Translating Global Models to the Local Scale

1
Translating Global Models to the Local Scale
Ed Maurer Civil Engineering Dept.
OCCRI Workshop Scenarios of Future
Climate October 28-29, 2009
2
California Water

• CA hydrology is sensitive to climate variations,
  climate sensitive industries (agriculture,
  tourism), 5th largest economy in world
• Water supply in CA is limited, vulnerable to T, P
  changes
• timing, location
• Changes already are being observed
• CA Executive Order supporting studies on climate
  change impacts

3
California Water Management

• 1400 dams
• gt1000 miles of canals and aqueducts
• SWP alone generates 5.8 billion kWh/yr
• SWP is Californias largest energy consumer (net
  user)
• Edmonston pumping plant biggest single energy
  user in state

4
Prospects for the Future

• Range of likely warming by end of 21st century
  variable
• By mid-21st century most differences smaller

1.8
3.4
4.0
2.4
2.8
5
Estimating regional impacts
2. Global Climate Model
4. Land surface (Hydrology) Model
1. GHG Emissions Scenario
5. Operations/impacts Models

• Downscaling

Adapted from Cayan and Knowles, SCRIPPS/USGS, 2003
6
Projecting Future Climate - GCMs

• Climate Models (GCMs) Necessary
• These have biases
• Spatial resolution
• Parameterization
• etc.
• At inappropriate scale for most impact analysis

7
Downscaling

• Dynamic
• Better representation of terrain captures local
  processes and feedbacks
• Computationally expensive
• Still contain biases
• Statistical
• Assumes stationary transfer function

8
BCSD Method BC

• At each grid cell for training period, develop
  monthly CDFs of P, T for
• GCM
• Observations (aggregated to GCM scale)
• Obs are from Maurer et al. 2002

• Use quantile mapping to ensure monthly statistics
  (at GCM scale) match
• Apply same quantile mapping to projected period

Wood et al., BAMS 2006
9
BCSD Method SD

• Use bias-corrected monthly GCM output
• Aggregate obs to GCM scale
• Calculate P,T factors relative to coarse-scale
  climatology
• Interpolate factors to 1/8 grid
• Apply to fine-scale climatology

Daily Values from rescaled historical values
10
Bookend Studies

• Brackets range of uncertainty
• Useful where impacts models are complex

11
Bracketing Future Warming for California
CA average annual temperatures for 330-year
periods Amount of warming depends on
our emissions of heat-trapping gases. Summer
temperatures increases (end of 21st century) vary
widely Lower 3.5-9 F Higher 8.5-18 F
Ref Luers et al., 2006, CEC-500-2006-077
12
Bracketing Future California PrecipitationStatewi
de Winter Average
Winter precipitation accounts for most of annual
total High interannual variability less
confidence in precipitation-induced changes than
temperature driven impacts.
Ref Hayhoe et al., 2004
13
Projected Impacts Loss of Snow

• Snow water in reserve on April 1
• Change (Sacramento-San Joaquin basin, 2 GCMs, 2
  emissions scenarios)
• -12 to -42 (for 20352064) (up to 1 Lake
  Shasta)
• -32 to -79 (for 20702099) (up to 2 Lake
  Shastas)

Ref Luers et al., 2006, CEC-500-2006-077
GFDL CM2.1 results
14
Availability of GCM Simulations

• 20th century through 2100 and beyond
• gt20 GCMs
• Multiple Future Emissions Scenarios

15
Which (and how many) GCMs to select?

• Most important to have ensembles of runs with
  enough realizations to reduce the effects of
  natural internal climate variability Pierce et
  al., 2009
• Little advantage to weighting GCMs according to
  skill

Source Brekke et al., 2008
16
Impact Probabilities for Planning
Snow water equivalent on April 1, mm

• Combine many future scenarios, models, since we
  dont know which path well follow (22 futures
  here)
• Choose appropriate level of risk

17
BCSD in mass production

• PCMDI CMIP3 archive of global projections
• New archive of 112 downscaled GCM runs
• gdo4.ucllnl.org/downscaled_cmip3_projections
• Allows quick analysis of multi-model ensembles

18
Use of U.S. Data Archive

• Approximately 400 unique users downloaded 2 TB
  of data
• From across US and outside
• Uses for Research (R), Management Planning
  (MP), Education (E)

19
What is missing from downscaled data archive?
20
Global BCSD

• Similar to US archive
• Allows probabilistic representation of
  projections
• Captures variability among GCMs
• www.engr.scu.edu/emaurer/global_data/
• http//climatewizard.org/

21
Changes to Inflows 15 Setiembre Reservoir

• Inflows to the major reservoirs decline by 13-24
• Drops in reservoir inflow July-August, 21 to 41.
• 20-year return low flow (firm hydropower
  generation indicator) 33-53.

22
Most commonly requested items
23
Need for enhanced downscaling

• Some impacts due to changes at short time scales
• Heat waves
• Flood events
• BCSD limited

24
Constructed Analogues
Given daily GCM anomaly
Analogue is linear combination of best 30 observed
Apply analogue to fine-resolution climatology
25
Can CA improve daily downscaled projections?

• Downscaled NCEP-NCAR Reanalysis for 1950-1999
• Use 1950-1976 as observed
• 1977-1999 as projected
• Monthly skill in reproducing Reanalysis P and T
  is high for both methods

26
Daily Temperature Extremes

• CA able to recover Reanalysis skill

Winter Cool Extremes (10 tile daily T)
Summer Warm Extremes (90 tile daily T)
r2
27
Daily Skill Dry Extremes
Dry Extremes (20 tile daily P)

• 20th percentile winter P
• r2 values shown
• 90 confidence line
• Low skill for both methods
• Daily large-scale data cannot counter lack of
  skill, poor relationship between scales
• No statistical difference for CA, BCSD
• Similar results for wet extremes
• Difficulty downscaling dry extremes

28
Peak Flow Differences

• Most sites comparable for both methods and Obs.
• Tuolumne R and Colorado R worse with CA than BCSD
• Room for improvement?

29
Differences between BCSD and CA

• CA uses daily GCM data BCSD uses monthly
  w/random resampling to produce daily values
• BCSD explicitly corrects for systematic GCM
  biases based on historic GCM performance
• CA corrects mean bias (using anomalies) but not
• spatial GCM biases
• variability biases

30
Looking in detail at one GCM cell

• At high and low extremes, reanalysis exhibits
  bias
• Accounting for bias in mean alone is insufficient
• Improvement Bias correct daily GCM data prior to
  CA BCCA
• Since BCCA is bias corrected, no need to anomalize

31
Schematic of Procedures
32
Effect of BCCA

• Compared to CA, BCCA improves
• simulation of annual flow volumes
• Simulation of flood peaks
• Problems remain for low flows, timing of snowmelt

Highlighted indicates downscaled different from
observed
33
Final Comments

• Statistical downscaling has skill, especially at
  monthly level
• Monthly downscaled data has substantial value to
  climate change impacts community
• Some daily skill from large (GCM) scale can be
  translated to regional/local scale
• Daily data (extremes) of interest for future
  studies
• For many measures, differences between
  downscaling methods are small

34
Final Question Have we captured most important
uncertainties?

• Perturbed physics experiments and theoretical
  feedback analyses extend tail to right
• Uncertainty in emissions is on same order if
  planning horizon includes end of 21st century or
  beyond

Roe and Baker, 2007
35
Thanks!