Structured Decision Making

Structured Decision Making
James T. Peterson
USGS Oregon Cooperative Fish and Wildlife Research Unit
TRRP Scientific Advisory Board
Homepage: http://people.oregonstate.edu/~peterjam/
SDM is the Application of the Scientific Method
to Natural Resource Decision-Making
The Scientific Method
Identify scientific phenomenon
of interest
Structured Decision Making
Identify alternative actions
Explicitly identify quantifiable objectives
Devise explicit falsifiable hypotheses
Create explicit models of ecological
process
(preferably framed as hypotheses)
Devise study with alternative
possible outcomes
Implement best management action
Conduct the study
Confront the hypotheses with the data
Monitor system and compare observed
patterns to model predictions
Multi-Stakeholder SDM
SDM requires agreement on objectives (not science)
Identify stakeholders
Significant/
deep rooted
conflicts?
Yes
Conflict resolution/
transformation
Yes
Conflicts
resolved/
transformed?
No
Initiate
SDM process
No
Organizing stakeholders:
Developing a governance system
Deciding how we decide
Essential before initiating multi-stakeholder SDM
Define structure
governing board
chair
Identify roles and responsibilities
Process for dispute resolution and decision making
Establish procedures for changing the governance
Reality check:
Decision Makers and Stakeholders
All decision makers are stakeholders
Not all stakeholders are decision makers
Decision makers
legal authority/mandate
management resources
My question for later discussion: in collaborative efforts, do decision makers give-up
decision making authority to stakeholders? To scientists?
Types of Decision Making
Autocratic
single entity maintains total control and ownership
responsible for outcome
Consultative
single entity receives stakeholder input
good for public relations
Democratic
one stakeholder one vote
generally perceived as fair
winners and losers
Consensus
collective agreement
preferred method
susceptible to gridlock (JP’s opinion)
Opportunities
for
collaboration
Case study: R.L. Harris Dam
Alabama Power Company’s most recent: April 20, 1983
Hydropeaking facility
Many Problems
Many stakeholders/objectives
No significant or deep rooted conflicts
Flow, m3/s
450
350
250
150
50
0
Hourly flow data: April 1 - August 31 1995
E. Irwin, ALCFWRU
Developed Governance and Charter
http://www.rivermanagement.org/charter.html
Alternative: Use existing structures
Structured Decision-Making Process
Identify the decision
situation and objectives
Identify the management alternatives
Break down and build model of the problem:
Identify the best alternative
Evaluate model sensitivity
Is further
analysis needed?
NO
Implement the best alternative
YES
All of this occurs
AFTER establishing
governance in multistakeholder SDM
Structured Decision Making Process
Step 1: Identify the problem / decision situation
Provides proper starting point and sideboards
Focus on the big problem
decision-makers
spatial dimensions
temporal dimensions
Identify the key elements of the problem
>> failure to identify problem hinders entire process<<
Step 2: Identify and Structure Objectives
Why emphasize objectives?
Decision
Model
Objectives
(must be
quantifiable)
Everything depends on your objectives
Everything depends on your objectives
Everything depends on your objectives
Basic types of objectives
Fundamental objectives: what the decision-maker really
wants to accomplish.
Means objectives: the things that need to be
accomplished to realize the fundamental objective
The importance of identifying and structuring objectives
common sticking point
Confusing fundamental and mean objectives
I
regimes
Water level
Stated (fundamental) objective of stream fishery manager:
Natural Hydrologic Regime
Time
Possible outcome: The flow regime is natural but….
all the fish are dead
?
I
regimes
Would the fishery managers be happy with the outcome???
Means objectives (sometimes) help
realize the fundamental objective
Maximize Fish
Population Size
Maintain natural
hydrologic regime
Maximize habitat
availability
Means objectives often are hypotheses about system
dynamics
More common problems
Dismissing potential
objectives due to
perceived conflicts
Dismissing potential
objectives due to
perceived lack of
information or complexity
Please leave your
model at the door
Values (objectives) masquerading as facts or process
Remember objectives must be quantifiable
Decision
Model
of system
Lots of approaches
Depends on objectives
Depends on stakeholders
Objectives
(must be
quantifiable)
Case study:
Quantifying CVPIA fundamental objectives
Fundamental objectives hierarchy
Naturally reproducing self-sustaining populations
Chinook salmon: Spring, Fall, Late-Fall, Winter
Spatial structure
No. of
spawning
popn per
diversity
group
No. of
returning adults
Abundance
Juvenile
abundance
Adult/juvenile
natural production
No. natural
juveniles/Natural
returning adults
Size at
emigration
Nat.
returning
adult/natural
spawner
Prop of
natural
spawners vs
hatchery
Juvenile size
distribution
All quantifiable entities
Weight each attribute and combine in single utility (score)
Diversity
Age
structure
Outmigration
timing
Return
timing
Harris dam case study: What if stakeholders can’t
agree on the relative value?
Decision
Dam operation plan A
Dam operation plan B
Dam operation plan C
Number of native
fishes
water levels in
the reservoir
Amount of
bank erosion
downstream
Number of
Downstream
float days
Show the stakeholders the individual estimated effects of each
dam operation decision and choose decision based on their
input
Step 3: Identify decision alternatives
What actions can be taken
sometimes limited- legal mandates, restrictions
within the authority of decision maker
Should be mutually exclusive and exhaustive
Can be a list of discrete actions, or the selection of an
action over some continuous range (e.g., a harvest
rate)
Must be specific and explicit
“Look at everything” approaches should be avoided
Structured Decision Making Process
Step 1: Identify the problem / decision situation
Step 2: Identify and structure objectives
Step 3: Identify decision alternatives
These 3 steps = most difficult aspects of SDM
Reminder:
Linking decisions to objectives
Decision
Model
Objectives
(must be
quantifiable)
Step 4: Model building
Construct the model (need to estimate the outcome!)
- Science-based
- Simple (simple is good!) preferred option
- Complex
Simple model example:
Least Chub Decision Model
Livestock grazing
management
Seasonal
precipitation
Water table
Emergent
vegetation
Water level
Extant of seasonal
marsh inundation
Restoration and
maintainence
Proximity to
exotic source
Wetland waterbody
types/condition
Continued
waterbody use
Refugial
populations
Ungulate
damage
Mosquitofish
presence
New
waterbody use
Waterbody occupancy
(persistence)
Net ecological
benefits
Reproduction
Complex model example:
CVPIA Green and white sturgeon models
CVPIA coarse scale alternatives
Screen diversions
Change diversion amount
Upper river rearing habitat
Lower river rearing habitat
Spawning habitat
Remove modify flood by...
Water availability
Current habitat
availability
(by type)
Current no. flood
control bypass
structures
Current amount of diversion
during juvenile presence
Future amount of diversion
during juvenile presence
Future unscreened
diversions
Water year
Stranding
Adult in river
pre-spawning
survival
Potential No.
returning adults
Delay in spawning
No arriving at
spawning area
Dredging
Ocean/
estuary
survival
Entrainment in
delta diversions
Delta inflows
July-Dec max.
water temp/ min DO
Predation
(egg- larval)
Spawning habitat
availability
Contaminants
Larval
survival
Harvest/bycatch
/poaching
Adult in river
post-spawning
survival
Water availability
Temperature
regulations
Upper mid river
rearing habitat
availability
Spawning success
(Larval production)
Non-spawning adults (t)
Climate
Current unscreened
diversions
Future no. flood
control bypass
structures
No. adults (t)
Current water
availability
No.
outmigrating
adults
Riverine
juvenile
survival
No. riverine
juveniles
Mid-lower river habitat
availability
No.esturine
juveniles
Estuary
carrying capacity
Juvenile
estuarine
survival
No. adults
(t+1)
Subadult
growth rate
No. subadults &
Marine juveniles
Juvenile
growth rate
Proportion of popn.
in each age class
No. reproducing popns
Management action
No. watersheds occupied
Population dynamic
No. adults
(reproductive size)
Age structure
Other
Growth
Utility
Model Complexity Considerations
Pallid sturgeon conceptual life history model
Need to have:
1) Values for each box!
2) Parameters for each arrow!
Monitoring data provide feedback
Multiple unobservable states = cannot differentiate among mechanisms/hypotheses
M. Colvin MSU
Step 5:
Identify key uncertainties--Sensitivity Analysis
•
An essential step
•
Basis for model simplification
•
Focus monitoring on decision-making
• what do we need to know
• how much is enough
•
Estimate value of information
• collecting monitoring data
• more studies
Example: CVPIA Sensitivity of green sturgeon model parameters
One way sensitivity analysis
Alternative larval survival models
Larval surv. pumping model water diverted
Reprd. success max success
Juvenile to adult transition
Adult ocean surv. poach
Transition young to old adult probability
Adult ocean surv. dredge mod
Larval surv. contaminants model no. impairments
Reprd. success female per pool
Adult ocean surv. dredge high
Prespwn mort. harvest
Juvenile estuarine surv. natural mort
Postspwn mort. prob. of stranding
Larval surv. predation model dry year
Larval surv. predation model wet year
Prespwn mort prob. of stranding
Age 2 riverine base survival
Abandon spawning if stranded prob.
Spawning delay when stranded
Alternative adult survival models
Reprd. success delay
Proportion adults spawning
Age 1 riverine base survival
Adult ocean surv. dredge low
Larval surv. pumping model unscreened
Juvenile estuarine surv. entrainment mort
Larval base survival
Juvenile river surv. pumping model unscreened
Proportion of larvae leaving river
Prespwn mort. temp LD100
Postspwn mort. harvest
Prespwn mort. DO LD100
Juvenile river surv. pumping model water diverted
Postspwn mort. temp LD100
Postspwn mort. DO LD100
Juvenile river surv. pumping model unscreened*diverted
Prespwn mort. temp LD50
Prespwn mort. DO LD50
Juvenile river surv. pumping model age2 effect
Postspwn mort. temp LD50
Postspwn mort. DO LD50
0.0
0.2
0.4
Utility
0.6
Real question: what uncertainties affect what managers decide?
Response profile sensitivity analysis
each line is a decision alternative, best decision top line
Optimal decision
changes once
Decision utility
0.6
0.4
0.2
0.03
0.06
Ocean/poaching mortality
0.09
Real question: what uncertainties affect what managers decide?
Response profile sensitivity analysis
Alternative hypotheses system dynamics
Age 0 survival Ha:
Adult survival Ha:
Behavioral
thermoregulation
Non-behavioral
thermoregulation
Decrease
amount of
diversion
10% at
Feather
River
Diversion and
entrainment
Modify flood bypass
control structure at
Freemont Weir
Create/improve
passage at
Daguerre Point
Dam
Predation
Contaminants
Increase inchannel
rearing
habitat 10%
at Lowermid
Sacramento
River
Habitat
limitation
Optimal decision changes multiple times!!
Reducing Uncertainty
• Retrospective studies
– Can provide a good initial basis for prediction
– Usually confounded with other factors
• Experiments
– Difficult to perform in many systems
– Uncertainty reduction is not directed at resource
objective (inefficient)
• Adaptive management
– Can be done in virtually any resource system
– No tradeoff necessary in resource objective
What do we need for ARM using SDM?
• Predictions under 2 or more models
• Sequential decision making
• MONITORING
Sequential decision making
Through time
Population
Population
Decision
Population
Decision
Population
Decision
Decision
time
In space and time
Site A
Site B
Site C
Site D
Site E
Site F
Site G
time
C. Moore GACFWRU
An illustration
Lower American River Chinook Salmon:
over simplified version Number juvenile outmigrants
What controls Chinook recruitment?
Spawning
habitat availability?
availability?
Juvenile habitat
Available
Spawningjuvenile
habitat habitat
availability
Management
action
System
dynamics
Estimated number
juvenile outmigrants
Juvenile habitat
300k
0.5
Create juvenile
habitat
150k
0.5
225
Spawning habitat
150k
Juvenile habitat
200
0.5
Create spawning
habitat
250k
0.5
Number outmigrants
Spawning habitat
Outcome: 210k outmigrants
0
1
2
3
4
Management
action
System
dynamics
Estimated number
juvenile outmigrants
Juvenile habitat
300k
0.39
Create juvenile
habitat
150k
0.61
209
Spawning habitat
150k
Juvenile habitat
211
0.39
Create spawning
habitat
250k
0.61
Number outmigrants
Spawning habitat
Outcome: 180 outmigrants
0
1
2
3
4
Adaptive Resource Management as
Dynamic SDM
•
Decisions are made
•
Sequential dynamic decision making
•
Two or more models represent alternative ideas of system dynamics
•
Model weights (information)
•
Monitoring
Information
System
statet
Information
t
Management
action t
Model A
System
statet + 1
Information
t+1
Management
action t+1
Predicted
State t+1
System
statet + 2
Model A
Predicted
State t+2
Model B
Predicted
state t+2
Bayes
Rule
Model B
Predicted
state t+1
Bayes
Rule
t+2
Single and double loop learning
Stakeholder objectives
Management alternatives
Integrated models
Implement ‘best’ management
alternative
Revise beliefs
(updating)
Observe outcome
(monitoring)
Single loop
Evaluate evidence
(analysis)
Double loop
Revise or develop new
Assess current assumptions
Objectives
Management alternatives
Models
Objectives
Management alternatives
Models
Deliberately
Obscured
Conflict Resolution
- - - - - - - - - - - - - - - - - - - - - - - - - - -- - - - -- - - - -
Complex
Structured
Decision
Making
Scenario
Planning
Adaptive
Management
(Dynamic SDM)
Concise
& Clear
Well
Understood
Uncertain
-------------------
OBJECTIVES CLARITY-COMPLEXITY
SDM Decision Space: When is SDM Appropriate?
Joint
Fact
Finding
Disputed
SCIENCE / TECHNICAL UNDERSTANDING
M. Runge PWRC