Regional modeling: key processes and uncertainties

1
Regional modeling key processes and
uncertainties Isaac Held GFDL/NOAA

• Improving models
• Assessing imperfect models

2

• Models are improving!
• The mean model is far better than any single
  model
• Regional projections in AR4 based on mean model
• probably the best we have ever had

Courtesy of T. Reichler (U. of Utah) based on a
set of metrics measuring quality of mean climate
3
2080-2099 (A1B) - 1980-1999 Europe
temperature
Precipitation
models with dP 0
4
2080-2099 (A1B) - 1980-1999 Africa
temperature
Precipitation
models with dP 0
5
(No Transcript)
6
2080-2099 (A1B) - 1980-1999 N. America
temperature
Precipitation
models with dP 0
7
(No Transcript)
8

• Downscaling studies played a relatively small
  role in AR4 projections
• as compared to CMIP-3 global model archive
• Why? Much more difficult to assess
• no organized archives of results
  (exception PRUDENCE)
• often biased by downscaling only 1
  global simulation
• High resolution atmospheric time slices
• Regional models
• Statistical downscaling

9
High resolution time slices There are
limitations even if one imposes SSTs from
coupled model, one does not capture the same
climate response as the coupled model (especially
for Asian Monsoon). Are there better simple
boundary conditions to use? Despite limitations,
I believe that we need an organized repository of
standardized high resolution time-slice
simulations Danger of using only one boundary
condition The Oouchi et al (20km) time-slice
generates increase in Atlantic storm activity,
but forced by an Atlantic SST response that is
an outlier among the CMIP3 models
MRI SST Response
CMIP3 Ensemble Mean SST Response
10
Regional modeling Once again organized
consistent archives are needed if these are to
impact the assessment process in a meaningful
way. PRUDENCE, (regional downscaling for Europe)
is best example, but only downscaled from two
models, one of which is an outlier with respect
to the magnitude of the poleward shift in the
circulation/stormtrack in the Atlantic. NARCCAP
(downscaling over North America) is patterned
after PRUDENCE Need to insure that the small
number of global models being downscaled does
not distort results
200 km
50 km
observations
Winter mean precipitation in Western U.S.
11
Statistical downscaling should be routine and
global Downscale the world not just the coast
of Norway Treat like MOS (Model Output
Statistics) in NWP refinement, not just
downscaling (can improve large scales in
projections as well) Standard procedure
station data global models projections for
each station Create accessible archive of
standardized statistical downscaled results for
globe
12
Evaluating models Weighting models to optimize
multi-model output Example when trying to
assessing weights to different model projections
in some region, do you consider primarily the
quality of a models regional climate, or the
quality of the global simulation? How do you
evaluate the answer to this question, or any
other proposal for weighting some models more
than others One perfect model approach
show that you can predict the future of a
model! For example show, by using the 20th
century simulations, that by some measure one
can distinguish between models that get wetter
and those that get drier in the Sahel then use
that measure to determine which models more
closely resemble observations The metric(s)
underlying discussion of seamless
prediction Short range, extended range,
seasonal, decadal forecasts which of these is a
good metric for determining the quality of a
global warming forecast?
13

• Improving regional projections
• 1) No magic bullet -- need to work on improving
  models on multiple fronts
• Tropics are the main concern
• (largest spread in model results, and potential
  for very serious impacts)
• For mean temperature large fraction of the
  scatter among models
• is due to spread in estimates of global
  (transient) climate sensitivity
• (other things -- precip, extremes - change
  because temperatures change)
• Push to much higher atmospheric resolutions will
  help especially in tropics
• and in analysis of extreme weather events

14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
T. Knutson, S. Garner, J. Sirutis, I. Held, R.
Tuleya BAMS, 2007)