogram #specialpleading
But why don’t you just make your own figures if these are not ok?
@PCKnappenberger
JC reflections
A whole host of interesting issues are raised in this exchange:

  • How to communicate complex data to a non technical audience?
  • How to best compare model predictions against observations?
  • What is are the most reliable sources of such plots to use in public presentations?

I’ll start with the third question first.  When selecting figures to use in presentations or testimony, I am looking for the most credible figures to use.  I use figures from the most recent IPCC assessment where possible.  A second choice is a figure from the published literature.  However, in public presentations and testimony, they are looking for the most up-to-date analysis with the latest observations (which aren’t in peer reviewed publications owing to research-publication time lags).  Hence I have often used Ed Hawkins’ update of figure 11.25 from the AR5 comparing model projections and surface observations (Ed was the author of fig 11.25).
With regards to John Christy’s figure, he is the author of one of the main observational data sets used in the comparison. I don’t know the source of the time series that Gavin provided, but the observations in gavin’s figure vs Christy’s figure do not look similar in terms of time variation.  I have no idea how to explain this.  I have to say that I think John Christy’s figure is more reliable, although some additional thought could be given to how to define the beginning reference point to eliminate any spurious influence from El Nino or whatever.
Why don’t I draw my own figures for such presentations?  Apart from the issue of lack of time and lack of artistic skill in making such plots look nice, I regard published diagrams or diagrams made by originators of the data sets to have a higher credibility rank, as well as being a source of analysis independent from the person summarizing the information (i.e. me).
The issue of comparing models to observations has been hashed out here (Spinning the climate model – observation comparison Part I, Part II, Part III) and at other blogs (e.g. Lucia’s, etc.).   How to compare depends on the point you are trying to make.  For example, in Gavin’s first time series plot, I am not sure what the point is of comparing to scenario R4.5.   I don’t see that the baseline matters that much, if you are mainly comparing trends.
I really like the histogram, this conveys exactly the point I wanted to make, although explaining this to a nontechnical audience in ~1 minute is pretty hopeless.  I also like Ed Hawkin’s figure 11.15.
A final comment on twitter vs blogs.  I find twitter to be a great source of links, but very frustrating for conducting conversation such as this.Filed under: Communication

Source