Developing Perturbations For Climate Change Impact Assessments
Eos,Vol. 84, No. 35, 2 September 2003
VOLUME 84
NUMBER 35
2 SEPTEMBER 2003
EOS,TRANSACTIONS, AMERICAN GEOPHYSICAL UNION
PAGES 337–348
mining frost occurrence, etc.As GCM grid
Developing Perturbations for
cells are aggregated, however, one finds that
the GCM output and the comparable observa-
tional data begin to converge.At the final level
Climate Change Impact
of aggregation, the global mean GCMs have
been shown to track temperature well even
Assessments
over the last century.
Nonetheless, for the unaware, the simplest
means of obtaining a regional climate change
PAGES 337, 341
The necessity for downscaling then becomes
scenario is to use the values of the GCM grid
an activity justified on the basis of needs and
cell most co-located with the region of interest,
possibly interpolating this further to a point
Following the 2001 Intergovernmental Panel
understanding not addressed through initial
location.Apart from the fact that GCM grid
on Climate Change (IPCC) Third Assessment
sensitivities based on the large-scale perturba-
cells are area averages and not point values, the
Report [TAR; IPCC, 2001], and the paucity of
tions.This assessment of the first-order sensi-
GCM grid cell value is typically well below the
climate change impact assessments from
tivity of a regional response develops the
skill scale of the model,especially with regard to
developing nations, there has been a significant
essential base understanding for further
precipitation—the variable most often required.
growth in activities to redress this shortcoming.
impact studies,and is as important in developed
This exacerbates the uncertainty associated
However, undertaking impact assessments (in
as well as developing nations.
with any derived regional-scale climate change
relation to malaria, crop stress, regional water
For climate change issues, there is much
scenario, and reduces the value of any impact
supply, etc.) is contingent on available climate-
uncertainty implicit in the choice of GCM
assessment intended for developing policy and
scale scenarios at time and space scales of rele-
and greenhouse gas forcing scenarios; an
adaptation strategies.
vance to the regional issues of importance.
uncertainty further compounded by the
Presented here is a simple approach for
These scales are commonly far finer than
additional complications introduced by
regional scenarios appropriate to the user
even the native resolution of the Global Cli-
downscaling.This further supports the argu-
needs described above. Using GCM simulation
mate Models (GCMs) (the principal tools for
ment that initial steps in impact studies are
output of future climate change, at spatial and
climate change research), let alone the skillful
better served by regional-scale scenarios that
temporal scales more associated with the
resolution (scales of aggregation at which GCM-
are plausible perturbations of the current
GCM skill resolution, one may develop “guided
observational error is acceptable for a given
climate, and which access the future climate
perturbations”—perturbations to baseline
application) of GCMs.
envelope, sidestepping factors such as optimal
observational data that are guided by, or in
Consequently, there is a growing demand for
downscaling solutions.To this end, a method-
accordance with, the GCM large-scale anomaly
regional-scale scenarios, which in turn are
ological approach that produces climate per-
under future climates.This allows the impacts
reliant on techniques to downscale from GCMs,
turbations that relate to the skill resolution of
researcher to investigate vulnerability and
such as empirical downscaling or nested
a GCM, and is thus guided by the large-scale
potential consequences of climate change at
Regional Climate Models (RCMs).These methods
response simulated by the GCM, may be a
the relevant regional scales,using a perturbation
require significant skill, experiential knowledge,
more robust means to address the initial
of the present climate that is guided by the
and computational infrastructure in order to
sensitivity questions in impact assessment
indicated change from the GCM. Further, this
derive credible regional-scale scenarios. In
research.
avoids potential errors associated with using
contrast, it is often the case that impact
A particular problem has arisen in this regard
raw GCM grid cell output, requires minimal
assessment researchers in developing nations
with the advent of the ready availability of
statistical or modeling expertise and
have inadequate resources with limited
GCM climate change products on the Internet.
infrastructure, and supports the development
access to scientists in the broader international
The provision of grid cell resolution GCM
of an initial understanding of regional sensi-
scientific community who have the time and
output invites the use of these at the GCM
tivity prior to drawing on more advanced
expertise to assist. However, where developing
grid scale. However, the skill resolution of the
downscaling techniques.
effective downscaled scenarios is problematic,
GCM is typically some spatial and temporal
it is possible that much useful information
aggregation of the native GCM resolution.“Skill”
can still be obtained for impact assessments
in this context may be considered the spatial
Methods
by examining the system sensitivity to larger-
and temporal scales of aggregation where the
The approach to developing GCM-guided
scale climate perturbations. Consequently, one
error between GCM output and observational
perturbations is conceptually simple, and
may argue that the early phase of assessing sen-
data does not preclude its use in a given appli-
readily undertaken with basic computational
sitivity and vulnerability should first be char-
cation. For example, comparison of a GCM grid
infrastructure and skills. Primarily, one seeks
acterized by evaluation of the first-order
cell value with point station observations is
to draw useful information from the GCM
impacts, rather than immediately addressing
not valid. Even after aggregating station data to
while accommodating the systematic bias of
the finer, secondary factors that are dependant
estimates of area averages comparable to a
different GCMs.The “useful information” here
on scenarios derived through downscaling.
grid cell average (a problematic process itself),
refers to the first-order (spatial large-scale)
one still finds significant errors between the
response of the GCM to greenhouse gas forcing.
GCM and the observations for parameters
This has the immediate benefit of avoiding
BY BRUCE HEWITSON
such as the number of rain days, or for deter-
the problems associated with the validity of
Eos,Vol. 84, No. 35, 2 September 2003
Fig. 1. HadCM3 30-year precipitation climatology (mm/day) for the control climate (a), and smoothed with a spatial moving average (b). Panel (c)
shows the bi-cubic spline surface fitted to the smoothed data of panel (b).
single-grid cell values.In essence,this approaches
For precipitation this may then further be expressed
anomaly.This may be applied to observational
the regional climate change question from
as a percentage change (anomaly divided by
data at any spatial scale—from station obser-
the view that there is a spatial large-scale change
the GCM control climatology), which is a
vations to gridded products.Temporal scales
signal that is closer to the skill resolution of the
typical treatment of precipitation anomalies
should be limited to some time average greater
GCM than individual grid cells, and which
in climate change studies (conversely, temper-
than daily (where GCMs have low skill in
provides a first-order anomaly for an impact
ature is more typically treated in terms of
reflecting observed climate variability).
assessment sensitivity study. It is explicitly
absolute anomaly). In this discussion we
In this example, the focus is on the DJF
recognized that there will be additional local-
focus on precipitation,which is commonly the
seasonal mean; and subjectively, it is suggested
scale variance of the climate change signal, but
variable of greatest concern. Figure 2 shows
that monthly means are likely to be close to
this remains the subject for subsequent and
the climate change percentage anomaly for
the finest temporal resolution one should use.
more refined impact assessment research.
DJF for two GCMs, the HadCM3 and the
In the example here, the GCM percentage
Inherent in using the GCM climate change
ECHAM4-OPYC GCMs, using the bi-cubic,
anomalies are applied to an experimental,
output is that significant uncertainty arises
spline-fitted surface.
high-resolution, 10-km gridded precipitation
from differences in GCMs.This is due in part
3. Perturb the observational climatology by
climatology for South Africa, demonstrating
to differing model physics parameterization
the fractional change indicated by the GCM
the application of the perturbation on a fine
of sub-grid scale processes.These differences
spline-fitted surface of the large-scale percentage
spatial resolution for a developing nation.
are reflected in various systematic biases and
climate sensitivities to greenhouse gas forcing.
The inter-model differences are important, and
reveal a measure of the uncertainty envelope
of projected future climate change. However,
the systematic bias and spatial errors in GCMs
obscure the climate change signal.The simple
methodology outlined here addresses these
issues in a manner appropriate to the objectives
outlined earlier for early phases of impact
assessments. Only a brief outline of the proce-
dural steps is presented, as the method is
simple. Full details are available in the support
documentation for the Africa guided pertur-
bation products on the Internet (available
late-2003,see http:// www.csag.uct.ac.za/AIACC).
The summary procedure,however,is as follows:
Fig. 2. Precipitation percentage change for DJF from the spline surface fitted to the current and
1. Derive a large-scale spatial response sur-
future climate means for (a) HadCM3 and (b) ECHAM4. Dashed lines are negative.
face from the GCM using a simple spatial filter.
In this example, a spatial 3 x 3 grid cell moving
average is used on the control and future GCM
simulation output. Figure 1a shows the un-
modified GCM 30-year current climate
climatology (1970–1999) from the HadCM3
GCM over southern Africa for the December-
February (DJF) season.The same field smoothed
with a 3 x 3 spatial smoother (panel b), and
the large-scale response surface using a bi-
cubic spline is applied to the data in panel (c).
2. Express the climate change anomaly
as the difference between the large-scale
response (spline) surfaces of the present and
future climate simulations, thereby removing
systematic bias which affects both the seasonal
Fig. 3. Precipitation anomaly (mm/day) of the DJF baseline climatology using the GCM percentage
and spatial magnitudes of the anomaly and
change.This represents a guided perturbation of the current observed climate, guided spatial
facilitating comparison between models.
large-scale anomaly from (a) the HadCM3 and (b) the ECHAM4 models.
Eos,Vol. 84, No. 35, 2 September 2003
The final GCM-derived perturbation of the
Such agreement does not necessarily indi-
in accord with the GCM first-order response
baseline climate (GCM percentage anomaly
cate that the models are right in their future
to greenhouse gas forcing. Using this approach
x baseline observed climate) is shown in the
projection of climate, but does suggest that
with a range of GCMs allows one to undertake
panels in Figure 3.
there is some common response, giving credi-
an assessment of fundamental regional sensi-
bility to the plausibility of the anomaly. For
tivities to climate change that are not arbitrary,
but guided by the envelope of future climate,
Assessment and Caveats
impact assessment research, this is exactly
what is needed in initial studies; namely, that
as characterized by GCMs.The approach is
By using the GCM percentage change, the
one has a plausible, credible perturbation on
computationally simple and appropriate for a
derived absolute magnitude of the perturbation
which to develop initial understanding of the
broad range of researcher sectors.The procedure
is naturally a function of the magnitude of the
regional sensitivities to climate change forcing.
is equally applicable to large areas and for
baseline climatology. Consequently, when
Thus, assuming that regional climate bound-
single station time series, and lends itself to
comparing Figures 2 and 3, it is apparent that
aries do not undergo dramatic lateral shifts,
data-sparse regions, as commonly found in
the spatial expression of the derived anomaly
the results suggest that this approach provides
developing nations. Hence, especially for many
has differences even where the spatial com-
a future climate perturbation at spatial scales
developing regions (although not excluding
ponent of GCM large-scale anomaly field is
appropriate for initial sensitivity studies in a
impact assessment work in developed nations),
the same for different locations.The perturba-
range of impact assessment activities. Related
this approach serves to provide a first look at
tion thus captures the spatial differences of
to this is the important requirement that the
the regional climate change envelope.
the existing climate,while maintaining agreement
GCM does not misplace, or fail to resolve, fun-
with the large-scale response of the GCM, the
damental physical climate boundaries. For
scale better associated with the model skill, as
example,if the GCM allows one climate domain
Reference
opposed to the single-grid cell values. Of note
(say, maritime) to extend into the adjacent
here is that the two models in Figure 3, which
but different climate region (say, continental
IPCC, Climate Change 2001: the Scientific Basis, J.T.
have notable differences in their control cli-
arid zones), this will result in inappropriate
Houghton, Y. Ding, and M. Nogouer (eds.), pp. 881,
matologies (not shown), indicate a degree of
application of the GCM anomaly in that region.
Cambridge University Press, 2001.
convergence in the regional anomaly pattern
This serves to highlight the need to carefully
as derived here. Both models clearly indicate
evaluate the GCM fields prior to application.
similar west-east patterns of wet-dry
Finally, although this is not a downscaled
Author Information
anomalies, and the HadCM3-derived anomaly
product, and does not include local feedbacks
is, to a large extent, very similar to a “dry-shifted”
and other forcings under future climates, it
Bruce Hewitson, ENGEO Department,University of
ECHAM4-derived anomaly.
does represent a regional-scale perturbation
Cape Town, South Africa; E-mail: hewitson@egs.uct.ac.za
on mantle dynamics as continental breakup
Geophysical Project in Ethiopia
proceeds to sea floor spreading, but there
remain fundamental questions regarding the
three-dimensional distribution of strain and
Studies Continental Breakup
melt as continents rift apart.
Our approach in EAGLE is to examine the
nature of crust and upper mantle along a
PAGES 337, 342–343
will be presented in a session at the “The East
highly extended,magmatically active continental
African Rift System: Development, Evolution
rift prior to the modifying effects of post-rift
and Resources” Meeting to be held in Addis
sedimentation, erosion of the uplifted rift
As continental rift zones evolve to sea floor
Ababa in June 2004.The lead Ethiopian insti-
flanks, and thermal decay. In the Ethiopian
spreading, they do so through progressive
tutions were the Geophysical Observatory, the
Rift we can (1) trace the evolution from broadly
episodes of lithospheric stretching, heating,
Department of Geology and Geophysics of
distributed to focused strain during rift devel-
and magmatism, yet the actual process of
Addis Ababa University,the Ethiopian Geological
opment; and (2) study the active processes of
continental breakup is poorly understood.
Survey, and the Petroleum Operations Depart-
continental breakup associated with a mantle
The East African Rift system in northeastern
ment of the Ethiopian Ministry of Mines.The
plume (or other upper mantle convective
Ethiopia is central to our understanding of
lead European and U.S. institutions were the
upwelling),while avoiding interactions between
this process, as it lies at the transition between
universities of Leicester, Royal Holloway Lon-
subducted slabs and asthenospheric flow; the
continental and oceanic rifting [Ebinger and
don, Leeds, and Edinburgh, together with
region has been tectonically stable since 600 Ma.
Casey, 2001].
Stanford, the University of Texas, El Paso,
We are exploring the kinematics and
Southwest Missouri State, and Penn State uni-
Ethiopia-Afar Rift Zone
dynamics of continental breakup through the
versities.The entire project was coordinated
There is general agreement that the broad
Ethiopia Afar Geoscientific Lithospheric
in Ethiopia by the Commission of Science
uplifted Ethiopia-Yemen plateau and Oligocene
Experiment (EAGLE), which aims to probe
and Technology of the Democratic Republic
flood basalt province have been affected by
the crust and upper mantle structure between
of Ethiopia.
one or more Cenozoic plumes [e.g., Nyblade
the Main Ethiopian (continental) and Afar
and Langston, 2002].A synthesis of 40Ar/39Ar
(ocean spreading) rifts, a region providing an
Models for Continental Breakup
data shows that flood basalts were erupted
ideal laboratory to examine the process of
across an ~1000-km diameter region at ~31
breakup as it is occurring. EAGLE is a multidis-
The three-dimensional structure of oceanic
Ma, presumably coincident with plume head
ciplinary study centered around the most
rifts is primarily controlled by the supply of
contact with Afro-Arabian lithosphere [e.g.,
advanced seismic project yet undertaken in
magma [e.g., Phipps-Morgan and Chen, 1993],
Hofmann et al., 1997]. Previous geophysical
Africa (Figure 1). Our study follows the Kenya
whereas that of youthful continental rifts is
studies show crustal thinning northward into
Rift International Seismic Project [e.g., KRISP
controlled by the spatial arrangement of large
the Afar depression. Refraction profiles in
Working Group, 1995], and capitalizes on the
displacement border faults [e.g., Hayward and
Afar, interpreted as near-one-dimensional
IRIS/ PASSCAL broadband seismic array
Ebinger, 1996].Thus, magmatic processes
structures due to the very small number of
[Nyblade and Langston, 2002], providing a
increase in importance as rifting proceeds to
shots and receivers, suggest thinned 25-km-
telescoping view of the East African Rift within
sea floor spreading, but there is no consensus
thick crust underlain by a 10-km-thick layer
this suspected plume province.
as to when or how this transition occurs.The
with anomalously low upper mantle P-wave
EAGLE fieldwork was undertaken between
volume of melt produced and its seismic
velocities above apparently normal mantle
October 2001 and March 2003. Many results
velocity structure provide critical constraints
[Berkhemer et al., 1975].
