Transcript lettenmaier_epa_climate_change_jan09

Where the Research Meets the Road: Climate
Science, Uncertainties, and Knowledge Gaps
Dennis P. Lettenmaier
Department of Civil and Environmental Engineering
University of Washington
First National Expert and Stakeholder Workshop
on Water Infrastructure Sustainability and
Adaptation to Climate Change
Arlington, VA
January 6, 2009
Issues in estimating (and planning
for) climate change impacts
• Design and management of most water-related
infrastructure is based on methods of assessing risk and
reliability (e.g., flood plain delineation based on 100-year
event; water system planning for 98 percent reliability)
• These methods are almost entirely based on analysis of
historic observations (e.g., fitting a flood frequency
distribution to an annual flood maximum series)
• These methods often are standardized (e.g., Water
Resources Council Bulletin 17b for estimation of flood
risk)
• Similarly standardized methods are not available when
the stationarity of historic observations cannot be
assumed
What methods are at our disposal for design
and planning in a changing climate?
•
•
•
•
•
•
Global climate models provide our best information as to future realizations
(time series) of climate
These is a scale mismatch (spatial, and perhaps temporal) that needs to be
bridged for water resources applications (simple ratio or difference methods,
statistical, dynamical)
There are considerable differences among GCMs, and these differences
need to be incorporated in planning as a representation of uncertainty in the
climate prediction process
Each GCM “scenario” is effectively an ensemble member from a particular
model (multiple ensemble members from a given GCM provide a measure
of another source of uncertainty, natural variability – this type of uncertainty
exists in a stationary world as well)
Multimodel ensemble approaches are well adapted to the risk and
uncertainty analysis required for water planning – these methods are
increasingly used in numerical weather prediction. One complication is that
lack of forecast verification opportunities in the climate prediction case.
The assessment process involves a modeling chain, starting with GCM
ensemble members
Climate
Scenarios
Global climate
simulations, next
~100 yrs
Hydrologic
Model (VIC)
Natural
Streamflow
Downscaling
Delta
Precip,
Temp
Performance
Measures
Reliability
of System
Objectives
Reservoir
Model
DamReleases,
Regulated
Streamflow
GCM Predicted Climate Change
Change in precipitation (%)
Change in precipitation and temperature for selected basins
40
30
20
10
0
-10
-20
-30
-40
40
30
20
10
0
-10
-20
-30
-40
40
30
20
10
0
-10
-20
-30
-40
0
Amazon
Amur
Mackenzie
Mekong
Mississippi
Severnaya Dvina
Xi
Yellow
Yenisei
1
2
3 4
5
GFDL_CGCM
CCCMA-CGCM1
2025
6
7
8
0
1 2
HCCPR-CM2
CCSR-CGCM
2045
3
4
5
6
7 8
Change in temperature (C)
2095
0
1
HCCPR-CM3
CSIRO-CGCM
2
3
4 5
6 7
8
9
MPI-ECHAM4
DOE-PCM3
BAU 3-run average
historical (1950-99)
control (2000-2048)
PCM
Business-as-Usual
scenarios
Columbia River Basin
(Basin Averages)
How can we best extract information for water
infrastructure design and planning from GCMs?
Some issues:
• Bias is a key issue, which must be addressed in any assessment
protocol
• All GCMs are not “created equal” – evaluations using current
climate demonstrates this
• Some GCMs perform better than others (in current climate
simulations), but these differences vary both geographically, and
depending on evaluation criteria
• How much effort should be allocated to generation of multiple
ensemble members for a given GCM, vs representation of the
largest possible number of GCMs (essentially comes down to the
role of inter-model uncertainty, vs natural variability)
• How to downscale?
Example of ensemble method
9000
cfs
7200
Week 22
5400
3600
1800
0
1
3
5
7
9 11 13 15 17 19 21
ensemble rank for the 2020s
Downscaling issues
• In general, statistical downscaling offers our best current hope of
integrating multimodel ensemble methods into water infrastructure
planning and management, based on computational issues alone
• Scale mismatch between the relevant spatial scale at which design
and planning analyses must be performed (e.g., for estimating of
reservoir inflows) and the GCM scale is not itself a justification for
dynamic downscaling
• Where there are key processes that affect hydrologic forcings
(especially precipitation) that are subgrid to the GCM, dynamical
downscaling may be necessary (example: urban stormwater design)
• Bias correction is an essential step whether or not dynamical
downscaling is used.
• There is also a a question as to whether statistical downcaling
methods are required rather than the so-called delta method – one
justification for the more complex statistical methods is transient, vs
“time slice”, analyses
Regional climate
models: The role of
spatial scale
Visual from NCAR
What needs to be done to incorporate climate
change into water planning in a routine
manner?
• Standardization of methods that lead to reproducibility of
results (a la Bulletin 17b) – very climate change study cannot
be a research project
• Standard archive data sets of climate model output, updated
periodically (IPCC archive has gone a long ways towards
meeting this requirement)
• A recognition of (and methods for dealing with) the fact that
climate change projections will be subject to change, albeit
periodically
• Movement away from critical period planning (analysis of the
worst period in a sequence of historical observations – e.g., the
worst drought)
What are the applications research needs?
• Need to better understand the elements of uncertainty
inherent in GCM ensemble members (is it possible to
represent inter-model uncertainty using resampling from a
single GCM)?
• Better archiving (and retrieval) of GCM output (especially
daily and shorter time scale output)
• Need for a more systematic approach to regional climate
simulation and downscaling (mostly computer time and
archiving!)
• A stronger push for applied research and development in
water resources, with specific emphasis on hydrologic change
(see ASCE JWRPM, Nov-Dec 2008)