Transcript Casino-21: Public Participation in Climate Simulation of

climateprediction.com
The Physics of Risk: Understanding and Predicting Global Climate Change
Emily Shuckburgh, DAMTP
with thanks to Myles Allen and David Stainforth, Department of Physics, Oxford
www.climateprediction.com
The World’s climate in danger
Executive summary: Monday January 22, 2001

“There is new and stronger evidence that most of the
warming observed over the last 50 years is attributable
to human activities”

“Human influences will continue to change atmospheric
composition throughout the 21st century”

“The globally averaged surface temperature is projected
to increase by 1.4 to 5.8° C by 2100”

The projected warming is very likely to be without
precedent during at least the last 10, 000 years”
www.climateprediction.com
“A collective picture of a
warming world”

Over the 20th century:







global-average surface temperature has increased (0.6° C)
temperature in lowest 8km increased in past 4 decades
global-average sea level has risen (0.1- 0.2 m)
snow & ice extent decreased - widespread retreat of glaciers
precipitation has increased (up to 1% per decade)
In Northern hemisphere increase in temperature in C20th
likely to have been largest of any century, 1990s the warmest
decade & 1998 the warmest year during past 1000 years.
Since 1750 there has been an increase of:


CO2 (31%), CH4 (151%) and N2O (17%)
present CO2 likely not exceeded in past 20, 000 years
www.climateprediction.com
The Arctic & Antarctica
www.climateprediction.com
Indonesia & Australia
www.climateprediction.com
Bangladesh & Mozambique
www.climateprediction.com
The science of climate change





How can we model weather and
climate?
How can we predict climate when we
can’t predict next week’s weather?
What are the main uncertainties in
climate prediction?
Quantifying risk: the science of
probabilistic climate forecasting.
Harnessing idle CPU on YOUR
computer for global climate prediction.
www.climateprediction.com
Weather and Climate

“Weather” - tropospheric events associated with atmospheric
flows of few 100 m & few days or less

Weather phenomena - chaotic, but atmospheric data
averaged over a month - more regular

But exists interannual variability

“Climate” - state of the atmosphere averaged over several
years +: the expected weather for a particular time of year.

It is determined by the boundary conditions of the
atmosphere-ocean system:

solar irradiance (power output of the sun)

atmospheric composition (greenhouse gases...)

positions of continents, ice-sheets etc.
www.climateprediction.com
A simple climate model
Sun (energy input Fs)
Earth (black body)

Fs = 1375 W/m2, tube area pa2

~30% reflected away into space: albedo a = 0.3

~70% emitted as thermal infrared F0, from area 4pa2

“effective temperature”, Te given by:
sTe4=F0=¼Fs(1-a)  240 W/m2

gives Te = 255K, lower than observed Ts = ~285K

why? - greenhouse effect
www.climateprediction.com
Solar and thermal radiation
F0
Atmosphere:
More solar radiation
transmitted ts than
thermal radiation tt
Atmosphere (Ta)
Fg
Earth (Tg)
Fg=F0(1+ts)/(1+tt) 1.6 F0 =sTg4
Tg  286 K
For doubled CO2, net radiation to space is reduced
from ~240W/m2 by ~4W/m2
Climate system adjusts to restore balance.
www.climateprediction.com
Change in outgoing fluxes on
1K tropospheric warming
Dir
ect
Wa
ter
v.
Alb
edo
Clo
ud s
T ot
al
3.5
3
2.5
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
Direct emission: 4sTe3
more energy to space
 Warm moist air
increases IR opacity:
less energy emitted
 Snow & ice melt: less
energy reflected
 Cloud amount and
properties change
 Net feedback factor:
l = ~1.5W/m2/K

www.climateprediction.com
History of numerical modelling







Lewis Fry Richardson (Kings, Nat. Sci., 1900)
Whilst working as an ambulanceman in WW1 he
produced the first numerical weather forecast,
using a slide-rule.
During intervals between transporting wounded
soldiers back from the front he made a 6-hour
forecast of pressure and wind, starting from
analysis of the conditions at 7am on 20 May 1910.
It took at least 6 months and was very inaccurate.
Proposed a “forecast factory” with some 26, 000
accountants
25 years later Jule Charney formulated equations
to be solved on a computer
First successful numerical prediction of weather in
April 1950 using ENIAC computer
www.climateprediction.com
What is a climate model?

Dynamical equations + parameterisations
www.climateprediction.com
Dynamical equations
Newton’s 2nd law in horizontal (forces:
pressure gradient and Coriolis)
 Hydrostatic equation (gravity and
pressure gradient)
 Thermodynamic equation (temperature
can change by “advection” or
evapouration/condensation)
 Continuity equation (conservation of
mass)
 Equation of State (perfect gas)
 Water Vapour equation (amount of water
vapour)

www.climateprediction.com
The equations for a sphere
www.climateprediction.com
NWP parameterisations





Radiation
Surface and Sub-surface
processes
Large-scale cloud and
precipitation
Convection and convective
precipitation
Gravity wave drag
www.climateprediction.com
Ocean-Atmosphere GCMs
For climate
modelling,
physical
processes not
important for
weather
modelling must
be included, in
particular a
representation of
oceanic heat
transfer
www.climateprediction.com
Evidence of “climate control”?
Response to anthropogenic, solar and volcanic forcing
www.climateprediction.com
And how fast?
Surface oceans warm faster than depths.
 Most initial warming occurs in top ~100m.
 More sensitive climates (higher DT2xCO )
respond slower.
 Ocean continues to adjust for centuries after
atmospheric composition stabilises.

2
www.climateprediction.com
Climate model predictions: I
Change In Near Surface Temperatures by the 2040s
www.climateprediction.com
Climate model predictions: II
Predicted change at model resolution
www.climateprediction.com
How uncertain are these
model predictions?





Models depend on “parameterisations” of
processes too small to resolve.
Parameterisations represent the feedbacks
between smaller and larger scales.
Many prescribed “parameters” (e.g. “ice fall
speed in clouds”) are poorly constrained.
What is the impact of different parameter
choices on model predictions?
Harder question: impact of model structure.
www.climateprediction.com
Simulating global mean DTs a
simple climate model:
www.climateprediction.com
www.climateprediction.com
Varying DT2xCO in a simple
climate model:
2
www.climateprediction.com
www.climateprediction.com
Varying ocean heat uptake in
a simple climate model
www.climateprediction.com
www.climateprediction.com
Varying both DT2xCO and
ocean heat uptake
2
www.climateprediction.com
www.climateprediction.com
www.climateprediction.com
Range of 50-year forecasts
consistent with recent change
www.climateprediction.com
But climate is more than DTs
Simple re-scaling of model predictions only
works for very large-scale variables.
 Better approach: vary parameters within
models and repeat the forecast.
 But how much to vary parameters?
 The Monte Carlo solution:




Vary parameters over very wide ranges
Simulate 1950-2050 changes with many models
Down-weight predictions from runs that fail to fit
observed changes over 1950-2000
www.climateprediction.com
How many simulations?





The problem of non-linearity: you can’t just
add up responses to different perturbations.
All combinations and permutations need to
be tried, at least in principle.
5 settings each of 9 parameters gives 59
permutations, or 2M simulations.
Current typical ensemble sizes with a
comprehensive climate model: 4.
An impossible task?
www.climateprediction.com
Climateprediction.com
Most computing power is now on desks or
in bedrooms, not supercomputing centres.
 100-year simulation with HadCM3L would
take 8 months on an up-to-date PC.
 Over 2M people have participated in
SETI@home…
 So we plan to:




Distribute ~2M versions of HadCM3L set up for…
Pre-packaged (unique) simulation of 1950-2050
Estimate uncertainty from collated results
www.climateprediction.com
Participants will be able to:
Run up to 110 years of the UM.
 View model output.
 Compare their results on the web.

Run and view results from simplified
models.
 Run impact models?

www.climateprediction.com
Viewing model diagnostics
Surface Temperatures
www.climateprediction.com
Impacts on Water
IS92a, GISS, 2050
Large cost
(21)
(6)
(20)
(11)
(9)
No cost
(90)
(16)
(3)
(4)
(10)
Large benefit (7)
www.climateprediction.com
The participants?
To date we have:
> 17,000 people
> 45,000 PCs
 Aiming for ~ 2 million PCs.
 Who are they?
Individuals
Small businesses
Schools
Large businesses ?

www.climateprediction.com
How many is “enough”?
There are many more than 9
underdetermined parameters in the model.
 We cannot screen all possible combinations.
 We can be more efficient by:




Screening parameter combinations first with a
simplified model (same atmosphere, slab ocean).
Using intelligent sampling techniques (e.g.
“genetic” algorithms).
We apply a range of future emissions
scenarios to the most realistic models.
www.climateprediction.com
Current emissions scenarios
www.climateprediction.com
Implications for future forcing
www.climateprediction.com
What of the real world?
Should we expect climate change in the real
world to lie inside the range of predictions?
 It depends on the quantity of interest:





Has the ensemble converged, or does perturbing
more parameters change the estimated range?
Is the ensemble consistent with observations in
this quantity?
Do we expect this variable to be well-simulated?
Basic problem: a probabilistic forecast
cannot be tested with a single event -- the
world can always surprise us.
www.climateprediction.com
Project Plan

UGAMP / researchers release.


Linux release.



Soon
Single parameter sensitivity tests with the slab model.
Year End
Multiple parameter perturbations with the slab model.
Initial condition ensemble.
Main windows release.

Next year
Casino-21: A physics ensemble experiment.
www.climateprediction.com
climateprediction.com
www.climateprediction.com