Aug 18 2007
Gavin Schmidt on Imperfect Data
Gavin Schmidt had a good response to a comment on his blog, and I didn’t want it to be lost (as if posting it here will help with that).
Underlying all of this is, I think, a big misconception about what replication means in an observational field like climatology. Everyone is working with ambiguous and imperfect data - whether that is from weather stations, paleo-records, models or satellites. Every single source needs to be processed in ways that are sometimes ad hoc (how to interpret isotopes, how to model convection, how to tie two satellite records and, yes, how to adjust for urban heating effects). Robustness of a result is determined by how sensitive the results are to those different ad hoc assumptions (do different climate models have the similar sensitivity, do results from south Greenland match those of North Greenland, does the UAH MSU record match the RSS MSU record). The important replication step is not the examination of somebody’s mass spec, or the analysis of lines of code, but in the broader results - do they conform with other independently derived estimates? Climate science is not pure mathematics where there is (sometimes) a ‘right’ answer - instead there are only approximations. These are outlined in innumerable papers which for decades have been the method of recording procedures, assumptions and sensitivities. To be sure, there are sometimes errors in coding (in the MSU record, or in climate models), but these come to light because the end results seem anomalous, not because there are armies of people combing through code. The independent replication of results is by far the more important stage in evaluating something.
Take the Greenland ice cores. The US and EU wanted to drill an ice core in central Greenland. They could have drilled one core , split the samples and done indpendent analysis of each sample. That would have given a good test of how good analytical techniques are. Fine. But what they elected to do was to drill two separate cores, 30 miles apart and do everything independently. This was much, much more useful. Firstly they found that for most of the cores the results were practically identical (which demonstrated the robustness of the analytics as well as sharing the samples would have), but more importantly, they found that the cores diverged strongly near the bottom - a sure sign that there was something wrong at one or (as it ended up) both sites. Without the independent replication that non-robustness would have not been uncovered. Thus while there are already two independent replications of the GISS temperature analysis (CRU and NCDC) - both of which show very similar features, there is always room for more. That would demonstrate something. Looking through code that does a bi-linear fit of data will not.
(No Ratings Yet)
Loading ...Related Posts:
6 Responses to “Gavin Schmidt on Imperfect Data”
To reduce spam, comments are automatically closed 30 days after the last comment. If you would like to comment on any closed thread, please use the contact form at the top of this page.
It’s not an either/or. Duplication by replication is useful, but so is reanalysis of data. In crystallography for instance, people would be aghast at the type of obstructionism that people like Mann practice. It is normal to get someone’s data and reanalyze to see if the structure is correct. Telling someone that they need to redo the entire data collection (which could even involve chemical synthesis) is silly.
Oh…and Gavin’s a lightweight who won’t allow arguments to continue on his blog. He blocks stuff like crazy. What a pussy.
[Response: I've deleted a few comments here. I don't get a lot, so I prefer to keep them all. But if I feel like someone is trolling, they get sent to the bit bucket!]
I troll a lot. But Gavin deletes posts when people are trying to debate an issue, to really get into the nitty gritty. It’s as if he thinks he is some herrdoktor professor and the purpose of the blog is education and actually debating something to brass tacks detracts from that. Of course, I think that’s silly. I think real men, real scientists, real mathematicians want to get as far as possible.
[Response: What Gavin does on his site is up to him. It appears that RealClimate has a fairly large following though, so he must be doing something right.
Here's just a particular situation when I think it would be okay to delete a comment: when someone says something untrue that sounds like science, and it would be difficult (take too long, be too technical for the audience, etc.) to refute, it may be better just to delete the comment instead of leaving it without responding.]
How do you distinguish that from when there is just a disagreement? Also, don’t you think that idiots expose themselves. Also, it’s only the hoi polloi on one side that end up getting censored. And in any case, what he does is censor people for getting detailed in discussion/argument.
Now, of course anyone can do anything they want when they have their own blog. But bottom line is that this sort of censorship will chill any kind of useful engagement. Gavin and Mike have censored Steve McIntyre, who for all of his flaws and I’ve cited them, is a completely useful person to debate given he has really read the literature and done data manipulation.
Honestly, I’m used to a more robust form of engagement than the Gavin types. But then I’m a real man from a non-politicised science.
The main purpose of RC very much is education, TCO.
No, the main purpose of RC very much is propaganda, unless you mean education that reflects ’socialist reality’ or some other such euphemism.
[Response: I've never seen the writers at RealClimate promote a socialist government.]
I would make a James Annan bayesian bet that the RC writers swing to the left. Also, they allow off-topic Bush-bashing (in moderation) and have done an interview on DKos.