View
215
Download
0
Embed Size (px)
Citation preview
Development of GCM Based Climate Scenarios
Richard Palmer, Kathleen King, Courtney O’Neill,
Austin Polebitski, and Lee Traynham
Department of Civil and Environmental EngineeringUniversity of Washington
December 13, 2006
Objective
Develop Future Climate Variable Database for Consistent Evaluation in the Region
Approach Take Global Climate Model output and
refine to local scale through a downscaling method
Downscaling Method Requirements Maintain local characteristics while
acknowledging changes in larger scale state The downscaling method must account for:
Effects of underlying climate trends (ie., warming) Effects of interannual variability (consecutive years
can be very different)
Seeking method that: Preserves full range of historic observed variability Creates steady-state representation of future climate
3 Stages to Develop Local Climate Variables
1) Downscale climate variables from the GCM scale grid to a regional scale grid
2) Bias-correct a single regional grid cell to an individual station location
3) Expand the station scale transient scenario into multiple, quasi-steady-state time scenarios with the full historic variability
Downscaling in a Nutshell
-6
-4
-2
0
2
4
6
8
10
0 0.2 0.4 0.6 0.8 1 1.2
Non-Exceedance Probablility
Ja
nu
ary
Av
era
ge
Te
mp
era
ture
(C
)
HadCM3 Cell (47.5, -120.0)
Regional Cell (47.5625,-121.8125)
-6
-4
-2
0
2
4
6
8
10
0 0.2 0.4 0.6 0.8 1 1.2
Non-Exceedance Probablility
Ja
nu
ary
Av
era
ge
Te
mp
era
ture
(C
)Regional Cell (47.5625,-121.8125)Snoqualmie Falls
Stage 1
Downscale climate variables from the GCM scale grid to a regional scale grid
Downscale from GCM to Regional Scale
Downscaling takes us from 107 km2 to a regional scale of 104 km2
Overview of Stage 1 Downscaling Process
Bias-correctionDownscale
CDFTransfer Function
RegionalGlobal
Quantile Mapping
Develop Transfer Functions from Historic Climate Simulation from GCM and Historic Observed Data
Use Transfer Functions developed to Bias-Correct Future Climate Output
Downscale Bias-Corrected GCM output to finer scale
CDF – cumulative distribution function,
Develop Transfer Functions and Bias-Correction Monthly temperature and precipitation
CDF calculated for same historic periodEach grid cell in GCM Each grid cell at regional scaleThe GCM and regional scale CDFs are used
to derive a set of transformation functions The process of relating the CDFs is
generically referred to as “Quantile Mapping”
Develop Transfer Functions and Bias-Correct Quantile mapping method is based on a bias correction
scheme for downscaling climate model output Assumes that shifts in climate variables occur with different
magnitudes at different points along the distribution Temperature and precipitation simulated by the climate
model are then bias corrected using the transfer function
-10-8
-6-4-2
024
68
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Non-Exceedence Probability
Jan
uary
TE
MP
Cedar
GFDL_R30
Downscale The bias-corrected model is downscaled and
disaggregated The bias-corrected model data are sampled onto the
1/8° grid The mean difference between the bias-corrected model
and the 1/8° data for each calendar month during the time period (1950-2000) is computed to form a perturbation factor
The factor is added to the monthly simulated variable of the simulated scenario Temperature Precipitation
Stage 1 Output
The output from Stage 1 is transient, monthly time-series at the 1/8° scale of GCM simulated climate
The daily, transient, regional climate grid is then be used as forcings in regional scale hydrologic models or further downscaled to specific locations
Stage 2
Bias-correct a single regional grid cell to an individual station location
Transformation from regional grid to station locations Data from the regional grid can be further downscaled to
individual weather station locations by an additional application of the Quantile Mapping method
Downscale to Location
The monthly transformation relationships are defined Historic climate CDFs from the regional cell to CDFs
from the observed station data Future regional scale climate data is downscaled
to the station location though use of developed relationships The difference (bias) between the regional gridcell
value and the station record tends to be considerably smaller than the bias seen when comparing GCM scale cells to regional cells
Stage 2 Output
A bias-corrected, transient, monthly GCM time series for each station location of interest
The output from this process is used to:Examine climate trends at the station scale Examine transient hydrologic phenomena
generated using a high resolution hydrologic model
Create Quasi-Steady-State Long-term Time Series
Stage 3
Create Expanded Time Series
Expanding Transient Time Series into Quasi-Steady-State Time Series Climate is defined as the average condition of the
weather over a period of time Assumes that the climate state being defined is stationary (Long
term average does not change over time) These averages do change
Influenced by the range of time Range of natural variability is often greater than the
magnitude of change expected over several decades This is NOT to imply that climate change impacts are
insignificant Need to include the full range of potential variability in
any estimate of future climate change
Extreme Events
Extreme events are the defining events when describing the sustainability of a water resource
It is important to include these events in any representation of potential future climate
Steady-state vs. Transient
By using a steady state approach to estimate climate conditions, it is likely that a significant amount of potential variability will be excluded
If a transient scenario is used (examining the entire time series) then it becomes hard to see the potential impacts of climate change at a specific point in time Each simulation is only a single realization of the
infinite number of possible combinations of events
Solution
Incorporate a step into the downscaling process that expands the climate time series so that it includes the full range of observed historic variability by creating an expanded time series
Expanded Time Series
Uses a quantile relationship similar to quantile mapping to develop transfer functions
Combines the climate variable distributions derived from one data subset with time series of events from different subset
Allows use of a shorter period to define the climate state, yet maintains the variability of the full historic record
Creating an Expanded Time Series
1. A 31-yr slice, centered on the Year of Investigation is extracted from the transient GCM data
These 31 years are considered indicative of the average climate for that period, so the climate of 2050 would be described by the years 2035-2065
2. Bias-corrected, transient, monthly GCM time series is divided by climate variable and by month into 24 (12 months x 2 variables) climate progressions
Creating an Expanded Time Series
3. Create CDFs of extracted climate data and aggregated historic observed data. Develop Quantile Maps between historic observed and GCM CDFs.
4. Output from mapping is historic time series shifted by GCM based climate. This is compared to historic monthly CDFs.
The differences in temperature and precipitation are computed as the difference in temperature (dT) and the quotient of the precipitation (dP).
The result is a full time series of monthly dT and dP values
Creating an Expanded Time Series
5. The monthly dT, dP time series is applied to the daily station level time series
dT values are added to temperatures dP values are multiplied by daily precipitation6. The output of this step is a daily time series of
temperature and precipitation that has the range of variability seen in the historic record, but also has the long-term climate properties of the GCM
Expanded Time Series
This procedure captures the climate change signal from the GCM with the shifts in the climate variable CDFs, while also creating a series that contains all of the extreme events in the observed record
Advantages of an Expanded Time Series The long-term climate trends from the GCM
are removed so that the station scale data set contains a long climatic sequence that is not complicated by the presence of an underlying trend
Instead it is a steady-state approximation of the climate during a window of time that contains the full range of potential variability
Climate Change
Changes in climate are unlikely to occur as a uniform shift in values In fact- they are highly non-linear
Current impact assessments use delta methodThese rely only on changes in the means of
climate variables to fully describe the range of potential impacts
New Method Improves Upon the Delta Method Allows for differential shifts in climate
variables at different rates of change at the extremes of climate distributionMonthly climate means simulated by GCMsProbability of extreme events
Why Use the Expanded Time Series Approach? The examination of climate change impacts to
water resources must be targeted to specific future periods Difficult due to combined effects of a constantly
shifting underlying climate trend and large year to year variability
System impacts are best described using a long time series that incorporates the full range of potential variability and represents a steady state approximation of climate as defined for a chosen future period
Why Use the Expanded Time Series Approach? The quantile mapping process used with
an expanded historic time series reproduces the desired statistics of the target time period while providing the length and variability of record needed for most system reliability assessments
Conclusion
This method is most appropriate for application to a water resources evaluation where: Natural variability can strongly affect system
performance Small changes in extreme events can have a
much larger impact than changes in the long-term means
Future Climate Variable Database
Goals of Website:
Disseminate Climate Data:Access to historic climate dataAccess to projected data from GCMsAbility to view trends from projected GCM
data graphically through simple manipulations on website
Data Sets – Overview Regions:
WRIA 7,8,9,and 10 Models:
IPSL A2 (‘Pessimistic’) GISS B1 (‘Optimistic’) ECHAM A2 (‘Average’)
Years 2000 2025 2050 2075
Climate Variables Temperature (Daily Minimum and Maximum) Precipitation Data (Daily Total)
Regions
29 Stations 5 regions
Northwest Southwest Central Southeast Northeast
Home Page
Acquiring Data
Step 1: Select a Region
What is This?
A ‘What is This?’ icon helps users navigate to data of interest and explain data available and format
Regional Divisions
Step 2: Select a Station
Step 3: Select a Scenario/Year
Step 4: Select a Model
Step 5: Download Data
Access to Raw Data
Downloadable as text or excel file by station
Available data consists of daily Tmin, Tmax (°C), and Precipitation (mm) values
Graphical Display of Trends
Look at Projected Trends
Through selection of region, station, climate variable, and GCM, user will be able to create graphical displays of projected trends
Generating Graphs
Monthly statistics for climate variables are easily visualized and navigated by users
Preliminary Results of the Downscaling Process
IPCC 2001: http://www.grida.no/climate/ipcc_tar/wg1/figts-17.htm
Scenarios Used in GCMs
Departure from Historic Temperature - ECHAM5
-2
-1
0
1
2
3
4
5
6
7
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
Mo
nth
ly A
vera
ge
Tem
per
atu
re (
Deg
rees
C)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
Departure from Historic Temperature- GISS
-2
-1
0
1
2
3
4
5
6
7
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
His
tori
c A
vera
ge
Mo
nth
ly T
emp
erat
ure
(d
egre
es C
)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
Departure from Historic Temperature - IPSL
-2
-1
0
1
2
3
4
5
6
7
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
His
tori
c A
vera
ge
Mo
nth
ly T
emp
erat
ure
(d
egre
es C
)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
Departure from Historic Temperature - Average of GCMs
-2
-1
0
1
2
3
4
5
6
7
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
Ave
rag
e M
on
thly
His
tori
c T
emp
erat
ure
(d
egre
es C
)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
USGS Palmer Station – Just Below Howard Hanson Dam
Departure from Total Monthly Precipitation - ECHAM5
-2
-1
0
1
2
3
4
5
6
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
To
tal A
vera
ge
Mo
nth
ly P
reci
pit
atio
n (
inch
es)
2000
2025
2050
2075
Historic
Departure from Total Monthly Precipitation - GISS
-2
-1
0
1
2
3
4
5
6
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
To
tal A
vera
ge
Mo
nth
ly P
reci
pit
atio
n (
inch
es)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
Departure from Total Monthly Precipitation - IPSL
-2
-1
0
1
2
3
4
5
6
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
To
tal A
vera
ge
Mo
nth
ly P
reci
pit
atio
n (
inch
es)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
Departure from Total Monthly Precipitation - Average of GCMs
-2
-1
0
1
2
3
4
5
6
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month
Dep
artu
re f
rom
To
tal A
vera
ge
Mo
nth
ly P
reci
pit
atio
n (
inch
es)
2000
2025
2050
2075
Historic
USGS Palmer Station – Just Below Howard Hanson Dam
Monthly Streamflows Forecasted w/ ECHAM5 Howard Hanson Inflow
0
500
1000
1500
2000
2500
3000
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month of Year
Flo
w R
ate
(cfs
)
DHSVM Historic
ECHAM5 2000
ECHAM5 2025
ECHAM5 2050
ECHAM5 2075
Monthly Streamflows Forecasted w/ GISS Howard Hanson Inflow
0
500
1000
1500
2000
2500
3000
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month of Year
Flo
w R
ate
(cfs
)
DHSVM Historic
GISS 2000
GISS 2025
GISS 2050
GISS 2075
Monthly Streamflows Forecasted w/ IPSL Howard Hanson Inflow
0
500
1000
1500
2000
2500
3000
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month of Year
Flo
w R
ate
(cfs
)
DHSVM Historic
IPSL 2000
IPSL 2025
IPSL 2050
IPSL 2075
Monthly Streamflows Forecasted w/ All GCM's Howard Hanson Inflow
0
500
1000
1500
2000
2500
3000
Octobe
r
Novem
ber
Decem
ber
Janu
ary
Febru
ary
Mar
chApr
ilM
ayJu
ne July
Augus
t
Septe
mbe
r
Month of Year
Flo
w R
ate
(cfs
)
DHSVM Historic
Average 2000
Average 2025
Average 2050
Average 2075