Comments on: Huybers #2: Re-Scaling

By: Pseudoproxies in Mann et al 2007 « Climate Audit

Pseudoproxies in Mann et al 2007 « Climate Audit — Thu, 22 Mar 2012 22:55:10 +0000

[...] me by Eduardo Zorita) and one using pseudoproxy networks constructed to emulate the MBH98 network (Huybers #2 and Reply to Huybers) also here and I’ll try to tie these three different studies [...]

By: Briffa on Yamal Impact « Climate Audit

Briffa on Yamal Impact « Climate Audit — Thu, 31 Mar 2011 17:47:45 +0000

[...] that they assert using either CPS or Mannian methods. The graphic below, taken from a CA post here from a few years ago shows the impact of replacing all the “proxy” series in the MBH98 [...]

By: TCO

TCO — Sat, 04 Aug 2007 15:48:33 +0000

Link: http://www.climateaudit.org/?p=836#comment-124625

By: TCO

TCO — Sat, 04 Aug 2007 14:41:51 +0000

But the issue is that he did not just use a simple AR1 or AR11 method. He used a method that extracts the noise from a record and creates new series based on it. How do we then know that all of what he extracted is really noise? Is it relevant to call it “just noise” if it has so much connection to the actual experimental evidence? (This is the circular logic I’m concerned about.) I will bump the thread (read the comments) where we finally got a better hand on what Steve was doing with his noise generation. You actually have to go to code 9including a subroutine) to understand it.

By: UC

UC — Sat, 04 Aug 2007 14:31:10 +0000

They are highly modeled noise derived from the actual sample set.

OK, Steve’s red noise is probably , where x is i.i.d zero mean unity std Gaussian vector. And you are afraid that carries information to that noise, as it is estimated from the proxies. Sounds a bit far-fetched, but should be easy to check. Just force to a covariance matrix of AR1 process, and re-run.

yet there is at least an open question as to whether there is a meaningful signal within those proxies.

Of course there is some meaningful signal in the proxies, trees don’t grow at all if temperature is 400 C, or -150 C. Other aspects of your part b remain unclear to me. But I Know I need to study MBH98 methods (and Steve’s articles) more. (Still kind of stuck to the evolving multivariate regression part.)

By: TCO

TCO — Thu, 02 Aug 2007 11:52:23 +0000

(Caveat) First, it’s not an allegation. It’s something I want to check.

a. My understanding is that Steve made his “just red noise” proxies by taking the acf of each of 70 specific proxies used in the actual study. The acf has 5+ parameters to it. That’s 350 parameters. Now maybe some are irrelevant. But still it worries me as a possible problem. The “just red noise” not being some simple set of duplicated AR1 or 11 proxies. They are highly modeled noise derived from the actual sample set.

b. In addition, the actual sample proxies were not detrended, yet there is at least an open question as to whether there is a meaningful signal within those proxies. Thus some aspects of signal (or at least debated, open to question) signal are being a priori defined as noise.

By: UC

UC — Thu, 02 Aug 2007 07:40:29 +0000

This is maybe the most interesting thread here, it is good to bring it up from time to time. However, I don't quite understand your comment, where's the circular logic? Where's the fitting? Note Mann's comment in #54. ICE or CPS + CIs from calibration residuals = upper-bounded errors, before seeing the data (remember Juckes and CVM). Not completely unlike a civil engineer who is about to design a bridge, and tells you the weight limit before choosing the material.

By: TCO

TCO — Thu, 02 Aug 2007 04:31:28 +0000

I suspect that there is some circular logic here. Steve says that what he is using is “just red noise”, but it has MANY parameters to it (a lot of fitting). It’s NOT a simple AR1 or (1,1) model. And it is based itself off of the bcps which at least are in question as to whether there is a century scale signal.

By: UC

UC — Thu, 05 Jul 2007 20:42:09 +0000

I wrote this ICE vs. CCE and associated CI stuff a bit more formally here:

http://signals.auditblogs.com/2007/07/05/multivariate-calibration/

Profs. Brown, Sundberg etc. probably will tell me where I go wrong with this, I’ll let you know

By: UC

UC — Thu, 21 Jun 2007 06:06:58 +0000

The situation is quite funny. Essentially ICE is based on assumption of stationarity, i.e. calibration temperatures are representative sample of past temperatures (no AGW). VS04 notes that if you use ICE, you'll underestimate past temperatures (if they exceed the calibration data range). Mann responds to criticism by claiming that VS04 'simulation was forced with unusually large changes in natural radiative forcing in past climates'. IOW, it is ok to use ICE-like estimators because past temperatures were in the range of calibration temperatures. And these high-school kids are in power.. Something related:

In the CPS method, the composite is typically scaled (e.g., Bradley and Jones 1993; Crowley and Lowery 2000; Esper et al. 2002; MJ03) to have the same variance as the target instrumental series . Such scaling is based on the as assumption that the contributing proxy series can be treated as actual local temperature indicators, rather than simply a set of statistical predictors. The CPS approach thus differs from a multivariate regression approach wherein the reconstruction is guaranteed to have less variance than the target series over the calibration interval .

(Mann et al 2005, Testing the Fidelity of Methods Used in Proxy-Based Reconstructions of Past Climate, my emph).