Saturday, May 19, 2012


date: Fri Oct 24 17:00:26 2008
from: Keith Briffa <>
subject: Re: Question on climate reconstructions, and a query on model
to: "Richard Baldwin" <>

see you then - cheers
At 16:55 24/10/2008, you wrote:

Hi Keith,
Thanks for the replies.
There are elements of standardisation that I'm still not quite sure of, so post-reading
and armed with questions I shall drop by one day next week when I'm in Norwich (I live
in London...) to go through some of this further.
Thanks again.
2008/10/24 Keith Briffa <[1]>

Hi Richard
sorry for delayed response - things a bit manic here
At 15:09 22/10/2008, you wrote:

Hi Keith,
In further query of your comments on issues with standardising climate data,
and in a way hopefully simplifying the process to myself, is the (simply
expressed) issue with the divergence between current instrumental climate data
and observed tree ring density that the change has been too rapid to allow a
'smoothing' of the high frequency variability to occur?

merely that the divergence is only apparent (at the NH average scale) in the
smoothed i.e. lower-frequency domain. You need to smooth the tree-ring records and
the temperature to see it. However, the divergence is largely an artifact of using
curve fitting (i.e. based on least-squares fitted regression lines or functions ) to
estimate the unwanted (biological) growth trend in the tree-ring data. These fits
are influenced by climate warming signals in the recent data , and this signal is
inadvertently removed in the standardising process. When non-curve fitting methods
are used (such as RCS) this problem is largely removed. I attach a recent papers
that goes into this - though you DO NOT NEED TO UNDERSTAND ALL THE DETAILS OF THIS.

Or is there something else lurking in there still? I must admit that having
read some of the reading list papers I'm still a bit unclear on some of these

If you wish to discuss this further after looking at the papers - come and chat

And secondly - with so much of the SH being ocean, is it the case that a
certain number of grid cells in GCMs will necessarily be made up of 'best
estimate' parameters due to a lack of a) direct observations and b) proxy

GCMs are based only on physical equations - no observational data or proxy data
really affect the formulation of the models. OK some reworking of model parameters
does in reality take place to get the model to better simulate observed conditions
- but only at a gross average scale . That is why we can justifiably compare
simulated model output with observations , where the models are forced with
realistic estimates of the net effects of processes that produce climate changes ie
volcanic activity, solar radiation changes , changes in atmospheric constituents -
all of which directly or indirectly produce the radiative forcing that ultimately
produces changes in regional and global climates.

and if so will this have an effect on the uncertainty levels of predictions
based on the model output? Or as it is all ocean can more assumptions on
parameter values be made along similar uniformitarian principles as with the
tree ring data?

The design of models is based on our physical understanding of the climate
system - true though that some components of the system are ignored or very
simplified - as was interactive vegetation until very recentlty. In as much
as the values ascribed to these processes may be poorly understood and may not
be valid on some timescales or through time the uniformitarianism principle
can be considered as applying here also - but not in the sense of
regression-based interpretations of proxy data (when these are merely
regressed against instrumental data with no consideration (often unavoidably)
for the underlying influence of other processes that may obscure, mask or
bias the apparent relationships thus established , and that are used to
retrodict climate over periods when the "other" processes may not act in the
same way.

2008/10/7 Keith Briffa <<[2]>[3]>
happy to chat about this after tomorrow's lecture if you wish - but in the
meantime ;
the distinction I make in the chapter is between empirical signal on the one
hand and theoretical signal on the other.
This of course is a frame of reference invented for convenience. The
theoretical signal in this chapter should be taken to be a measure of the
representation, within the chronology or chronologies , of the specific
climate variability with which we are concerned. This could be , for example
the average of June,July and August temperature as measured by some
instrumental record for the region.
What I mean by the statement is that if I am interested in reconstructing the
past variability of this specific climate variability at long time scales -
i.e. how mean JJA temperature changes on time scales of a century or more, the
chronologies must be processed (i.e. standardised) in such a way as the
expressed empirical signal (i.e. the expression of the common variability
actually contained within the trees we have sampled) is at least potentially
preserved at this same long time scale. This is not to say that preserving the
long time scale information will ensure a good representation of the
theoretical signal as it is expressed by the chronology. Rather that, even if
tree growth in an area is influenced by summer temperatures at this long time
scale, if we process the measured ring-width data in such a way that the long
time scale variance is removed (effectively high-pass filtering the
chronology) no evidence of long time scale temperature variability can
possibly be recovered from these standardised data. In fact , in some
situations, it is better to sacrifice this "potential" information in the
chronologies in order to ensure the reliability of the preserved (higher
frequency) variance. In doing this we can often get a more reliable
reconstruction , although of only the high-frequency part of the variance
spectrum. This is because in some situations preserving the low-frequency
involves accepting low reliability of this information in the chronology , or
because the low-frequency information preserved in the trees is simply not
well correlated with the low-frequency evidence of measured temperatures in
the area. You will see in the later lecture that , depending on the approach
we use to scale (calibrate) the tree-ring variability against the the climate
series we seek to reconstruct, it can be better to throw away the
low-frequency information and scale directly against only the equivelent time
scale climate information. We can discuss this in more detail later.
For now , hope this answers your question - we need to make this point clear
because it has wide relevance in the use of various proxy interpretations.
At 13:48 07/10/2008, you wrote:
I'm reading through your Ch5. and have a query regarding the following
phrase in section 5.5.2:
"If the required theoretical signal involves long-timescale variability, a
very conservative approach must be adopted when standarizing..."
In that context, what is meant by 'theoretical signal'? Or can I removed
'theoretical' from it and simply think of it in terms of signal and noise
as was discussed in the lecture?
Professor Keith Briffa,
Climatic Research Unit
University of East Anglia
Norwich, NR4 7TJ, U.K.
Phone: +44-1603-593909
Fax: +44-1603-507784
Richard Baldwin
07878 37 49 64

Professor Keith Briffa,
Climatic Research Unit
University of East Anglia
Norwich, NR4 7TJ, U.K.
Phone: +44-1603-593909
Fax: +44-1603-507784

Richard Baldwin
07878 37 49 64

Professor Keith Briffa,
Climatic Research Unit
University of East Anglia
Norwich, NR4 7TJ, U.K.

Phone: +44-1603-593909
Fax: +44-1603-507784

No comments:

Post a Comment