Improvements in Numerical Modelling During the ACSYS Decade

1
Improvements in Numerical Modelling During the
ACSYS Decade

• Gregory M. Flato
• Canadian Centre for Climate Modelling and
  Analysis
• Meteorological Service of Canada

2
Outline

• This overview is necessarily selective and
  focuses primarily on large-scale models,
  particularly global climate models.
• I will say something about the following topics
• Global climate models
• Sea Ice
• Snow
• Hydrology
• Atmospheric Circulation
• Ocean Modelling
• As you will see, Model Intercomparison Projects
  were a growth industry during this decade.

3
Introduction

• At the beginning of the 1990s, most climate
  modelling was done with atmosphere-slab ocean
  models
• These provided equilibrium estimates of climate
  change, typically with doubled CO2
• They highlighted projected enhancement of
  warming at high latitudes, particularly in
  winter.

4
CCCma
2xCO2 winter (DJF) temperature change from three
early climate models (IPCC, 1990). High-latitude
amplification is attributed to positive feedbacks
involving sea-ice albedo over ocean and snow
albedo over land.
GFDL
UKMO
5
Illustration of the effect of sea-ice feedbacks
on CO2-induced warming response
Interactive Sea ice
Fixed sea ice
6

• Transient climate change simulations require
  coupling of 3-D ocean models.
• This began in earnest in the early 1990s.
• In parallel, there was growing appreciation that
  the meridional overturning circulation in the
  North Atlantic might be sensitive to freshwater
  perturbations in the Nordic Seas.
• This raised the profile of modelling sea-ice
  motion so as to represent ice flux out of the
  Arctic through Fram Strait.

overturning state
collapsed state
Weaver and Sarachik, 1991
7

• By the mid-1990s there were 15 coupled models
  available and the Coupled Model Intercomparison
  Project (CMIP) was launched.
• However, only 4 of these had a physically-based
  representation of sea-ice dynamics.
• Despite the fact that rather sophisticated
  sea-ice models had been available since the late
  1970s (e.g. Hibler, 1979).

8
Global Climate Models of the mid 1990s

? Motionless ice with a prognostic equation for
ice growth and melt. 2 Prognostic equations for
growth/melt and ice motion, including
representation of internal ice stress. 3
Prognostic equation for ice growth/melt, ice
motion diagnosed as a function of ocean surface
current.
9
Modelled ice extent in the 12 model CMIP ensemble
10 of models have less ice than this. Median
ice edge. 10 of models have more ice than this.
Interestingly, median model ice edge agrees well
with observations.
G. Flato for IPCC 2001
10
NH Ice Extent and its Change CMIP2 model
ensemble (CO2 increased at 1 per year for 80
years the time of doubling
Initial Ice Extent
Ice Extent Change
No obvious connection between error and ice model
characteristics
11

• As mentioned earlier, climate warming is
  enhanced over sea ice.
• However, this is also the location of largest
  disagreement.
• Representation of sea-ice processes and
  feedbacks is implicated.

NH ensemble mean temperature change (C)
NH intermodel standard deviation (C)
12
SH ensemble mean temperature change (C)
SH intermodel standard deviation (C)
There is a need to evaluate/improve
representation of sea-ice processes
13
ACSYS NEG Sea-Ice Model Intercomparison Project
(SIMIP)

• Initial intercomparison focused on dynamics
• Objective was to quantitatively evaluate
  performance of different dynamic schemes used in
  climate models.
• viscous-plastic model performed best it is
  appearing in many of the new coupled models.
• A follow-on project, SIMIP2, focussed on
  thermodynamics, is underway now.

Kreysher et al., JGR, 2000
14
Stand-alone sea-ice models, forced with
historical re-analysis of atmospheric quantities,
have also provided insight into aspects of
sea-ice behaviour that we cannot observe.
15
Hilmer and Jung, 2000
16
Snow

• Has a profound effect on surface energy balance.
• Rapidly-evolving and heterogeneous material.
• Interacts with vegetation in complex ways.
• Sophisticated snowpack models have been developed
  for applications such as avalanche forecasting.
• Too computationally demanding for use in GCMs
• GCM representations of snow typically use 1 or a
  few layers, with simplified physics.
• Various approximations led to a considerable
  diversity in early climate models.
• As well see, more recent models show
  considerable improvement.

17
Snow cover error in AMIP1 models (early 1990s)
Frei and Robinson, 1998
18
Snow cover error in AMIP2 models (late 1990s)
Frei et al., 2003
19

• Snow Model Intercomparison Project (SnowMIP)
• www.cnrm.meteo.fr/snowmip
• considered two alpine sites in Europe and two
  sites in N. America.
• point simulations at sites without tall
  vegetation.
• simple snow schemes used in GCM and hydrological
  models as well as sophisticated schemes.

Sleepers River, VT, USA
Weissfluhjoch, Switzerland
Large differences at Vermont site are attributed
to differences in the representation of
sublimation.
20
Connections to Hydrology

• Project for Intercomparison of Land-Surface
  Parameterization Schemes (PILPS) involves 21
  models.
• PILPS 2d 18-year simulation of seasonally
  snow-covered grassland (Valdai, Russia).
• PILPS 2e basin in northern Scandinavia.
• Sublimation of snow was a large source of
  intermodel discrepancy.

PILPS 2e
Observed
(Added by GF)
Provided by R. Essery, U. Aberystwyth
21
Atmospheric Circulation the Arctic Oscillation
http//www.cpc.ncep.noaa.gov/products/precip/CWlin
k/all_index.html
22
Climate change scenario
CCCma model Fyfe et al., 1999
Observations to 2002
GISS model Shindell et al., 1999
Stratosphere included
No Stratosphere
23
AMIP (the Atmospheric Model Intercomparison
Project) allowed evaluation of various aspects of
atmospheric circulation. Mean sea-level pressure
exhibits certain biases, but there is a lot of
variation from model to model.
Bitz et al., J.Clim., 2002
24
Thickness field from a sea-ice model driven by
different forcing
Observation-based Forcing
Climate Model Forcing
Bitz et al., J.Clim., 2002
25
Circulation biases, along with other errors,
impact other quantities, such as precipitation
Range of observational estimates
Walsh et al., J. Clim., 1998
It is hard to isolate or identify shortcomings in
process representation in a global model,
particularly for some region
26

• Arctic Regional Climate Model Intercomparison
  Project (ARC-MIP)
• Joint project of ACSYS/CliC NEG and GEWEX Working
  Group on Polar Clouds
• RCM experiments using common domain and boundary
  conditions.
• 5 RCM groups participating.
• Comparing with observations from SHEBA year
  Oct/97 Oct/98

http//cires.colorado.edu/lynch/arcmip/background.
html
27
Ocean Modelling
Models of Arctic Ocean Circulation have improved
significantly during the ACSYS decade, in concert
with availability of improved atmospheric
forcing. Model studies have contributed
substantially to understanding variability in the
Arctic Ocean. Karcher, Gerdes, Kaucher and
Koeberle, JGR, 2003
28

• Arctic Ocean Model Intercomparison Project
  (AOMIP)
• Initial evaluation of existing Arctic ocean model
  output.
• Coordinated model experiments underway now.
• Example shows transport streamfunction from
  various Arctic Ocean models, forced in various
  ways.
• Steiner et al., Ocean Modelling, 2003

http//fish.cims.nyu.edu/project_aomip/overview.ht
ml
29
A recent trend is to make use of alternate model
grid configurations in global models to better
resolve ocean (and ice) processes in the Arctic.
These examples are from the POP ocean code, used
in the NCAR community climate model.
http//climate.lanl.gov/Models/POP/index.htm
30
Summary

• Feedbacks involving the cryosphere lead to
  amplification of projected climate warming in the
  Arctic.
• These feedbacks also amplify model errors
• Although climate models are improving, errors in
  representing Arctic climate remain large and
  must be improved.
• The last decade has seen an increased focus on
  modelling Arctic climate.
• Various intercomparison projects yield
  quantitative evaluation of model shortcomings.
• Representation of snow in climate models has
  improved demonstrably.
• More sophisticated sea-ice models are being
  employed, and alternative grid configurations are
  being used to improve resolution of Arctic ice
  and ocean processes.

31
The End
32
Sea Ice and its Response to CO2 Forcing in
Global Climate Models G.M. Flato Canadian
Centre for Climate Modelling and Analysis

• Results from an ensemble of climate models are
  compared
• mean state and response to increasing CO2
  concentration
• find a large range in model results that are not
  obviously connected to representation of sea-ice
  processes. Connection to initial ice stated is
  also not compelling.