AFS - Spokane Apr 27 to May 1, 2002ESSA Technologies A decision analysis of adaptive management experiments: Is it worth varying flows to reduce key uncertainties?
Post on 26-Mar-2015
216 Views
Preview:
Transcript
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
A decision analysis of adaptive management experiments: Is it worth
varying flows to reduce key uncertainties?
An application to Columbia River whitefish management
AFS - Spokane;
Apr 27 to May 1 2002
Developed byESSA Technologies Ltd.
Clint Alexander, Paul Higgins*, David Marmorek, and Calvin Peters
* Funded by BC Hydro Power Supply & Watershed Management
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Outline
• Study Area / Problem• Objective• Methods
– Decision analysis
– Model
• Results: Is it worth it?• General conclusions
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Problem IProblem I: Increased egg mortality from dam operation
Flow during spawningFlow during spawning
Flow during Flow during incubationincubation
stage
Proportion eggs in de-watered area
Some flexibility to regulate flows during spawning (January 1 - 21)
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Problem IIProblem II: Uncertainty True whitefish recruitment dynamics?
No reliable baseline information
Alternative Hypotheses
-
5,000
10,000
15,000
20,000
25,000
0 5 10 15 20 25
Eggs Just Prior to Hatching (millions)
Age
4 W
hite
fish
Very Sensitive
Sensitive
Neutral
Insensitive
Very Insensitive
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Study Objective
Use Decision Analysis to evaluate benefits and costs of alternative
spawning flows & monitoring programs
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Methods
• Decision Analysis
• Age-structured simulation model– 1) population dynamics– 2) simulated mark-recapture egg & age4 abundance
estimates
– site specific stage and wetted-area at depth data generated from hydraulic flow simulations (HEC2-RAS model)
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Management Objective
• Maintain “least cost” whitefish population nearest
to or greater than 45,000 adults
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Stage 1 - Base Decision Analysis
Columbia RiverFlows During
WhitefishSPAWNING
Kootenay RiverFlows During
WhitefishSPAWNING
Min. Columbia RiverFlows Prior to
WhitefishHATCHING
Egg-Age4RecruitmentRelationship
EggAbundance
Abundance4+ Recruits
50 kcfs Model
30 kcfs
20 kcfs
85 kcfs
80 kcfs
20 kcfs
15 kcfs
10 kcfs
55 kcfs
25 kcfs
20 kcfs
15 kcfs
85 kcfs
a3, b 3
a2, b 2
a1, b 1
a5, b 5
a4, b 4
ForegonePower
Revenues
......
ManagementActions States of Nature and their Probabilities Outcomes
Min. KootenayRiver Flows Prior
to WhitefishHATCHING
20 kcfs
15 kcfs
10 kcfs
55 kcfs
...
40 kcfs
45 kcfs
55 kcfs
60 kcfs
65 kcfs
70 kcfs
Natural variability in flow Uncertainty due to lackof understanding / data
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Stage 1 Results: Current Uncertainty
HKD Spawning Q
(kcfs)
Expected Adults (yr50)
Expected de-watering mortality
Annual Foregone Revenue
($mil)
Maximum Annual
Potential Revenue ($mil)
Expected Egg Abundance (millions)
20 30,190 0.003 3.75 0.02 6.430 36,143 0.010 1 0.5 9.240 41,754 0.034 0.4 1.25 12.445 43,967 0.054 0.15 1.61 13.850 45,286 0.087 0 1.95 14.755 45,827 0.127 0 2.15 15.360 45,776 0.172 0 2.4 15.565 44,768 0.230 0 2.6 15.070 43,274 0.292 0 3 14.280 38,707 0.429 0 5.75 11.785 36,309 0.492 0 9 10.3
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Experimental flowsduring whitefish
spawning + Populationmonitoring
Kootenay RiverFlows During
WhitefishSPAWNING
Min. Columbia RiverFlows Prior to
WhitefishHATCHING
Assumed truestate,
includingprocess error
eA+10meH
ModeleA+10meL
eP10meL
eC10meH
20 kcfs
15 kcfs
10 kcfs
55 kcfs
25 kcfs
20 kcfs
15 kcfs
85 kcfs
a1, b 1
......
AdaptiveManagementExperiments States of Nature and their Probabilities Outcomes
Min. KootenayRiver Flows Prior
to WhitefishHATCHING
20 kcfs
15 kcfs
10 kcfs
55 kcfs
...
a 3, b 3
a 2, b 2
a 1, b 1
a 5, b 5
a 4, b 4
1. Using updated probabilities, repeatdecison analysis
2. Compare value of what youwould do with and without the newinformation from AM
3. Assess trade-offs, and "pick" thebest experiment!
UpdatedProbabilities
for hypotheses
Cost of Experiment(collecting information
+ foregone powergeneration)
Revised Probabilities
Sensitive - Low natural varation
0
0.2
0.4
0.6
H1 H2 H3 H4 H5
P
Current knowledge
eP10meL-$2.3
eA+10meL-$20.3
eA+10meH-$19.23
eA10meL-$4.55
eC10meL-$1.55
eP10meH-$1.23
eA10meH-$3.48
eC10meH-$0.48
...
Repeat stage 1...
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Stage 2 - Simulated learning from flow experiments and monitoring
Uses same model and uncertain components but...
Actions are now alternative experimental Actions are now alternative experimental flow regimes + monitoring programsflow regimes + monitoring programs
Assume a true relationship for population Assume a true relationship for population dynamics with process errordynamics with process error
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
4 experimental flow regimes4 experimental flow regimes
Year Active Constant Active+1 35 55 202 65 55 75
etc… … … …
Average January flows used in "passive" experiments
60
75
60
7580
70
35
5550
45
30
50
70
90
1 2 3 4 5 6 7 8 9 10
Simulation year
Q (kcfs)
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Learning from AM experiments: a function of what practitioner can and cannot control
Spatial / temporalcontrast in mgmt.
actions(e.g., flow)
Level precision/investment in
monitoringNatural variability
(added noise)
Ability to distinguish alternative hypotheses w AMexperiments
Value of information for decisions
Under AM practitioners control
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Would you change if you knew the “truth”?Expected adult N, year 50
25,000
30,000
35,000
40,000
45,000
50,000
20 30 40 45 50 55 60 65 70 80 85
HKD Spawning Q (kcfs)
N
Minimum desired
Current Uncertainty
Sensitive
Insensitive
10
5
2.5
7.5
$Cnd mil
Max. potential power revenues (per yr)
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Stage 2 Results: Good monitoring is critical for learning; flow manipulation has less effect than
expected.
Flow manipulation
High Meas. Error
Low Meas. Error
High Meas. Error
Low Meas. Error
Constant 0.55 ($0.48) 0.88 ($1.55) 0.51 ($0.48) 0.74 ($1.55)Passive 0.60 ($1.23) 0.92 ($2.3) 0.57 ($1.23) 0.85 ($2.3)
Active 0.63 ($3.48) 0.92 ($4.55) 0.54 ($3.48) 0.85 ($4.55)
$ CDN MillionsBlue = things under AM practitioners controlRed = beyond AM practitioners control
Probability identify insensitive population (10-year experiments)
Low Nat Variability High Natural Variability
Natural Variability and Measurement Error
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Stage 2 Results: It’s easier to identify a sensitive population than an insensitive one.
Flow manipulation
High Meas. Error
Low Meas. Error
High Meas. Error
Low Meas. Error
Constant 0.84 ($0.48) 0.90 ($1.55) 0.81 ($0.48) 0.82 ($1.55)Passive 0.90 ($1.23) 0.99 ($2.3) 0.88 ($1.23) 0.97 ($2.3)
Active 0.84 ($3.48) 0.90 ($4.55) 0.84 ($3.48) 0.85 ($4.55)
$ CDN MillionsBlue = things under AM practitioners controlRed = beyond AM practitioners control
Probability identify sensitive population (10-year experiments)
Low Nat Variability High Natural Variability
Natural Variability and Measurement Error
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Low measurement error critical to differentiating hypotheses
Inensitive - eC
0
0.2
0.4
0.6
H1 H2 H3 H4 H5
Current knowledgeNatVar-L - MErr-L-$1.55NatVar-H - MErr-L-$1.55NatVar-L - MErr-H-$0.48NatVar-H - MErr-H-$0.48
Inensitive - eA
0
0.2
0.4
0.6
H1 H2 H3 H4 H5
Current knowledgeNatVar-L - MErr-L-$4.55NatVar-H - MErr-L-$4.55NatVar-L - MErr-H-$3.48NatVar-H - MErr-H-$3.48
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
AM can “pay for itself”
Flow manipulation
High Meas. Error
Low Meas. Error
High Meas. Error
Low Meas. Error
Constant $0.2 (2.4) $0.6 (2.6) $0.2 (2.4) $0.2 (7.7)
Passive $0.2 (6) $0.6 (3.8) $0.2 (6.15) $0.6 (3.8)
Active $0.6 (5.8) $0.6 (7.6) $0.2 (17.4) $0.6 (7.6)
$Cnd millionsNumbers in brackets = experimental pay-back interval in yearsBlue = things under AM practitioners controlRed = beyond AM practitioners control
I ncrease in annual power revenues from operating with experimental information (insensitive population only, 10-year experiments)
Low Nat Variability High Natural Variability
Natural Variability and Measurement Error
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Caveats about the Keenleyside Results
• Small difference in performance of alternative experiments was surprising. Why?
– Large uncontrollable natural variation in flows, at both spawning and hatching, creates year-to-year variability in egg mortality (Kootenay R influence)
– “Passive” flows not actually that passive (large spawning flows informative)
– Model added too much measurement error (true detection probability higher than shown)
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Is AM worth it?Is AM worth it?
“Yes” IfNew information leads to choice of a
different management action that better satisfies a particular objective
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
FactorUnder AM practitioners control
Benefits of AM decision analysisBenefits of AM decision analysis
Management objective(fish vs. power $)
Ability to do well designed experiments
Initial level of uncertainty in alternative hypotheses
Magnitude of natural variability in the system
What “truth” really is
Inherent sensitivity of best action to uncertainty
Yes
Yes
May be known
No
No (can’t know without doing the experiment)
No
Yes
Yes
Yes
Yes
Yes
Yes
Can evaluate implications using decision analysis?
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
Natural variation in recruitmentNatural variation in recruitment
Low natural variation: sigma = 0.25
0
20000
40000
60000
80000
1 11 21
High natural variation: sigma = 0.45
0
20000
40000
60000
80000
1 11 21
ESSA TechnologiesAFS - Spokane Apr 27 to May 1, 2002
General ConclusionsGeneral Conclusions
• Value of AM potentially large
• Whether to proceed depends on “the kind” of system you are in (i.e. previous factors)
• Decision Analysis is very helpful for evaluating these benefits
– Decisions more robust to uncertainties (reduces risk - explicitly accounts for uncertainties)
– forces clarification of problem & uncertainties
– Determine which uncertainties have strongest effect on choice of “best” management decision (identify research priorities)
top related