Uncertainty and Learning in Sequential DecisionMaking: The Case of Climate Policy

1
Uncertainty in Integrated Assessments
Mort Webster
Department of Public Policy, University of North
Carolina at Chapel Hill
Watson Institute Frontiers of Environmental
Change Research 15-16 September 2005
2
Persons pretending to forecast the future shall
be considered disorderly under subdivision 3,
section 901 of the criminal code and liable to a
fine of 250 and/or six months in prison.
Section 889, New York State Code of Criminal
Procedure
3
Outline

• Motivation
• Propagating Uncertainty Through an Integrated
  Assessment Model
• Uncertainty in Decision Analytic Framing
• Reduction of Climate Uncertainty by Observing
  Climate Variables

4
Motivation How should we choose near-term
climate policy?

• Example Senate Energy Bill Debate, Summer 2005
• McCain-Liebermann (20/tonC)
• Bingaman Bill (7/tonC)
• Hagel Resolution (Voluntary 0/ton)
• What is the right level of effort?

5
Bottom Line Climate Policy as Risk Reduction

• Need to have a long-term strategy to guide
  short-term decisions
• Climate Policy How can we reduce/manage the risk
  of severe climate change impacts?
• To manage risk need to know the relevant
  uncertainties to best of our current knowledge.

6
Cascade of Uncertainties

• Uncertainties in socioeconomic and technological
  trends
• Uncertainties in climate system
• Uncertainties in regional impacts
• Uncertainties/Disagreements in (economic)
  valuation of impacts

7
Using Uncertainty Characterizations

• Range of Possible Outcomes
• Assessing the importance of climate change
• Decision-focus
• Use uncertainties to look for Robust Strategies
• NOTE Risk Reduction ? Uncertainty Reduction
• Learning Which uncertainties are likely to be
  reduced (and within what time frame)?
• How does this change the robust (optimal)
  near-term strategy?
• Value of Information
• Which reducible uncertainties will most change
  the policy decisions?

8
MIT Integrated Global System Model (IGSM)
Joint Program on the Science and Policy of Global
Change
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
9
Uncertainty Analysis of Climate Projections
Uncertain Parameters

• Economic Technological Uncertainties
• Labor Productivity Growth
• Autonomous Energy Efficiency Improvement Rate
• Emissions Factors for Industrial Pollutants
• Climate Uncertainties
• Climate Sensitivity
• Heat Uptake by Deep Ocean
• Aerosol Forcing Strength

10
Uncertainty Analysis of Climate Projections
Outcomes

• What is the uncertainty (PDF) in
• Global Mean Temperature Change
• Sea Level Rise
• As a result of
• No Climate Policy
• One Possible Path of GHG Reductions
• How does the risk of extreme outcomes change?

11
Sources for Parameter PDFs

• Economic Parameters
• Literature, e.g., std err on regression
  coefficients
• Expert Elicitation
• Climate Parameters
• Constrained by Climate Detection (Forest et al)
• Combined with Expert Judgments

12
Uncertainty Propagation through IGSM

• Using Latin Hypercube Sampling, draw samples from
  all parameter PDFs, imposing correlation (n250
  n1000)
• Propagate through EPPA to obtain emissions of all
  GHGs and other relevant substances
• Propagate through climate-chemistry model to
  obtain climate outcomes

13
Global CO2 Emissions
14
CO2 ConcentrationsMedian and 95 Bounds
15
Total Radiative ForcingMedian and 95 Bounds
16
Global Mean TemperatureChange Median and 95
Bounds
No Policy
Policy
17
Global Mean Temperature Change by 2100
1 in 20 chance of exceeding 3.2o
1 in 20 chance of exceeding 4.9o
18
Communicating the Odds of Temperature Change
19
Communicating the Role of Policy
Stringent Policy
No Policy
20
Decision-Analytic Framing of Climate Policy
Problem

• Simple Example 2-period decision
• Choose mitigation level for 2010-2030
• Choose mitigation level for 2030-2100
• Use 2 extreme assumptions to bound the range
• No Learning Both decisions made under
  uncertainty
• Complete Learning All Uncertainties resolved
  before second decision

21
Decision Analytic Framework
2010 Decision
2030 Decision
Future Uncertainties
22
Use Uncertainties in Decision Framework
23
Value of Information within Simple Framework
24
Using Observations to Reduce Climate Uncertainty

• How to reduce climate uncertainty?
• Experiments and Research
• Observations (e.g., global mean surface
  temperature)
• Using observations, how long until uncertainty
  reduced?
• How might this change todays decision?

25
Applying Bayes Rule

• Suppose we observe global mean surface
  temperature increase at some future time (e.g.,
  2050).
• Focus on Climate Sensitivity
• Use Bayes

26
Numerical Implementation

• Perform Monte Carlo on Prior PDFs to 2050
• Choose 0.05, 0.5, 0.95 from PDF of DT2050
• For each possible observation, update the prior
  PDF of Climate Sensitivity and obtain new
  posterior PDF
• Use posterior PDF of sensitivity for Monte Carlo
  simulations of temperature change from 2060 to
  2100.

27
Compute Temperature Change Conditional on Climate
Sensitivity
28
Prior and Posterior Distributions for Climate
Sensitivity (2030)
29
Prior and Posterior Distributions for Climate
Sensitivity (2050)
30
Posterior also depends on Observed Emissions and
on Climate Noise
31
Uncertainty Bounds (5-95) on Temperature Change
32
Use Updated Uncertainty in Decision Model
Near-term decision
Revise Policy
Observe Climate
Remaining Uncertainty
33
Optimal Mitigation with Partial Learning
34
Sensitivity of Near-Term Policy to Climate Damage
Uncertainty
35
Improvements to Partial Learning Calculations

• Right now Only climate sensitivity updated,
  based on GMT,
• Better Update the JOINT PDF of sensitivity, heat
  uptake, aerosol forcing, based on spatial pattern
  of temperature in atm. ocean,
• Appropriate treatment of climate noise,
• Update every decade
• Model costs of learning (observations, RD)

36
Questions to Explore

• Relative potential for uncertainty reduction in
  economic, climate, impacts
• Better estimates of value-of-information
• Guidance for overall level of effort for
  near-term policy
• What to do about negative learning?

37
Major Challenges for Climate Change Research

• Regional scale impacts
• Regional climate, ecosystems effects,
  hydrological, agricultural, etc.
• Ways to think about technological change
• Ability to model more realistic policies
• Ways to deal with uncertainty and design scenarios

38
Calvins View on Risky Decisions
39
Uncertainty in Policy Costs
40
Challenges to Uncertainty Analysis

• Empirical Challenges
• Past behavior does not fully determine the future
• Sparse data
• Expert judgments required cognitive biases
• Methodological Challenges
• Combining experts
• Model uncertainty

41
Challenges to Uncertainty Analysis (II)

• Institutional Challenges
• How to structure formal assessment processes?
• Focus on consensus
• Expert judgments in a political context
• Appropriate venue for uncertainty analysis?
• Philosophical Challenges
• Frequentist vs. Bayesian
• Differing views on future social development

42
Possible Next Steps to Improve UA

• Constructing PDFs for socio-econ. parameters
• Using historical data to inform
• Multiple Experts
• assessments across wider range, intercomparisons?
• More focus on impacts (beyond DT)?
• More links between standard scenarios and
  probabilistic information?
• Other ideas?

43
Impacts of Period 1 Decisions
44
Using Data to Constrain Climate Parameter
Distributions
From Forest et al, Science 295, 113-117
45
Uncertainties in Emissions/Costs
Webster et al (2002), Atmos. Env., 36(22),
3659-3670.
46
Uncertainties in Climate System
Forest et al (2002), Science 295, 113-117.
47
Uncertainties in Impact Valuation
Roughgarden and Schneider (1999). Energy Policy
27 415-429.