Climate Science for Decision Support .

Download Report

Transcript Climate Science for Decision Support .

Climate Science for Decision Support
Dave Stainforth
SoGAER, Exeter University;
Centre for the Analysis of Timeseries, London School of Economics.
Tyndall Centre for Climate Research
Oxford University Centre for the Environment
SIAM Minisymposium
From Global Predictions to Local
Action
January 2008
Communication / Knowledge Transfer Is Tricky
Increasing GHGs / Increasing Temperatures and Sea Level
Source: IPCC Fourth Assessment Report
Climate Forecasts: A problem of extrapolation
We have no hope of confirming a climate forecasting system the way we
might other systems; including weather forecasting.
- The lifetime of models is short.
- Predictions are for states of system for which we have no observations.
- Observations are in-sample.
- Confirmation of weather forecasting systems using related models can
not inform us about the long timescale processes in question; or their
interaction with short timescale processes.
[Q: It is not sufficient but is it a necessary condition for informative
climate forecasts?]
- We will only have one verification point (no good for probabilistic
verification) and that will come too late to be of practical value.
Uncertainty analysis is therefore critical.
Climate: The Distribution of Weather
• Climate change is a change in this distribution.
• Most if not all decision support is sensitive to more than the mean of
that distribution.
• Climate prediction is the attempt to predict this changing distribution.
Uncertainty analysis is therefore critical.
What’s the Purpose of Climate Modelling?
1. Pursuit of knowledge / Process Understanding
2. Answering the questions:
Is anthropogenic climate change a significant global problem?
3. Guiding (policy) decisions by governments, industry, societies and
individuals.
Adaptation Decisions?
•
•
•
•
•
•
Here are some possibilities. We need more/better in order to direct climate science well.
How will climate change affect life expectancy? (By region, social class etc.?) [Insurance
industry]
How should we design the next Thames barrier or the New Orleans flood protection?
How much will sea level rise in SE England? How will the characteristics of storm surges
change? What are they now?
[London/UK Government]
What are the relative risks to my pipeline from permafrost destabilisation for various designs
and routes?
[Oil/gas industries]
Given that climate change is likely to have a significant impact on agriculture, how do we
optimise the investment in a developing country’s transport infrastructure for long term
economic value and to minimise the impact of disasters?
Which bridges/roads are/will be most susceptible to flood damage? How will climate change
affect migration?
[Aid Agencies, UN, DFID, USAID etc.]
What sort of electricity cables should be lain under the streets of London?
[EdF, Energy Companies, Energy Regulators, Government]
Interpretation of Models: An Economist’s View
•
•
“Using climate models that follow basic physical laws, scientists can now
assess the likely range of warming for a given level of greenhouse gases in
the atmosphere.”
“It is currently impossible to pinpoint the exact change in temperature that will
be associated with a level of greenhouse gases. Nevertheless, increasingly
sophisticated climate models are able to capture some of the chaotic nature of
the climate, allowing scientists to develop a greater understanding of the many
complex interactions within the system and estimate how changing
greenhouse gas levels will affect the climate. Climate models use the laws of
nature to simulate the radiative balance and flows of energy and
materials. These models are vastly different from those generally used in
economic analyses, which rely predominantly on curve fitting. Climate
models cover multiple dimensions, from temperature at different heights in the
atmosphere, to wind speeds and snow cover. Also, climate models are tested
for their ability to reproduce past climate variations across several dimensions,
and to simulate aspects of present climate that they have not been
specifically tuned to fit.”
Source: The STERN report.
Interpretation of Models: IPCC
“There is considerable confidence that climate models provide credible quantitative
estimates of future climate change, particularly at continental scales and
above. This confidence comes from the foundation of the models in accepted
physical principles and from their ability to reproduce observed features of current
climate and past climate changes. Confidence in model estimates is higher for
some climate variables (e.g., temperature) than for others (e.g.,
precipitation). Over several decades of development, models have consistently
provided a robust and unambiguous picture of significant climate warming in
response to increasing greenhouse gases.”
Source: IPCC Fourth Assessment Report
Interpretation of Models: UK Climate Impacts Programme
• UKCIP (the United Kingdom Climate Impacts Programme) is
contracted to provide to UK industry a new set of scenarios in 2008.
“The UKCIPnext climate change scenarios will be
presented … as probability distributions.”
They will be available for 25km grid boxes.
“Model outputs will include changes in
temperature, precipitation, snowfall, wind speed,
humidity, cloud cover, solar radiation, air pressure
and soil moisture content.”
Source : UKCIPnext Consultation
Uncertainty Analysis – Probabilistic but not Probabilities
•
•
•
Multi-model analyses – Model Intercomparison Projects.
Perturbed physics ensembles.
Grand ensembles – perturbed physics / initial conditions / forcing.
(climateprediction.net)
•
•
Computing power is a significant issue.
But how should we design these experiments?
Sources of Uncertainty In Climate Forecasts
•
•
•
•
•
Forcing Uncertainty.
Microscopic Initial Condition Uncertainty.
Macroscopic Initial Condition Uncertainty.
Model Inadequacy.
Model Uncertainty.
Sources of Uncertainty
and How to Include Them In a Climate Forecast
• Forcing uncertainty:
Changes due to factors external to the
climate system e.g. greenhouse gas
emissions (natural and anthropogenic),
solar radiation etc.
Solution: Scenarios for possible futures.
Source: IPCC Fourth Assessment Report
Source: IPCC, Third Assessment
Sources of Uncertainty
and How to Include Them In a Climate Forecast
• Microscopic Initial Condition Uncertainty
How is the prediction is affected by our
imprecise knowledge of the current state
of the system at small, rapidly mixing,
scales?
Response: Initial Condition Ensembles
Source: IPCC, Third Assessment
Source: Large (50 member) IC ensemble from climateprediction.net.
What is climate?
What is climate under climate change?
•
•
Under constant boundary conditions weather is chaotic and climate
may be taken as the distribution of states on some attractor of weather.
Climate is the distribution of weather variables.
Under changing boundary condition the behaviour is not chaotic but
pandemonium [Spiegel, 1987].
Climate has changed from the initial attractor but the distribution itself
is in a transient state which may eventually stabilize towards some
other attractor when boundary conditions are again constant.
Climate is still the distribution of possible weather but it can not be
evaluated in the real world.
It can be defined for a model but its description requires very large
initial condition ensembles; something we don’t currently have.
Sources of Uncertainty
and How to Include Them In a Climate Forecast
• Macroscopic Initial Condition Uncertainty
How is the prediction is affected by our
imprecise knowledge of the current state of
the system on relatively large, slowly mixing,
scales?
• Response: Better Observations / Directed
Observations
•
Ocean temperature and salinity structure.
Sutton and Hodson, Science, 2005
•
State of the quasi-biennial oscillation.
Sources of Uncertainty
and How to Include Them In a Climate Forecast
• Model Inadequacy
All models are unrealistic representations of many
relevant aspects of the real world system.
• Response: A context for all climate forecasts.
•
•
•
Processes known to be important are absent.
e.g. ice sheet dynamics, atmospheric and oceanic chemistry,
stratosphere circulation.
Parameterized processes are unlikely to capture small scale
feedbacks.
Inadequate simulation of some processes which should result
from the fundamental processes included.
e.g. hurricanes, diurnal cycle of tropical precipitation.
Sources of Uncertainty
and How to Include Them In a Climate Forecast
• Model uncertainty:
Climatic processes can be represented in models in different ways e.g.
different parameter values, different parameterization schemes, different
resolutions. What are the most useful parameter values and model versions
to study within the available model class? What is the range of possibilities?
Solution: Perturbed-Physics Ensembles
Stainforth et al.2006
Exploring Uncertainty: The Climateprediction.net Experiment
Initial Condition
Ensemble
•
Forcing Ensemble
•
Overall Grand
Ensemble
Standard model
set-up
Perturbed Physics
Ensemble
10000s
10s
Statistics
• > 300,000 participants.
• > 24M years simulated.
• > 110,000 completed simulations.
(Each 45years of model time)
10s
•
•
To quantify uncertainty we need 100s
of thousands of simulations.
Impossible with super computers but
possible with distributed computing.
At www.climateprediction.net people
can download the model to their PC.
Using a GCM means we can get
regional detail as well as global
averages.
ClimatePrediction.net : What it looks like.
Teams.
P2P?
climateprediction.net Screensavers
First Results: Grand Ensemble Frequency Distribution of Climate Sensitivity
From Stainforth et al. 2005
Since then many studies
have shown the
possibility of high
sensitivities (>6 °C)
(e.g. Andronova et al.
(2001), Forest et al.
(2002), Knutti et al.
(2002), Murphy et al.
(2004))
Distributions of Regional, Seasonal Precipitation
Mediterranean Basin
Northern Europe
Winter
Winter
Summer
Summer
Annual
Annual
From Stainforth et al. 2006
Some Issues in the Interpretation of Perturbed-Physics and Multi-Model
Ensembles
•
Climate models are not independent.
– Climate models share methods, parameterizations, code. They are all
constrained by the same limits on resolution and computational structures.
– Exploration of parameter space is at the very least dependent on the
definition of parameters.
•
We have no empirically adequate models.
– Objective weighting by observations would rule out all models.
– Given that all models are unrealistic and we are trying to extrapolate into
the future, how do we know which are the most useful for predicting future
changes?
•
We have no hope of verification.
– Present day climate models will be long size abandoned by the time we
have the time we have observations of the climate of 2030/2050/2100.
– Even then we will have only one observational point. Not very useful in
defining the distribution which is climate.
•
Making over-confident claims today risks undermining climate science
just at the point when it is on the verge of providing valuable information.
Nevertheless we can be confident about some aspects of the future.
•
•
•
•
•
•
Global mean annual mean temperature will continue to rise.
Annual mean temperatures will probably rise almost everywhere.
Sea levels will continue to rise.
The climate in 40 years time will be very different from the climate now
or 40 years ago.
Society will therefore be less well adapted to the climate of the future
than it is to the climate of today.
Therefore we do need to mitigate.
And we do need to adapt.
We have information which may be useful in adaptation decisions.
Identifying it is a significant challenge.
Combined Distributions of Regional, Seasonal Temperature and Precipitation
Precipitation
6, 7. Downscale and
evaluate impact.
Temperature
Frequency of outcomes.
Sample The Range of GCM responses to Inform Decisions
8. Distribution of climate
sensitive factor.
5. Sample uniformly
4. Extract range of
GCM variables.
Climate sensitive decision factor e.g.
agricultural production (tonnes/yr)
How Do We Go Further?
•
•
•
How do we explore the boundaries of the types and extremes of
behaviour consistent with current models?
How do we develop models and experiments to provide better info in
the future?
How do we design experiments to:
– Guide model development.
– Inform process understanding.
– Guide climate change influenced decisions.
•
•
•
•
How do we balance exploration of model uncertainty with exploration of
initial value uncertainty?
P2
How should we explore parameter space?
High
What is the role of emulators?
Stnd
Does the distribution of model behaviour
in parameter space have any relevance to
Low
probabilities of real world behaviour?
Low Stnd
High
P1
How Do We Go Further(2)?
•
•
•
•
How (when?) should we link impact
/ economic/ user system models
with climate model ensembles?
How can we propagate uncertainty
between different components of the
climate system?
How often do we want new models?
How much should be in them?
When is better to concentrate on
specific aspects to understand them
better before constructing big
models. Maybe waiting for cpu to
catch up with what believe is
necessary.
Can we constrain / weight models?
The End