Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Polasky decision making under great uncertainty
1. Decision-Making under Great
Uncertainty: Environmental
Management in an Era of Global
Change
Stephen Polasky
University of Minnesota
2. Introduction
Two large related issues of ecosystem
management:
Management of ecosystems to provide multiple
ecosystem services
Spatially-explicit integrated ecological-economic
analysis
Management under uncertainty in an era of
global change
Global change clouds our ability to accurately predict
future conditions
3. Current state of affairs
Ecosystems provide a wide array of goods and
services of value to people (“ecosystem
services”)
Human actions affect ecosystems and the
services they provide
The provision of ecosystem services often is not
factored into important decisions that affect
ecosystems
As a result, we often do a poor job of ecosystem
management
7. Introduction
Global change issues are complex and
the consequences of decisions are often
highly uncertain
Large spatial and temporal scales
Large stakes involved
Vitally important to try to take account of
present and potential future
consequences in decision-making
But very difficult to do so
8. Introduction
How can we best guide decision-making
to meet present and future human needs*
in an era of global change given pervasive
uncertainty?
Note: could substitute alternative
objectives for human needs
Biodiversity conservation
Environmental health
Evolutionary process
9. Outline
An approach to ecosystem management taking
account of multiple ecosystem services at
landscape scales
Examples from Minnesota and Oregon
Approaches to decision-making under
uncertainty:
Decision theory
Scenario planning
Thresholds approach
Resilience thinking
Thoughts on integrated assessment of
ecosystem management under great uncertainty
13. “InVEST”
Integrated Valuation of Ecosystem
Services and Tradeoffs
http://www.naturalcapitalproject.org/InVEST.html
Frontiers of Ecology
and Environment
Feb 2009
14. Modeling multiple ecosystem services and
tradeoffs at landscape scales
Nelson et al. 2009. Frontiers in Ecology and Environment 7(1): 4–11.
15. Modeling multiple services under
alternative scenarios
Three scenarios of land use / land cover
change for the Willamette Basin
developed by the Willamette Partnership
for 1990 – 2050
• Plan trend
• Development
• Conservation
16.
17. Modeling multiple services under
alternative scenarios
Model outputs: service provision and biodiversity
Water quality
Storm peak mitigation
Soil conservation (sediment retention)
Climate stabilization (carbon sequestration)
Biodiversity (species conservation)
Market returns to landowners (agricultural crop
production, timber harvest and housing values)
20. The Impact of Land Use Change on Ecosystem
Services, Biodiversity and Returns to Landowners:
A Case Study in the State of Minnesota
Photo by Raymond Gehman, National Geographic
Polasky et al. Environmental and Resource Economics 2011
21. Introduction
Use InVEST to analyze how changes in
land use in Minnesota affect ecosystem
services and biodiversity conservation
Compare the impact on ecosystem
services & biodiversity from:
Actual land use change from 1992- 2001
Alternative land use change scenarios
22. Land use scenarios
Use National Land Cover Database (NCLD) for 1992 to
2001 for data on actual land use change in Minnesota
Alternative land use scenarios:
No agricultural expansion
No urban expansion
Agricultural expansion into highly productive soils
Forestry expansion into highly productive forest parcels
Conservation: low productivity ag land and ag land within a 100
m buffer of waterways in MN River watershed were converted to
pre-settlement vegetation
23. InVEST outputs
Ecosystem services
Carbon sequestration
Water quality (phosphorus exports in the Minnesota River Basin)
Biodiversity
Grassland bird habitat
Forest bird habitat
Overall biodiversity (all natural habitat)
Returns to landowners
Value of agricultural production
Value of timber production
Value of urban/suburban development
28. Change from 1992 to 2001 by scenario: market
returns to agriculture, forestry, urban
Agriculture
Forestry
Urban
Million 1992 US $
29. Annual value from land use change
scenarios 1992-2001
Actual land No ag No urban Ag Forest Conser-
use expansion expansion expansion expansion vation
Change in total value:
carbon, water quality, ag
& forest production,
$3,328 $3,407 $3,040 $2,742 $3,300 $3,380
urban using actual
prices (M1992 $)
Change in returns to
landowners: ag & forest
production, urban using $3,320 $3,343 $3,027 $3,418 $3,292 $3,221
actual prices (M1992 $)
30. Summary
The failure to incorporate the value of ecosystem
services in land use planning can result in poor
outcomes
Low level of ecosystem services
Low value of total goods and services from landscape
Agricultural land use change had a bigger effect on
ecosystem service value and biodiversity than
urbanization
Result is largely due to the fact that there is far more agricultural land
than urban land
Urban land: generates negative externalities but the direct value of
urban land use is high
Agriculture: generates negative externalities but with lower direct land
use value
31. Challenge
Analysis assumes we know the links between
Action and ecosystem impact
Ecosystem functions and ecosystem services
Ecosystem services and human well-being
Each link is subject to considerable uncertainty
Global change means that future cause-effect
relationships may look quite different than
current relationships
33. Decision theory
Standard approach for rational choice under uncertainty
Specify objective function (e.g., maximize expected
human well-being)
Define uncertainty: probability of alternative states
Use best available information to specify how states and
actions combine to form outcomes (probability of
outcomes)
Define the net benefits of different outcomes
Choose action that maximizes the objective function
34. Example: guilty or not guilty?
Actual condition
Guilty Innocent
Guilty Correct False conviction
(Type I error)
Decision decision
Innocent False release Correct
(Type II error)
decision
35. Example: guilty or not guilty?
Version 1
Actual condition
(Unknown)
Guilty (p) Innocent
(1-p)
Guilty 20 -40
Decision
Innocent -5 10
36. Guilty or not guilty?
How high does the probability of guilt need
to be before you issue a guilty verdict?
Expected value of guilty verdict:
20P+(-40)(1-P) = -40+60P
Expected value of innocent verdict:
-5P+10(1-P) = 10-15P
As long as the probability of guilty is at
least 2/3, then issue a guilty verdict
(otherwise not)
37. Example: guilty or not guilty?
Version 2
Actual condition
(Unknown)
Guilty (p) Innocent
(1-p)
Guilty 20 -400
Decision
Innocent -5 10
38. Guilty or not guilty?
As long as the probability of guilty is at
least 0.9425 then issue a guilty verdict
(otherwise not)
If the cost of false imprisonment is high
then you must have a very high probability
of guilt before issuing a guilty verdict
“Innocent until proven guilty” – high cost of
false imprisonment (US criminal law)
39. Application to climate change and
mitigation actions
Actual condition
(Unknown)
Low sensitivity High sensitivity
to climate to climate
change (p) change (1-p)
Business-as- 20 -400
usual
Decision
Reduce GHG -5 10
emissions
40. Decision theory
Would only choose to continue on
business-as-usual path if you placed a
probability of 0.9425 of low sensitivity to
climate change
41. Decision theory
Decision theory is NOT hypothesis testing
Do not need to be 95% sure that climate
change is real before we act
Decision theory applied here minimizes
expected losses
Even if potential loss is uncertain, the optimal
action is often to avoid large potential losses
42. Risk versus uncertainty
Critics of decision-theory say that you may not
know what the probabilities are
“True uncertainty”
Unknown probabilities (P)
Unknown outcomes (don’t know possible states)
Former Defense Secretary Donald Rumsfeld:
“As we know, there are known knowns. There are
things we know we know. We also know there are
known unknowns. That is to say we know there are
some things we do not know. But there are also
unknown unknowns, the ones we don't know we don't
know.”
44. Risk versus uncertainty
With known probabilities of all possible
events – maximizing expected utility is a
reasonable rule
With “unknown unknowns” what do you
do?
Almost all important global change issues
have some element of “unknown
unknowns”
45. Decision theory response to
uncertainty challenge
Assign subjective probabilities then proceed to maximize
expected utility
Subjective probabilities – combination of available
information and best guesses (or opinion)
Predicting future mean global temperature in 100 years
with triple the greenhouse gas concentration from pre-
industrial times
• With some basic physics we can probably “narrow” the possible
range
• Could assign equal probability to every temperature in this range
• Or use best available information to assess probabilities, for
example, a normal distribution with a mean of +3C from today and
standard deviation of +/- 2C
46. Other approaches to decision-
making under uncertainty
Social-ecological systems are complex systems
The future trajectory of social-ecological systems
under global change is subject to considerable
uncertainty
Hard to understand system behavior and harder
still to predict the likely impacts of decisions
Complexity has led some to think the decision
theory approach is not well suited to provide
guidance for managing social-ecological
systems under global change
48. Scenario planning
Scenario planning is a method for thinking
creatively and systematically about complex
futures
Scenarios: sets of plausible stories, supported
with data and simulations, about how the future
might unfold from current conditions under
alternative human choices
Scenarios can address many important
uncertainties and contrasting beliefs about the
future of the system
Example: IPCC scenarios of future emissions
and consequences
49. Scenario planning
Scenarios were first used in the analysis of
global change during the 1970s
More recent efforts:
Global Environmental Outlook
Special Report on Emissions Scenarios (SRES) of the
Intergovernmental Panel on Climate Change (IPCC)
Millennium Ecosystem Assessment
These studies explored widely contrasting
alternative visions using quantitative models and
a diverse set of quantitative indicators
50. Scenario planning
Advantage of scenario planning is that it expands horizons (“blue
sky thinking”) that gets people thinking about potential outcomes
Weakness: difficulty of assessing the likelihood of alternative
futures
SRES presented six scenarios – all “equally sound”
High uncertainty prevented realistic assessments of probabilities
Criticism that the lack of probabilities limited the value of the scenarios
to decision-makers
Scenario planning can be combined with more quantitative decision
theory analysis
Scenario planning as first stage – get universe of potential outcomes
Robust decision-making to analyze more and less desirable decisions in
light of potential outcomes
52. Thresholds Approach
Social–ecological systems are complex
adaptive systems that can exhibit
nonlinear dynamics, historical
dependency, have multiple basins of
attraction and limited predictability
When crossed, thresholds between
multiple basins of attraction can lead to
fundamental transformations in system
feedbacks and dynamics
53. Thresholds Approach
Focus attention on critical boundaries that have major
consequences if crossed
Examples of the application of thresholds in global
change:
Planetary boundaries (Rockstrom et al. 2009)
Limits on emissions to avoid dangerous climate change (e.g. cap
of 450 ppm CO2e)
Thresholds are often used in regulatory or legal contexts to
distinguish permissible from impermissible activities
Thresholds can be used as a screen to rule out actions
thought to have too high a risk of crossing a threshold or
to rank actions based on risk
54. Thresholds approach
There is often uncertainty about the exact level
of a threshold
Decision-making involves choices about what
risks are acceptable: putting more stress on the
system can increase current benefits but at a
cost of having a higher probability of crossing a
critical threshold
Thresholds have been criticized as giving a false
impression that degradation below the threshold
level is ‘safe’ and improvements beyond a
threshold are of no value
56. Resilience thinking
Resilience thinking focuses on critical
thresholds for system performance
If current situation is desirable then
manage in ways to increase resilience
Resilience: ability to withstand shocks/perturbations
and remain within one basin of attraction (Hollings
resilience)
Build capacity to recognize and respond to
emerging transformations before they occur
Build capacity to adapt should transformation
occur
58. Resilience
Ways to shift between stable equilibria
Source: Scheffer et al. Nature 2001
59. Resilience thinking
Resilience thinking:
How can we remain within certain dynamic
limits to avoid crossing thresholds and remain
in a desirable state
In addition, how can be build capacity to cope
with change should a change of regime occur
Resilience thinking is more of a general
framework for thinking than specific
management prescriptions
61. Application to climate change
Question: how much and how fast should we
reduce GHG emissions?
Dueling approaches to answering this question
Nordhaus Dynamic Integrated Climate Economy
(DICE) Model
Stern Review of Climate Change
Framing climate change as minimizing risk of
crossing a threshold or maximizing expected
utility
62. Stern Review on the Economics of
Climate Change
It is a “stern review” but in reality it is
named for Nicholas Stern, an economist,
who headed the study
UK Treasury Department 2006
Created a splash
Proponents of strong action to address
climate change trumpet the report
Critics say the report is fundamentally flawed
and the conclusions are unwarranted
63. Stern Review on the Economics of
Climate Change
Main findings:
“…the benefits of strong, early action considerably
outweigh the costs.”
“…if we don’t act, the overall costs and risks of climate
change will be the equivalent to losing 5% of global GDP
each year, now and forever. If a wider range of risks and
impacts is taken into account, the estimates of damage
could rise to 20% of GDP or more…”
“Resource cost estimates suggest that an upper bound
for the expected annual cost of emissions reductions
consistent with a trajectory leading to stabilization at
550ppm CO2e is likely to be around 1% of GDP by
2050.”
64. Nordhaus DICE Model
Strong action at present is at odds with most prior
economic analyses of the climate change (e.g. Nordhaus
analysis)
Nordhaus policy recommendations: “climate change
ramp” - low initial price of carbon that ramps up over
time
Costs of moderate action are lower than rapid adjustment
Technological change lower future costs
Society will be richer in future and better able to afford costs
associated with climate change
This approach will not result in stabilization at 550 ppm
CO2 – concentrations will continue to rise
65. Why do Nordhaus and Stern
disagree?
Discounting
Estimates of damages
Uncertainty – and general approach to
climate change policy question
66. Uncertainty and general approach
to climate change policy question
Nordhaus: question of balancing costs
and benefits
Takes estimates of benefits of avoided
damages along with costs
Use integrated assessment model and inter-
temporal optimization to find the efficient
climate change policy
67. Uncertainty and general approach
to climate change policy question
Stern: risk assessment and cost
First do an analysis of risk and find a target for atmospheric
concentrations (550 ppm of CO2e)
Second figure out cost-effective way of achieving the target
On the standard cost-benefit analysis approach:
“As I have argued, it is very hard to believe that models where
radically different paths have to be compared, where time
periods of hundreds of years much be considered, where risk
and uncertainty are of the essence, and where many crucial
economic, social, and scientific features are poorly understood,
can be used as the main quantitative plan in a policy argument.”
Modeling of costs and benefits to optimize emissions is “still less
credible”
68. Application to ecosystem
management
Social-ecological systems: dynamic and interconnected
Climate change problem is almost easy by comparison
Single dimension – CO2 concentration
Ecosystem management – multiple dimensions
Water quality
Water flow (flood protection, drought mitigation)
Habitat and species
Agriculture & timber productivity
Livelihoods and jobs…
Do we understand systems well enough to predict short-
term and long-term consequences of management
actions on services?
69. Application to ecosystem
management
Path forward in ecosystem management
Building systems models to improve understanding of
potential system dynamics
Verification of models with data
Sensitivity analysis to probe uncertainty
Scenario thinking to broaden scope of analysis
(reduce blindspots)
Focus on potential thresholds with dramatic
consequences for system performance
Provide early warning to avoid thresholds and plan for
adaptation in case thresholds are crossed
70. Conclusions
Paul Krugman writing about the financial
crisis (September 2009):
“…an all-purpose punch line has become
‘nobody could have predicted. . . .’ It’s what
you say with regard to disasters that could
have been predicted…”
71. Conclusions
Essential role for conservation
professionals
Provide better understanding of social-
ecological systems
Highlight potential large-scale changes that
could have detrimental impacts
Provide early warning signs of danger
Provide guide-rails to keep us from crossing
thresholds
72. Conclusions
We do not know enough BUT…
We know enough to improve on current
performance
The long road rather than the quick fix:
Better science to improve understanding
Better institutions/policy that incorporates
science and reduces risks
Adaptive process that learns through time
Notes de l'éditeur
Should you get rid of one of these – confusing.. Does not change relative ranking across scenarios
Actual prices in red versus no price change in green. No price change reflects the change from land use Prices dropped and cost rose between 96-01 for agriculture products. Urban and forestry prices rose.