вход по аккаунту



код для вставкиСкачать
Atmos. Sci. Let. 6: 145–147 (2005)
Published online 5 August 2005 in Wiley InterScience ( DOI: 10.1002/asl.107
Comment on the paper: On the design of practicable
numerical experiments to investigate stratospheric
temperature change, by S. Hare et al. (2005)
John Austin1,2 *
Boulder, CO, USA
Geophysical Fluid Dynamics Laboratory, Princeton, NJ, USA
*Correspondence to:
John Austin, NOAA Geophysical
Fluid Dynamics Laboratory,
Princeton Forrestal Campus Rte.
1, 201 Forrestal Rd., Princeton,
NJ 08542-0308, USA.
If stratospheric temperature trends are to be understood, coupled chemistry climate models
will need to be run. Simulations with fixed ozone trends might provide a misleading
indication of future temperature trends. Copyright  2005 Royal Meteorological Society
Received: 19 May 2005
Revised: 24 May 2005
Accepted: 25 May 2005
In the above article, Hare et al. (2005) compare results
from the coupled chemistry climate run of Austin and
Butchart (2003) and results from uncoupled models
to conclude that the benefits of coupled chemistry
models are disproportionate to their cost. They suggest
that it is preferable to complete a small ensemble
of runs of an uncoupled model. In this comment I
give an alternative view, namely, that if stratospheric
temperature trends are to be properly understood,
coupled chemistry climate models will need to be
run, whether as ensembles or not, and that simulations
with fixed ozone trends might provide a misleading
indication of future temperature trends.
1. Model details
First, I would like to clarify a few important details
regarding UMETRAC (Unified Model with Eulerian
Transport and Chemistry). The chemistry is coupled
every physics timestep (30 min), although in practice
it is the radiation timestep (3 h) that is more significant. Improvements in transport such as using a semilagrangian method can reduce the overhead of running
chemistry, e.g. to a factor of 2.2 in the Geophysical
Fluid Dynamics Laboratory (GFDL) Model. This is
assuming that the sea surface temperatures (SSTs) and
sea ice are fixed. In a coupled ocean model, the additional resources for the ocean would mean that the cost
of coupled chemistry would be only about a factor
of 1.6 above that of a model without chemistry. The
balance of whether or not to include chemistry then
becomes much less clear. In comparison, in the last
two years computer capability at GFDL has increased
Copyright  2005 Royal Meteorological Society
effectively by a factor of about 6, divided evenly
between hardware and software, so that here we can
contemplate both coupled chemistry and ensembles
at higher overall spatial resolution than UMETRAC.
Institutions that are unable to keep up with world-class
facilities could consider investigating simpler problems that do not require such facilities or to work in
partnership with those that do have the facilities.
An issue not raised by the authors is the quality
of the underlying chemistry and transport models,
and this could be far more important than statistical
uncertainty in the trends. As indicated in Austin
et al. (2003), different models can give a wide range
of results, but some assumptions need to be made
regarding future ozone trends if future temperature
trends are to be predicted.
2. Trend uncertainties
Although it was not our primary goal in Austin
and Butchart (2003), we partially addressed the issue
of trend attribution by comparing the model trends
for the period 1980–2000 with those for the period
2000–2020. The difference in temperature trend is
due primarily to changes in the ozone (Austin and
Butchart, 2003, Figure 5c). The globally averaged
values are also compared with results from a slightly
different version of the Unified Model (Butchart et al.,
2000), which indicated a very nearly linear trend in
temperature for a slow exponential growth in CO2
The UMETRAC results show a trend difference,
attributable to ozone change, which is not quite significant throughout a large domain but which is much
Pressure hPa
Pressure hPa
Pressure hPa
Copyright  2005 Royal Meteorological Society
more coherent and can be more readily explained than
the Hare et al. results. One would not expect a significant change throughout the middle stratosphere since
model ozone trends there are small. A strong negative signal exists in the lower stratosphere throughout
a range of latitudes from about 45 ◦ S to 60 ◦ N, implying larger cooling for the past than for the future.
There is also a coherent model signal over Antarctica,
implying stronger cooling in the past than predicted
for the future. The upper stratosphere throughout the
broad latitude range 45 ◦ S to 90 ◦ N shows stronger
cooling for the past than for the future. Except near
the equator this is statistically significant at the 95%
confidence level. This is a very basic result due to
the past chlorine change having a direct influence on
the ozone; yet in their Figure 3, Hare et al. indicate a
lower result than might be implied from Austin and
Butchart (2003) by a factor of two. Unfortunately, the
analysis of Austin and Butchart (2003) (as presumably is Hare et al.) was incomplete in not taking into
account the quasi-biennial oscillation (QBO) which is
interpreted as noise by the statistical package and can
mask the underlying trend. The Rayleigh friction run
of Austin and Butchart (2003) does not have a QBO
and trend uncertainties are consequently smaller in the
tropics (this comment, Figure 1). However, this run
has a different dynamical behaviour than the nonorographic gravity wave forcing run; so the ozone trend
was different in the lower stratosphere in southern
mid-latitudes (Austin and Butchart, 2003, Figures 5,
7, 8).
Rather than calculating a linear trend in temperature
throughout the 40-year period 1980–2020 as in Hare
et al., an alternative approach is to recognise the
change in the physical processes from about the
late 1990s as the ozone trend changes from loss to
recovery. This follows from the fact that the chlorine
rate of change slowed considerably after the year 1997
(Newchurch et al., 2003). Over the period 1980–2020
the resulting change of temperature evolution would
lead to increased uncertainty in the computed trends,
since the statistical package assumes temperature is
a linear function of time, and any divergence from
linearity is interpreted as noise.
In compensation, a 40-year period has a longer
data record. However, ignoring the linearity issue,
taking the simple model that ozone decreases from
1980 to 2000 at a rate α with uncertainty and
then remains constant to 2020 and assuming that
each year’s data are independent, over the 40-year
period the trend√would be α/2 and the uncertainty
approximately / 2. For the 20-year period, the trend
is α/ standard errors, while√for the 40-year period
the trend is approximately 1/ 2. α/ standard errors.
So, even assuming no recovery, it is more difficult to
detect a trend over the full 40 years, than over the
first 20, all else being equal. In some parts of the
atmosphere, the model predicts significant recovery
after 2000 thus making it even more difficult to detect
a trend over the longer period. These arguments apply
J. Austin
Figure 1. Zonal and annual average temperature trends
computed for the Rayleigh friction run of Austin and Butchart
(2003). (a) the period 1980 to 2000. (b) middle panel refers
to the period 2000 to 2020. (c)the difference in trends for
the two periods 1980–2000 minus 2000–2020. Light shading
indicates a value significantly different from zero at the 80%
confidence level. Dark shading indicates the regions of 95%
statistical significance
to the ozone trend, but the same argument would also
apply to the temperature trend, assuming CO2 has an
offset to the figures.
3. Dynamical consistency
Hare et al. raise the issue of dynamical consistency between the coupled and uncoupled simulations regarding the Antarctic ozone hole. However,
the absence of anything resembling the Antarctic signal in their Figure 3 is remarkable in comparison with
Atmos. Sci. Let. 6: 145–147 (2005)
Investigation of stratospheric temperature change
Austin and Butchart’s Figure 5 or Figure 1 in this
comment. Dynamical inconsistencies between the simulations of UMETRAC and imposed trends may be
more extensive and warrant further investigation. Perhaps the UMETRAC ozone results were imposed in a
way that is different to the coupled chemistry climate
model. For example, the diurnal tides in the coupled
and uncoupled models will not be identical. In the
tropics at 1 hPa, the coupled model will have about 4%
less ozone at noon than in the zonal average (Huang
et al., 1997), although I do not see why this should
necessarily affect the trend results. It is also quite conceivable that there has been a trend in transient wave
activity which would have affected ozone amounts but
which are unrepresented in the monthly averages used
by Hare et al. An increase in wave activity for example has been identified by Butchart and Scaife (2001)
as contributing to an increase in the strength of the
Brewer-Dobson circulation.
It is unclear to me how to apply offline chemistry in order to reduce costs. The fact that chemistry
and dynamics are coupled would mean that considerable manual intervention is needed while models are
running. The task of keeping models running continuously for many months of wall-clock time is already a
difficult enough task, requiring considerable organisation. In any case, for the Unified Model, the bulk of the
cost of the chemistry lies in transporting the tracers.
For the GFDL model, the tracer transport cost is somewhat less and the resources here are excellent for the
full task. The one area that could significantly reduce
costs is simply to do the chemistry on every other
horizontal gridpoint and interpolate in between. In a
massively parallel environment this requires careful
programming to ensure that results are independent of
the number and distribution of processors. This could
lead to a very economical model with only modest loss
of detail.
4. Cost analysis
For small ensemble sizes, the total cost of the coupled
and uncoupled model simulations are not vastly different, although the details will vary according to the
individual climate model. Taking the (GFDL) climate
model (with specified SSTs) as unit cost 1 and the
coupled chemistry model as unit cost 2.2, four ensemble runs require a total expenditure of 6.2 (one coupled chemistry run and four uncoupled runs). For 5%
more resources, three coupled runs would be available.
The uncertainty
√ in the trends would presumably scale
inversely as n − 1, where n is the number of runs.
Hence four uncoupled runs would have a trend uncertainty only about 20% lower than for the three coupled
runs. In practice, the uncertainty reduction may well be
greater as different chemistry runs might visit different chemical regimes introducing increased variability
than would occur in an ozone specified ensemble. So,
statistical uncertainty might be less in uncoupled runs,
but this a false sense of progress as there would still
be confusion surrounding predicted future atmospheric
5. Chemistry simplifications
Simplifying the chemistry could also be problematic.
Parameterised chemistry such as Cariolle and Déqué
(1986) sounds attractive, in principle, but with such
parameterisations you cannot be certain that the model
will represent the correct physics, and it may proceed
along its own independent trajectory. The authors may
care to inspect Figure 9 of Austin et al. (2003). The
outlyer in the upper panel is the only model with
parameterised chemistry. It really depends on what
risks one is prepared to take.
Copyright  2005 Royal Meteorological Society
The UMETRAC simulations reported here and in Hare et al.
were completed at the UK Meteorology Office and partially
funded by the CEC Framework 5 EuroSPICE project. Neal
Butchart provided some helpful comments on the text and
Sian-Jian Lin clarified some of the details of the GFDL tracer
transport technique.
Austin J, Butchart N. 2003. Coupled chemistry-climate model
simulations for the period 1980 to 2020: ozone depletion and the
start of ozone recovery. Quarterly Journal of the Royal Meteorolical
Society 129: 3225–3249.
Austin J, Shindell D, Brühl C, Dameris M, Manzini E, Nagashima T,
Newman P, Pawson S, Pitari G, Rozanov E, Schnadt C, Shepherd TG. 2003. Uncertainties and assessments of chemistry-climate
models of the stratosphere. Atmospheric Chemistry and Physics 3:
Butchart N, Scaife AA. 2001. Removal of chlorofluorocarbons by
increased mass exchange between the stratosphere and troposphere
in a changing climate. Nature 410: 799–802.
Butchart N, Austin J, Knight JR, Scaife AA, Gallani ML. 2000. The
response of the stratospheric climate to projected changes in the
concentrations of the well-mixed greenhouse gases from 1992 to
2051. Journal of Climate 13: 2142–2159.
Cariolle D, Déqué M. 1986. Southern hemisphere medium-scale waves
and total ozone disturbances in a spectral general circulation model.
Journal of Geophysical Research 91: 10 825–10 846.
Hare SHE, Gray LJ, Lahoz WA, O’Neill A. 2005. On the design
of practicable numerical experiments to investigate stratospheric
temperature change. Atmospheric Science Letters 6: 123–127. DOI:
Huang FT, Reber CA, Austin J. 1997. Ozone diurnal variations
observed by UARS and their model simulation. Journal of
Geophysical Research 102: 12 971–12 985.
Newchurch MJ, Yang ES, Cunnold DM, Reinsel GC, Zawodny JM,
Russell JM III. 2003. Evidence for slowdown in stratospheric ozone
loss: first stage of ozone recovery. Journal of Geophysical Research
108. DOI: 10.1029/2003JD003471.
Atmos. Sci. Let. 6: 145–147 (2005)
Без категории
Размер файла
101 Кб
asl, 107
Пожаловаться на содержимое документа