Climate modelling

Download Report

Transcript Climate modelling

Climate models – prediction
and projection
Nils Gunnar Kvamstø
Geophysical Department
University of Bergen
Climate models
• Best tool for projections
• The best tool for attribution of
observed climate change
• Potential for realistic regional
and local projections
• The reality of global warming
is based on much more than
climate model results
Climate model
results used for
• Local climate change
• Mitigation studies, e.g.
emissions corresponding
to two degrees GW
• Effects of climate change
(e.g. extinction, food
supply, climate refugees,
water management,
economy)
Vilhelm Bjerknes 1862-1951 –
proposed weather prediction models in
1904
Painting: Rolv Groven
• Doctor deg. 1892
• Ass. Prof. in
mechanics
Stockholm 1893
• Prof. Stockholm
1895
• Kristiania 1907
• Leipzig 1912
• Bergen 1917
• Oslo 1926
Bjerknes’ vision on scientific
weather forecasting
Bjerknes’ vision
1. The state of the atmosphere must be known
for a specific time (from observations)
2. Then future states might be computed from
conservation laws for mass, energy and momentum
Model equations
Numerical methods:

  1
dV
  fk  V  p  PV
dt

d
 P
dt

d
    U
dt
p
  g
z
Finite differences
Spectral methods
d




 u v  w
dt t
x
y
z
One set of prognostic
variables in each grid box
Parametrisation of sub-grid scale processes
Resolved topography
Sub-grid topography
PV  k Z
Sub-grid processes for parameterisation
Sub-grid processes are prameterized, often as a function of
grid-point values. Horizontal derivatives are not involved =>
Easier to parallellize these computations.
Numerical Weather Prediction Models
1. State at a specific time S0 (wind, temperature,
pressure, humidity clouds) determined from
observations.
2. Future stated by solving (non-linear) conservation
laws:
dS
 H (S0 )
dt
t1
S1   H ( S 0 ) dt
t0
S1  H ( S 0 )  t
Predictability for weather forecasting
Lorenz attractor
y
(analogy to weather behaviour)
Ed Lorenz
(1918-2008)
x
Limitation in predictability of the
weather
Theoretical limit
score
Today’s limit
Limit for
useable
prediction
Tomorrow’s limit
predictability (days)
Predictability for climate
change different from that of
weather forecasting
• Weather forecasting: predictability
of first kind (the actual weather)
• Prediction of climate change:
predictability of second kind
(statistical properties of the
weather over several years)
• Actual weather models determines:
present limits for weather
prediction
• Actual climate models determine:
present ability to predict climate
change
Increase in complexity of climate models
FAR: First Assessment Report (IPCC 1990)
SAR: Second Assessment report (IPCC 1996)
TAR: Third Assessment Report (IPCC 2001)
Source: IPCC AR4 WG1
NorESM framework and model components
Atmospheric chemistry
CAM
CICE
CLM
River routing
MICOM
HAMOCC
Components in blue communicate
trough a coupling component.
Components in red are subroutines of
blue components.
Computer platforms
Shared memory
Distributed memory
memory
network
Combination
node
• gridur/embla (2002), 2 nodes, 384 + 512 = 896 cores, 1.0 Tflop
• njord (2006), 62 nodes x 16 cores = 992 cores, 7.5 Tflop
• stallo (2007), 704 nodes x 8 cores = 5632 cores, 60 Tflop
• hexagon (2008), 1388 nodes x 4 cores = 5552 cores, 50 Tflop
Specs Bergen Climate Model
Res atmosphere (1.9x 2.5 deg 20 layers) 96*172*20
Res ocean: (1deg, 20 layers) 180*360*90
90 procs Hexagon 6.7 yr/d
Output 6h, d, mon -> 5Tb per 100 years
Dipole grid (default CCSM)
Tripole grid
EXPERIMENT TYPES
Emission scenarios from IPCC, includes also air pollution
giving aerosols
ppm
Projections of global temperature change
Norges mål:2 grader
IPCC
IPCC 2007
Changes in precipitation (percent) by the end of
the century
Winter
Summer
Conclusions
• Climate models solve well known physical equations
from hour to hour, from day to day, from year to year
• Climate models have no tuning to fit observations
• Climate models simulate the observed global warming
during the latest decades
• Best tool for future projections and attribution of sources
for climate change
• Decadal prediction with models a large research area
• Climate drift a problem
• Next generation models will include the carbon cycle
• Feed-back from clouds and effects of aerosols a
notorious problem
Questions
• Why are there 'wiggles' in the output?
• What is tuning?
• What is robust in a climate projection and
how can I tell?
• Are the models complete? That is, do they
contain all the processes we know about?
• Do models have global warming built in?
• What is the difference between a physical
model and a statistical model?