Regional Performance of the IPCCAR4 Models in Simulating PresentDay Mean Climate

1
Regional Performance of the IPCC-AR4 Models in
Simulating Present-Day Mean Climate

• Junsu Kim and Thomas Reichler
• University of Utah, Salt Lake City, USA

2
Introduction

• Previous work
• How well do coupled models simulate todays
  climate? (Reichler and Kim 2008, BAMS, JGR)
• 3 model generations CMIP-1 to CMIP-3
• Focus Global performance skill

3
Error
4
Introduction

• Previous work
• How well do coupled models simulate todays
  climate? (Reichler and Kim 2008, BAMS, JGR)
• 3 model generations CMIP-1 to CMIP-3
• Focus Global scale
• Basic idea of this model intercomparison work
• Realistic simulation of current climate is a
  necessary condition for confidence in simulation
  of future
• This work
• Regional variations in model performance
• CMIP-3 models (IPCC-AR4)

5
How to Evaluate Model Performance?

• Problem of objectiveness
• measure of error (or goodness)
• choice of quantities/processes
• relative weights
• Method
• current (79-99) mean climate and seasonal cycle
• multivariate approach aggregate errors from many
  climate quantities into a single index
• rational
• complex interrelationship amongst individual
  components of climate
• it is not enough to focus on just one particular
  quantity of interest
• to have confidence in a model, it must simulate
  every aspect of climate well

• moments of climate
• timescale
• observational uncertainty
• spatial domain

6
Methodology

• Normalized error variance
• Regional error index
• Overall performance index

lt1 Better than average
How capable is a model in simulating regional
climate relative to the average performance on
the global scale?
Equal weighting

• We evaluate
• 24 CMIP-3 models (excluding BCC-CM1)
• average model
• multi-model mean
• NCEP/NCAR reanalysis

7
Regions
Land 22 regions Giorgi and Francisco
(2000) Ocean 10 basins
8
Climate Elements
Physics (12)
Oceans (9)
Land (1)
Dynamics (9)
9
Results
10
Average Model Performance
Tropics generally less well (50) simulated
than extratropics (-20 to -50) India and Tibet
most problematic (100)
11
Southern Asia (India)
Breakdown by Quantity
median
Error
individual models
climate elements

• most quantities show larger than average errors
• v850 and prw are most difficult

12
Average Model Performance
Tropics generally less well (50) simulated
than extratropics (-20 to -50) India and Tibet
most problematic (100)
13
Mediterranean
Breakdown by Quantity
Error

• most quantities well simulated
• Z500 most faithfully

14
Individual Models
15
Multi-Model Mean
NCEP/NCAR Reanalysis

• Problems over Antarctica, Tropics, Tibet
• Oceans better than land
• Does well over India (plenty of observations)

• Better than NNR for every region

16
Conclusion

• Performance index is useful to compare models and
  to track model changes
• Large inter-model differences
• Good models do well over all regions and all
  quantities
• Extratropics are generally better simulated than
  Tropics
• Multi-model mean outperforms even the best
  individual model and even the reanalysis
• Important to keep in mind (Retto Knutti)

Good performance in current climate increases
credibility of a model simulation but it is not a
guarantee for a reliable prediction of future
climate
17
Thank You
Reichler, T., and J. Kim (2008) Uncertainties in
the climate mean state of global observations,
reanalyses, and the GFDL climate model, J.
Geophys. Res., 113 Reichler, T., and J. Kim
(2008) How Well do Coupled Models Simulate
Today's Climate? Bull. Amer. Meteor. Soc, 89,
303-311.
18
CMIP-3
19
Southern Asia (India)
Breakdown by Models
India
Other regions
20
Case Study Precipitation
NNR
Multi-model mean
GFD21
Average model