Dynamical Downscaling of NASA/GISS ModelE Using WRF

1
Dynamical Downscaling of NASA/GISS ModelE Using
WRF

• Tanya L. Otte1, Jared H. Bowden1, Jerold A.
  Herwehe1, Christopher G. Nolte1, and Greg
  Faluvegi2
• 1Atmospheric Modeling and Analysis Division, U.S.
  EPA, Research Triangle Park, NC
• 2NASA Goddard Institute for Space Studies and
  Columbia University, New York, NY
• 8th Annual CMAS Conference, Chapel Hill, North
  Carolina
• 20 October 2009

2
Motivations

• EPA has a need to predict the regional impacts of
  climate change on water, air, and ecological
  systems.
• Primary focus is on extreme events (e.g., heat
  waves, droughts, flooding, stagnation events) and
  the frequency of such events, in addition to
  changes in local mean temperatures and
  precipitation.
• Create strong partnerships with external
  institutions that have established credible
  research programs in global climate modeling.
• Using dynamical downscaling, the regional
  climate simulations must remain true to the
  climate trends that are projected by the global
  climate model.
• Need to achieve a delicate balance between the
  amount of constraint in the regional climate
  simulations given to the ModelE and the freedom
  of WRF to simulate its own mesoscale features.

3
Dynamical Downscaling
Global climate model (GCM) creates gridded future
climate with world-wide coverage.
Coarse spatial (2 x 2.5) and temporal (3-6 h)
intervals.
Regional dynamical model uses GCM as initial and
surface and lateral boundary conditions to
generate gridded higher-resolution climate
predictions over focal area.

• More detail in local effects from
• appropriate physics
• topography land/water interfaces
• urban areas (population centers)
• precipitation patterns
• increased temporal output (1h)

4
Approach to Developing Downscaling Methodology

• Use reanalysis data as a verifiable surrogate for
  GCM
• Apply reanalysis at GCMs spatial and temporal
  resolutions
• Validate against observations and/or
  high-resolution analyses
• Similar approach to many other downscaling groups
• BUTalso simultaneously developing downscaled
  fields using GCM output to test downscaling
  methodology on a parallel track.
• Relatively unique approach within regional
  downscaling community
• Allows sanity check of conclusions drawn from
  reanalysis (i.e., can we make the leap of faith
  between reanalysis and GCM methods?)

5
Reanalysis vs. GCM for Downscaling Using WRF

• NCEP/NCAR Reanalysis Project (1988)
• Well-respected, often used
• Multiple observation sources and advanced data
  assimilation techniques create reanalyses
• 2.5 x 2.5, 6-h fields
• 28 s layers up to 3 hPa
• Use 17 pressure levels up to 10 hPa
• Validate WRF vs. 32-km NARR

• NASA/GISS ModelE (ca. 2002)
• 1 of 3 U.S. GCMs in IPCC AR5
• Coupled atmosphere-ocean model creates GCM
  forecast
• 2.0 x 2.5, 6-h fields
• 40 s-p layers up to 0.1 hPa
• Use native vertical layers in downscaling
• Validate WRF vs. ???

• WRFv3.1 Model Configuration for Downscaling
• 108-km domain, 34 layers, model top at 50 hPa
• RRTMg LW and SW Radiation
• WSM6/WDM6 Microphysics
• ACM2 PBL
• Pleim-Xiu Land-Surface Model
• Kain-Fritsch Cumulus Parameterization

Evaluate larger-scale constraint - Through
LBCs only - With analysis nudging - With
spectral nudging
6
Annual Mean 2-m Temperature WRF minus
Reanalysis (1988)
No Nudging
Analysis Nudging
Spectral Nudging
WRF generally warmer than reanalysis. Larger
differences (bias or error) without nudging than
with nudging. Spectral and analysis nudging very
similar.
7
Annual Mean 2-m Temperature WRF minus GCM (ca.
2002)
No Nudging
Analysis Nudging
Spectral Nudging
WRF generally warmer than GCM, and more
pronounced than reanalysis. Largest differences
in complex terrain. Spectral and analysis nudging
very similar, and cooler than no nudging. (Is
this systematically reducing bias or
systematically reducing temperature?)
8
January 500 hPa Geopotl Height WRF minus
Reanalysis (1988)
No Nudging
Analysis Nudging
Spectral Nudging
Nudging mitigates but does not remove anomalous
height errors over subtropical waters. Nudging
significantly reduces pattern error in upper
Rocky Mountain region.
9
July 500 hPa Geopotl Height WRF minus
Reanalysis (1988)
No Nudging
Analysis Nudging
Spectral Nudging
Larger errors without nudging than with either
spectral or analysis nudging. Nudging works to
correct pattern errors seen in no nudging
run. Nudging exacerbates a positive height
anomaly in the Plains (lee of Rockies).
10
A Few Thoughts

• For both reanalysis and GCM, 2-m temperature
  differences (i.e., biases) are lower when either
  spectral or analysis nudging is used than without
  nudging.
• Differences between WRF simulation and GCM are
  more pronounced (higher error) than WRF minus
  reanalysis.
• Is this related to higher-resolution detail in
  WRF than coarse GCM, and, if so, is this adding
  value?
• Or is this related to different behavior of GCM
  fields than reanalysis in WRF, or different
  years?
• How do we know? How do we relate statistical to
  physical differences?
• Seasonal differences in 2-m temperature are
  qualitatively similar to annual differences for
  both reanalysis and GCM (not shown).
• Across the domain, spectral and analysis nudging
  are qualitatively similar for 2-m temperature and
  500-hPa height under this configuration of WRF.
• No clear advantage for either technique yet.
• Using either nudging technique is better than no
  nudging at all.

11
Daily 2-m Temperature (1988)Analysis vs.
Spectral Nudging Reanalysis (by region)
Regional errors in daily average 2-m temperature
track very closely within each region for
analysis and spectral nudging. Changes in weather
patterns captured consistently with both methods
of nudging.
12
Daily Precipitation (1988)Analysis vs. Spectral
Nudging Reanalysis (by region)
Regional errors in precipitation reveal seasonal
and regional differences in nudging
approach. Often errors are more pronounced in
summer (convection?). Larger, widespread
differences seen with spectral nudging. At
times, analysis nudging has dry bias.
13
Daily 2-m Temperature (ca. 2002)Analysis vs.
Spectral Nudging GCM (by region)
Analysis and spectral nudging perform similarly
for each region, but more distinction in
amplitude than with reanalysis. Regional behavior
of bias and magnitude of bias is different with
GCM than with reanalysis.
14
Daily Precipitation (ca. 2002)Analysis vs.
Spectral Nudging GCM (by region)
Larger regional totals with spectral nudging. At
times, analysis nudging has dry bias. Magnitude
of difference in regional precipitation is much
larger with GCM than reanalysis. Significant
seasonal differences between nudging methods in
the regions
15
Annual 2-m Temperature and Precipitation from
GCM SW Region
Its not just about whether or not nudging is
used but how it is used. Subtle differences in
the nudging technique can make a BIG difference
in WRF. 2-m temperature fairly consistent with
both cases, but precipitation is very different.
Precipitation
Analysis Nudging (Case 1)
2-m Temperature
Precipitation
Analysis Nudging (Case 2)
2-m Temperature
16
Annual 2-m Temperature for Reanalysis and GCM
Plains Region
GISSE MAE 0.9C bias almost 0 seasonal
NNRP MAE 1.4C cold bias
Similar configurations of WRF can have different
qualitative impacts with reanalysis and GCM in
same region (e.g., reverse seasonal biases).
17
A Few Final Thoughts

• In our testing, RCM simulations are far more
  sensitive to nudging vs. no nudging than
  changing physics options (not shown).
• Both spectral nudging and analysis nudging are
  improvements over no nudging at all.
• Simulations that are only constrained via LBCs
  rapidly deviate from the larger-scale forcing
  fields. (Think forecast model.)
• Although spectral nudging is the hot topic for
  regional climate modeling, analysis nudging
  should not be ignored.
• There is little indication of decrease in skill
  with using nudging over time in an annual
  simulation. (Could be significant for AQM.)
• Method of nudging (including variables, strength,
  and where in atmosphere) can have important
  effects on simulation.
• Regional behavior with reanalysis fields (often
  used to establish downscaling methods) is not
  necessarily analogous to behavior with GCM
  fieldsat least looking at a single-year
  simulation.
• Near-surface fields (e.g., 2-m temperature and
  precipitation) are easiest to verify, but they do
  not tell the whole story. Need to look aloft,
  too!