Downscaling and Regional Climate Change Statistical and Dynamical Downscaling

1
Downscaling and Regional Climate
ChangeStatistical and Dynamical Downscaling

• Eric Salathé
• JISAO Climate Impacts GroupUniversity of
  Washington

Ruby Leung Qian Fu PNNLYongxin Zhang
CIG Valérie Dulière CIG Cliff Mass, Rick Steed,
Mike Warner Atmos Sci
2
NOAAs Regional Integrated Sciences and
Assessments (RISA)

• NOAA supports university-based teams across the
  U.S. to analyze how climate impacts key sectors
  within a region and how climate information could
  help with resource management and planning within
  that region.
• RISAs create strong university partnerships with
  federal, state, and local stakeholders within a
  region.
• Example topics covered include Agriculture,
  Wildland Fire, Water Resources, Drought Planning,
  Fisheries, Public Health.

3
Why do we want to simulate the regional climate?

• Climate Impacts Applications
• Streamflow and flood statistics
• Water supply
• Ecosystems
• Human health
• Air Quality
• Process studies
• Extreme events
• Attribution
• Land-atmosphere interactions
• Orographic precipitation
• Regional Reanalysis

4
http//cses.washington.edu/cig/
5
Climate Change Impacts Assessment
Air Quality Model
6

• The Columbia Basin Climate Change Scenarios
  Project
• Lead Alan Hamlet, JISAO/CSES Climate Impacts
  Group Dept. of Civil and Environmental
  Engineering
• University of Washington

Study Partnerships Funding Partners WA
Department of Ecology (via HB 2860) Bonneville
Power Administration Northwest Power and
Conservation Council Oregon Water Resources
Department BC Ministry of the Environment Collabo
rative Partners Montana Department of Natural
Resources Idaho Department of Water
Resources USBR, Boise Regional Office USACE,
Seattle and Portland Districts
7
Study Objectives Provide comprehensive
hydroclimatological data to support water
planning at a range of spatial and temporal
scales in the Columbia River basin and
PNW. Increase spatial resolution of hydrologic
models to capture smaller basins relevant to
planning. Improve range of products and services
available, and construct tools and data
processing methods to make future updates easier
(and less expensive) to produce.
8

• Overview of Downscaling Approaches
• Delta Method
• Realistic daily time series and spatial
  variability.
• 91 years of variability associated with each time
  frame and emissions scenario.
• Only incorporates changes in mean T and P.
• Transient Method -- Bias Corrected and
  Statistically Downscaled GCM Data (BCSD)
• Incorporates more information from the GCMs, but
  as a result may also inherit undesirable aspects
  of GCMs as well.
• Facilitates trend analysis, examination of
  potentially altered variability.
• Hybrid Methods
• Takes time series and spatial behavior from the
  observed record, but incorporates changes in full
  probablilty distribution from the GCMs.
• 91 years of variability associated with each time
  frame and emissions scenario.
• Dynamic Downscaling
• Uses regional climate models. (Not proposed for
  this study.)

9

• Typical Applications of Each Downscaling
  Approach
• Delta Method
• Sensitivity studies
• Summary of all GCM projections in one run
  (limited runs to identify the central tendency
• Transient Method (BCSD)
• Trend Analysis of Hydrologic Variables
• Ensemble uncertainty analysis for 30-year windows
  at any time in the 21st century (flexible time
  period of analysis)
• Hybrid Methods
• Ensemble analysis of water systems over 90 years
  of variability.
• Flood and low flow analysis
• Any application that needs very realistic time
  series behavior, spatial extent of storms, etc.

10
What is Downscaling?

• Something you do to a 20th-Century climate model
  simulation to reproduce the observed climate.
• Will also give the projected regional climate
  change when applied to a future climate model
  simulation.

11
Empirical Downscaling

• Some set of parameters from a coarse-scale data
  set are used as Precictors
• Some parameter from a fine-scale observed data
  set are used as Predictand (may be station data
  or gridded data)
• An empircal relationship is found between
  Preditors and Predictand during the Tuning Period
• This relationship is used to map projected values
  of the Predictors onto projected values of the
  Predictand

12
The Predictand OSU PRISM
13
The Predictors From Global Model
BCSD
Widmann, Bretherton, Salathé, 2003
14
Regional Climate Models (aka dynamic downscaling)

• WRF (NOAH LSM)
• ECHAM5 forcing
• CCSM3 forcing
• HadRM (PRECIS) HadCM3 forcing

15
Land-Atmosphere Interactions
Wintertime Change from 1990s to 2050s
Snow Cover Change
Temperature Change
Change in fraction of days with snow cover
Change in Winter Temperature (degrees C)
Salathé et al 2008
16
MM5 Compared to Climate model
2020s
2050s
2090s
17
Winter Trends at Various Stations
18
Winter Trends at Various Stations
19
Fine-scale InformationBase Climate vsClimate
Change Signal
20
Statistical Downscaling CCSM3
Temperature
Precipitation
21
Statistical Downscaling CCSM3
22
WRF CCSM3
Temperature
Precipitation
23
WRF CCSM3
CCSM3-WRF simulation 2030-2060 minus 1970-2000
Regional average precipitation changes are
comparable to global forcing Big contrasts
around topography in regional model
Regional average temperature change generally
follow global model Amplified warming over high
terrain
24
Value Added by Regional Model
Percent Change Fall Precipitation along 48N
Cascades
Olympics
25
Things we know vsThings we need to learn
26
Extreme Precipitation
27
Extreme Precipitation
28
Ensemble Uncertainty
29
Trends in Extreme Temperature (1970-2000)
Frost Days
Summer Nights
Cool Days
Warm Days
Number of HeatWaves
30
Trends in Extreme Precipitation (1970-2000)
Precip Intensity
Total Precip
gt 10mm
95th Percentile
Annual Maximumone-day total
31
Trends in Extreme Precipitation (1970-2000)
Precip Intensity
32
Variability of Extreme Precipitation (ENSO)
Precip Intensity
Total Precip
gt 10mm
95th Percentile
Annual Maximumone-day total
33
climateprediction.net
Oxford University
Regional Climate Experiment Simulate the period
1900 to 2100 with many versions of a climate
modeladapted to run on personal computers under
the BOINC distributed computing framework.

• Based on the PRECIS regional modelling project.
• HadRM3 regional model, 50km resolution for
  Southern Africa, 25km resolution for Pacific
  Northwest.
• Regional model driven daily by winds, moisture
  etc. on boundaries from global model, running
  alternately.
• Simulate selected decades 1960s, 1980s, 2000s,
  2030s 2080s.

With Richard Jones, Myles Allen, Phil Mote, Bruce
Hewitson
34
Summary

• Different tools for different objectives
• Global model projections present the largest
  uncertainty
• Ensemble methods are critical
• Regional models are essential research tools and
  best method for exploring some climate change
  effects
• but still need to be downscaled or bias corrected
  for many applications
• Statistical downscaling is an essential step in
  impacts assessments