To: <santer1@llnl.gov>

Subject: RE: Possible error in recent IJC paper

Date: Fri, 31 Oct 2008 11:01:46 -0000

Cc: "Jones Philip Prof \(ENV\)" <P.Jones@uea.ac.uk>, "Gavin Schmidt" <gschmidt@giss.nasa.gov>, "Thorne, Peter" <peter.thorne@metoffice.gov.uk>, "Tom Wigley" <wigley@cgd.ucar.edu>

Dear Ben,

many thanks for the full response to my query. I think my confusion arose from the

discussion on RealClimate (which prompted our earlier communication on this topic), which

clearly suggested that the observed trend should be expected to lie within the spread of

the models, rather than neccessarily being close to the mean as the models are stochastic

simulations (which seemed reasonable). I've just re-read that post, the key paragraph from

[1]http://www.realclimate.org/index.php/archives/2007/12/tropical-troposphere-trends/ is as

follows:

"The interpretation of this is a little unclear (what exactly does the sigma refer to?),

but the most likely interpretation, and the one borne out by looking at their Table IIa, is

that sigma is calculated as the standard deviation of the model trends. In that case, the

formula given defines the uncertainty on the estimate of the mean - i.e. how well we know

what the average trend really is. But it only takes a moment to realise why that is

irrelevant. Imagine there were 1000's of simulations drawn from the same distribution, then

our estimate of the mean trend would get sharper and sharper as N increased. However, the

chances that any one realisation would be within those error bars, would become smaller and

smaller. Instead, the key standard deviation is simply sigma itself. That defines the

likelihood that one realisation (i.e. the real world) is conceivably drawn from the

distribution defined by the models."

I had therefore expected the test to use the standard deviations of both the models and the

observations (which would give a flat plot in 5B and there would be an obvious overlap of

the uncertainties in 6a at say 500hPa).

best regards

Gavin

-----Original Message-----

From: Ben Santer [[2]mailto:santer1@llnl.gov]

Sent: Fri 10/31/2008 4:06 AM

To: Cawley Gavin Dr (CMP)

Cc: Jones Philip Prof (ENV); Gavin Schmidt; Thorne, Peter; Tom Wigley

Subject: Re: Possible error in recent IJC paper

Dear Gavin,

Thanks very much for your email, and for your interest in our recent

paper in the International Journal of Climatology (IJoC). There is no

error in equation (12) in our IJoC paper. Let me try to answer the

questions that you posed.

The first term under the square root in our equation (12) is a standard

estimate of the variance of a sample mean - see, e.g., "Statistical

Analysis in Climate Research", by Francis Zwiers and Hans von Storch,

Cambridge University Press, 1999 (their equation 5.24, page 86). The

second term under the square root sign is a very different beast - an

estimate of the variance of the observed trend. As we point out, our d1*

test is very similar to a standard Student's t-test of differences in

means (which involves, in its denominator, the square root of two pooled

sample variances).

In testing the statistical significance of differences between the model

average trend and a single observed trend, Douglass et al. were wrong to

use sigma_SE as the sole measure of trend uncertainty in their

statistical test. Their test assumes that the model trend is uncertain,

but that the observed trend is perfectly-known. The observed trend is

not a "mean" quantity; it is NOT perfectly-known. Douglass et al. made a

demonstrably false assumption.

Bottom line: sigma_SE is a standard estimate of the uncertainty in a

sample mean - which is why we use it to characterize uncertainty in the

estimate of the model average trend in equation (12). It is NOT

appropriate to use sigma_SE as the basis for a statistical test between

two uncertain quantities. The uncertainty in the estimates of both

modeled AND observed trend needs to be explicitly incorporated in the

design of any statistical test seeking to compare modeled and observed

trends. Douglass et al. incorrectly ignored uncertainties in observed

trends.

I hope this answers your first question, and explains why there is no

inconsistency between the formulation of our d1* test in equation (12)

and the comments that we made in point #3 [immediately before equation

(12)]. As we note in point #3, "While sigma_SE is an appropriate measure

of how well the multi-model mean trend can be estimated from a finite

sample of model results, it is not an appropriate measure for deciding

whether this trend is consistent with a single observed trend."

We could perhaps have made point #3 a little clearer by inserting

"imperfectly-known" before "observed trend". I thought, however, that

the uncertainty in the estimate of the observed trend was already made

very clear in our point #1 (on page 7, bottom of column 2).

To answer your second question, d1* gives a reasonably flat line in

Figure 5B because the first term under the square root sign in equation

(12) (the variance of the model average trend, which has a dependence on

N, the number of models used in the test) is roughly a factor of 20

smaller than the second term under the square root sign (the variance of

the observed trend, which has no dependence on N). The behaviour of d1*

with synthetic data is therefore dominated by the second term under the

square root sign - which is why the black lines in Figure 5B are flat.

In answer to your third question, our Figure 6A provides only one of the

components from the denominator of our d1* test (sigma_SE). Figure 6A

does not show the standard errors in the observed trends at discrete

pressure levels. Had we attempted to show the observed standard errors

at individual pressure levels, we would have produced a very messy

Figure, since Figure 6A shows results from 7 different observational

datasets.

We could of course have performed our d1* test at each discrete pressure

level. This would have added another bulky Table to an already lengthy

paper. We judged that it was sufficient to perform our d1* test with the

synthetic MSU T2 and T2LT temperature trends calculated from the seven

radiosonde datasets and the climate model data. The results of such

tests are reported in the final paragraph of Section 7. As we point out,

the d1* test "indicates that the model-average signal trend (for T2LT)

is not significantly different (at the 5% level) from the observed

signal trends in three of the more recent radiosonde products (RICH,

IUK, and RAOBCORE v1.4)." So there is no inconsistency between the

formulation of our d1* test in equation (12) and the results displayed

in Figure 6.

Thanks again for your interest in our paper, and my apologies for the

delay in replying to your email - I have been on travel (and out of

email contact) for the past 10 days.

With best regards,

Ben

Cawley Gavin Dr (CMP) wrote:

>

>

> Dear Prof. Santer,

>

> I think there may be a minor problem with equation (12) in your paper

> "Consistency of modelled and observed temperature trends in the tropical

> trophosphere", namely that it includes the standard error of the models

> 1/n_m s{<b_m>}^2 instead of the standard deviation s{<b_m>}^2. Firstly

> the current formulation of (12) seems at odds with objection 3 raised at

> the start of the first column of page 8. Secondly, I can't see how the

> modified test d_1^* gives a flat line in Figure 5B as the test statistic

> is explicitly dependent on the size of the model ensemble n_m. Thirdly,

> the equation seems at odds with the results depicted graphically in

> Figure 6 which would suggest the models are clearly inconsistent at

> higher levels (400-850 hPa) using the confidence interval based on the

> standard error. Lastly, (12) seems at odds with the very lucid

> treatment at RealClimate written by Dr Schmidt.

>

> I congratulate all 17 authors for an excellent contribution that I have

> found most instructive!

>

> I do hope I haven't missed something - sorry to have bothered you if

> this is the case.

>

> best regards

>

> Gavin

>

--

----------------------------------------------------------------------------

Benjamin D. Santer

Program for Climate Model Diagnosis and Intercomparison

Lawrence Livermore National Laboratory

P.O. Box 808, Mail Stop L-103

Livermore, CA 94550, U.S.A.

Tel: (925) 422-3840

FAX: (925) 422-7675

email: santer1@llnl.gov

----------------------------------------------------------------------------

References

1. http://www.realclimate.org/index.php/archives/2007/12/tropical-troposphere-trends/

2. mailto:santer1@llnl.gov

## No comments:

## Post a Comment