2011-07-14 06:33:42Critique of Lindzen & Choi 2011
John Cook


We haven't yet addressed Lindzen's 2011 paper which supposedly corrects criticisms of his earlier paper (which we criticise at SkS). Here's an email about the new paper:

You may remember I was asking about Lindzen & Choi 2011.
There is a discussion thread on Climate etc,


and Rob Dekker makes a sharp criticism of LC11 (find by looking for

This is copied from his comment:


Lindzen and Choi obtain different feedback numbers from the same ERBE data than Trenberth 2010 and two other papers, and Lindzen claims
(unsurprisingly) that his method is more accurately reproducing feedback numbers.

When I looked at the details of his method however, I found something very concerning :

The Lindzen and Choi method of doing FLUX/SST analysis (called “lead and lag” by Lindzen) seems to have a (strong?) bias towards negative feedback.

Here is why :
L&C analyze fragments of SST changes that are either rising or falling, and then measures the FLUX response over the same period.
No problem there, has been done many times before by numerous other scientists.

The difference is that Lindzen is looking back and forth (lead and lag) in time, and finds the FLUX response that has the highest correlation with the SST change.

First remember that the FLUX (response) has significant noise on it.
Let’s note that if you do not look back and forth in time (no lead or lag), then on average the FLUX response will tell you the average FLUX response to that SST change.

But also remember that the FLUX response with the highest correlation with SST will always be the response that starts at one extreme and ends at the other extreme. All other responses will correlate less, since they will show opposite slopes at the start and/or end points, which obviously don’t correlate well with the SST.

So, if you are allowed to look back and forth in time through that noisy signal, you have a high chance of finding a lead or lag time where the FLUX response is larger (and thus correlates better) than the no-lag response alone.

So Lindzen and Choi method will (for each fragment of SST analysed) find the lead or lag time where the FLUX response is the largest !

When the FLUX response is larger for a certain SST change, the calculated feedback will be lower, and thus this method has a bias towards lowering the feedback calculated from the ERBE data.

Let me note that the effect (bias) will be stronger the more lead or lag time is allowed, since there will be more start and end-points in the noise to consider, and the largest response will correlate the best.

So for short lag times and strong negative feedback (large FLUX response), Lindzen’s method will be approximately correct. But for no-feedback or positive feedback the lead-lag bias will be very significant.

In fact Lindzen mentions himself that his method works best for large negative feedbacks .

He also mentions that his method works less well for small feedbacks (and consequently) large lag times, which, as I showed above is consistent with increased bias.

Interestingly enough, he does not show what feedback parameter number he obtains for a system with no feedback or positive feedback, in which case the lead-lag-noise bias will be greatest.

Needless to say that maybe Lindzen drew some very premature conclusions when he discards other scientists’ work (Trenberth et al, Dessler et al) who do NOT use his (biased) lead-lag-correlate method.

Now I have not quantified this bias yet, but this bias should be very easily reproducible using Lindzen’s (Spencer’s) “simple model” simulation,

What I’m saying is that in a climate system without feedback (such as a planet without an atmosphere) the Lindzen and Choi method will find a negative feedback just because of the noise. And that is simply incorrect

Interesting ?

Rob Dekker

Hope this is helpful

2011-07-14 15:57:13
Ari Jokimäki


Someone said somewhere that they haven't even corrected all the flaws (or, rather, deliberate biases) in L&C 2009. I haven't looked at the paper, though.