Downscaling as a contributor to developing robust …...Downscaling as a contributor to developing...
Transcript of Downscaling as a contributor to developing robust …...Downscaling as a contributor to developing...
Downscaling as a contributor to developing robust messages of change
Bruce Hewitson
Information framework
Limitations of current offering
A case study example
The multi-model challenge
CORDEX and AR5
A: The Information Framework
Pick as role as a stakeholder trying to accommodate climate change
What has already changed?
Is that any different from variability?
What is the future?
When is the future?
How do you know that?
Where do you get your information?
Do you “believe” it?
How do you know how good it is?
Would you spend your own money based on this information?
At the root of the issue, what level of information do you trust?
DataClimate models, historical
observations, trends,
downscaling, projections, event
frequency, …
InformationMeasures of vulnerability and
risk, threshold exceedence,
combinatory impacts,
uncertainty and confidence,
regional scale variations, …
KnowledgeAssessing options,
understanding consequences,
evaluating responses, informing
decision making, …
A basis for actionPolicy development to balance
competing priorities, strategic
investments in adaptation and
mitigation, new research
avenues, coordination of
response frameworks, …
Generated by models,
analyses, downscaling,
observations …
We are not always sure
when we have “information”
Comes with close coupling
between science and society
Actions are risky, and takes
place within a multi-stressor
context
De
live
red
by s
cie
nce
Need
ed
by
so
cie
ty
WG
1 Obs & past
trend
Circulation
changes
Framework for integrating information on the physical climate system
for regional impacts and adaptation needs
Regional Integration and Understanding
Data products with
articulated uncertainty
Storylines and robust
messages of changeWG
1 /
WG
2-C
h 2
1
Adapted from Hewitson et al., 2010
Contextualization
around real world
questions
WG2 Part A
WG2 Part B
WG
2
Stakeholders
Downscaled
change
GCM
change
General classes of downscaling
Local climate = f (larger scale predictors) + locally forced variance
Dynamical
Two approaches
Empirical-statistical
Three main classesPerturbed observed
RCM Hi-res GCM
Weather Generators Transfer Functions
Trained on long term
time series and
atmospheric re-analysis
data, conditioned by
GCM parameters to
capture low frequency
variance
Trained on time series
that spans range of
variability, residual local
scale variance added
stochastically
Index / analogues
Requires long term data
sets and uses weather
typing or historical
analogues
Even then … it’s not only about the data!
Users face multiple dependencies in seeking value for
adaptation decision making
Articulation
of relevant
thresholds
Understanding
natural
variability
Effective
communication
between
knowledge
provider and
user
Tailored
information
products
Quantified
uncertainty
Iterative and
sustained
re-examination
Synergy between
process change and
local change
Accommodation of
feedbacks and tipping
points
Assessment of
error
Balancing
multi-stressor
factors
Etc …
Downscaling
Information source
(AO)GCMs – CMIP3
Downscaling – RCMs/SD
Process changes
Historical changes
User communities
Research scientists
Policy / mitigation
Vulnerability / Impacts
Adaptation
Cre
atio
n
Tra
nsfo
rma
tio
n
Inte
rpre
tatio
n
Each source has different:
- attributes of signal and noise
- limitations on interpretation
- degrees of uncertainty
- methodologies of evaluation
Each community has different:
- definitions / terminology
- priorities of need
- scales of interest
- access to information
What we would like to accomplish …
B: The Confusion of offerings
A proliferation of portals and data sets, with poorly articulated
uncertainties, weakly explained assumptions and dependencies,
data implied as information, and communicated to a user
community poorly equipped to understand the limitations
GCM data presented for local application
Fig 3: From Tabor & Williams, 2010
The proliferation of change factor “downscaling” as credible local scale information
Often users do not grasp the multiple assumptions and dependencies, and the fact that the data is not robust at fine scales.
Large vulnerability on change factor of precipitation at GCM grid cell resolution
Multiple portals springing up based on the GCM change factor approach
1
2
3 4
1
2
3 4
1
2
3 4
1
2
3 4
Typical statistical
downscaling
products at the
time of the AR4,
for Africa
Usually spatially
discontinuous
with weak spatial
cohesion
Significant degree of “noise” in RCM
downscaling
Robustness of Future Changes in Local
Precipitation Extremes. (Kendon et.al. 2008)
Examines the contribution of natural variability to
the spread of future climate projections of
precipitation extremes
“In general, where climate noise has a significant
component varying on decadal time scales, single
30-yr climate change projections are insufficient to
infer changes in the extreme tail of the underlying
precipitation distribution.”
FIG. 1. Percentage changes
in extreme precipitation in the
three different ensemble
members of HadRM3H
[DJF – left, JJA – right]
C: An example of multimodel downscaling complications
Downscaling may not lead to convergence or information products
of value – new complications can be intrroduced
Zambia 13.55S 32.58E – downscaled control
Examples of where things go right and wrong
Zambia 13.55S 32.58E – downscaled anomaly
Examples of where things go right and wrong
Surat, India – downscaled control
Note the ok-looking downscaled GCM
And some not-ok
Examples of where things go right and wrong
Surat, India – downscaled anomaly
The “ok” downscaled
control climate GCM!
Examples of where things go right and wrong
Calcutta, India – downscaled control
Interesting clustering of
downscaled GCMs – not
so clear in raw GCM data
Examples of where things go right and wrong
Calcutta, India – downscaled anomaly
Examples of where things go right and wrong
D: An case study
Drawing on multiple lines of evidence to build a
message of change
AOGCM multi-model
projected changes in sea
level pressure and surface
winds
Sea level pressure multi-
model median anomaly
Surface wind multi-model median anomaly
Raw GCM projections: Wind speed
and direction90th percentile
Median“Best estimate?”
10th percentile
Raw GCM projections: rainfall
75th percentile
Median“Best estimate?”
25th percentile
AR4 multi-model
median anomaly:
(2045-2064) - control
GCMs perspective
mm/month: max change = ~15%
DownscaledAR4 multi-model
median anomaly:
GCMs downscaled
precipitation change
(2045-2064)
Downscaled
PROJECTION
GCM
Information source Message Discussion
Historical trends Core winter wetting dominantly in the mountains
Shoulder season drying
Marginal indications of a possible wetter summer
The region is spatially inhomogeneous in trend magnitude, although the dominant trends can be seen to greater or lesser degrees across the region
GCM changes in circulation / processes Increased subsidence due to a stronger mid-latitude high pressure inducing drying
Deeper thermal surface trough over the continent increasing west coast pressure gradient and possibly summer convection in the east
Poleward shift in mid-latitude flow decreasing frontal intensity
Increased longshore west coast wind promoting stronger upwelling, colder coastal waters, and consequent drying on the west coast.
The models are in good agreement on these large scale circulation changes, albeit with a range of differing magnitudes of change. The change further is physically consistent with the anticipated first order response of the climate system.
GCM grid cell changes General drying over the region
A weak suggestion of possible summer wetting in the north east
The models are in strong agreement on the drying message for the region, but it is clear that the spatial detail related to local scale topography is absent.
Local scale downscaled changes A general drying in the west with modest wetting to the east, modulated by the topography
Core winter wetting in the important water catchments in the core winter season
Small decreases in rainfall frequency in the west and small increases in the east
Changes in dry spell duration commensurate with the above changes.
The downscaled projected changes across all models are robust in spatial pattern although vary in magnitude, and the projected changes in some regions are too small to be of consequence. Of importance is the drying in regions of non-irrigated agriculture in the west, and while core winter wetting in the key catchments is indicated for the near term, later in the century this reverses. Taken with an increase in temperatures, the indication is for problematic increases in water stress.
Assess, distill, conclude, communicate a message:
E: CORDEX
For Africa at least, the promise of a major advance
Multiple GCMs + process change assessment
Multiple RCMs
Multiple statistical downscaling
Common experiment framework
NARCCAP
CLARIS
ENSEMBLES
RCMIP
+ polar regions From Colin Jones
CORDEX: changing the landscape of information?
Evaluating the first CORDEX results over AfricaColin Jones and Grigory Nikulin, Rossby Centre, SMHI
Closing thoughts
Downscaling is a key (but only one) foundation for
developing regional projections of change
Perhaps the biggest challenge relates to the
integration of multi-model multi-method results