cc: dprowell@meto.gov.uk, ckfolland@meto.gov.uk
date: Tue, 20 Apr 1999 17:44:29 +0100
from: Mark Rodwell <mjrodwell@meto.gov.uk>
subject: Nature paper
to: t.osborn@uea.ac.uk

Dear Tim,

Thanks for your email regarding your concerns about our figure 1.
We understand your points but feel we can strongly defend our approach.

Letters to Nature are necessarily short, with strict limits on the amount of
text and space allowed. Hence only the most important points can be made and
shown. Normally, letters should not exceed 2.5 pages but ours is already a full
page more than this. Whilst there are many important points we would have liked
to have made, many of these must wait for the companion paper. For figure 1, our
major results (and the only ones that are quantified in the text) were the high
correlations seen at seasonal and multi-annual timescales and the simulation of
the sign of the NAO. Neither of these are affected by the scaling (which is
explained in the figure caption). The disadvantage of your suggestion is that it
would make visual comparison of these results rather difficult, and give a
misleadingly poor impression of the potential seasonal and multi-annual forecast
skill for this period. Far from discouraging further work in the field, this
paper has generated a great deal of interest. If anything, it is leading to
increased research effort, not less. Below, we argue these points more fully and
note that we feel that Kushnir's 'New and Views' article in the same issue of
Nature gives a fair interpretation of our work and its implications.

This paper is fundamentally aimed at predictability of the NAO (e.g. the last
sentence in the abstract and lines 8-11 in the last paragraph of the paper) and
the mechanisms by which such predictability may arise. We show that the NAO may
be forced by SSTs but, since we use an atmospheric model, we are not able to
quantify the predictability of these SSTs. Instead, we rely on the knowledge
that persistence gives some predictability of SSTs a month or season ahead and
also reference Sutton and Allen who suggested that extratropical SSTs may be
predictable up to several years ahead. Beyond these timescales, strong
ocean-atmosphere coupling (in both directions) is clearly going to be important
and it is difficult to see how there will be any significant predictability of
the SST tripole pattern over the 30-year trend from the 60's to the 90's.
Certainly talking at present about 30-year predictability of the NAO based on
known SSTs is meaningless. Hence, we concentrated on the seasonal (upper curves)
and multi-annual (lower curves) timescales.

As you know, an unnormalised 50 year timeseries shows that the ensemble mean
consistently under-estimates the strength of the NAO (both in its positive and
negative phases). (Individual ensemble members have an NAO standard deviation of
around 75% of that observed). This consistency implies that the difference
between observed and ensemble mean is not simply noise. In addition, since the
downward trend to the 60's as well as the upward trend thereafter is
underestimated, it is not simply a question of overlying an externally forced
upward trend. What we have done by standardising the data is a standard practice
in prediction, which aims to achieve the best possible probability distribution
of the forecasts by allowing for systematic model errors. This point is
discussed in Potts et al (1996) in J. Climate, p46, Fig. 6. Hence, in addition
to simply wanting to display clearly our correlations, our approach in figure 1
is not 'incorrect' - it can be easily justified from a predictability point of
view. Ours and your approaches are just different ways of looking at the data,
with different aims in mind and each with its strengths and weaknesses - neither
is definitely right or wrong. It is often the case that there is no single valid
approach to a scientific problem.

Despite this, over 50% of the trend is simulated by the ensemble. This is
considerable and we have talked to collegues around the world about it (if you
wanted to quote actual numbers for the trend in your NAO paper, we would be very
happy to let you have these). The point is that the trend was not the focus of
the paper - our concluding paragraph summarizes what we saw as the major focus
and results. In addition, the figure caption specifies that the curves have been
normalised and the y-axis scaling is consistent with this.

We do not feel that we implied that everything had been fully explained. For
example, the abstract says that "much of...the variability...may be
reconstructed" (i.e. not all of it). We say that "the correlation rises to 0.74"
(i.e. a lot of the variance is not explained by this forcing). The last two
sentences of the paper emphasise the need for this mode to be explored in
coupled models. Indeed, as we noted earlier, this paper is resulting in
increased interest and work in this field.

Kushnir's 'News and Views' entry (pages 289-91 of the same issue of Nature)
contains a very balanced view of the paper. He talks about the "synchronicity
between model-simulated and observed NAO fluctuations" so he clearly understood
our wish, through our use of standardisation, to emphasise the coherency of the
time variations rather than their magnitudes. He also points out that a lot of
work needs to done to understand the predictability of the SST field and that
others "need to be able to reproduce Rodwell and co-workers' result with other
atmospheric models".

We hope that this allays your concerns.

Best wishes, Mark, Dave and Chris
________________________________________________________________

   Mark Rodwell                Tel:   (+44/0) 1344 856751
   Hadley Centre               Fax:   (+44/0) 1344 854898
   London Road                 Pager:        07669 035 269
   Bracknell                   Email: mjrodwell@meto.gov.uk
   RG12 2SY
   UK
________________________________________________________________