Conflict, Coordination, & Control: �Do we understand the actual rules used to balance flooding, energy, and ag tradeoffs?
Julianne Quinn, Patrick Reed*,
Matteo Giuliani and Andrea Castelletti
1 Cornell University
2 Politecnico di Milano
3 ETH Zurich
10 June 2022
March 21, 2019
2
10 June 2022
Key Points
Model-based understanding of the complex evolution of food-energy-water systems as well as their “risks” and “resilience”
Must be able to capture extremes and real failure modes.
Is heavily influenced by human preferences, tradeoffs in conflicting demands, and high-fidelity representations of candidate actions
Should create a platform for understanding state-action-consequence feedbacks as a function of the information available to the actual humans managing the systems
1
2
3
3
10 June 2022
Red River Basin
Second largest river basin in Vietnam
Capital city of Hanoi sits in delta, threatened by floods
In 2002, UNDP estimated annual damages of 130M USD in the delta, 50M USD in Hanoi1
1Hansson, K., and Ekenberg, L. (2002). Flood Mitigation Strategies for the Red River Delta, in: International Conference on Environmental Engineering, An International Perspective on Environmental Engineering, Canada.
4
10 June 2022
Red River Basin
To provide flood protection to Hanoi and the delta, the Vietnamese government has started constructing reservoirs
But how should they be coordinated to meet multi-sector demands?
5
10 June 2022
Multi-sector reservoir demands
Dams provide hydropower
Hydropower currently represents 46% of Vietnam’s total installed electric power capacity
Reservoirs provide water supply
70% of Vietnamese population employed in agriculture,
76% of Vietnamese agriculture is irrigated
6
10 June 2022
But will these demands change? How?
Population growth in Hanoi could increase water demands
Or urbanization could reduce water demands
7
10 June 2022
Will the climate change? How?
8
10 June 2022
Red River System Goals
Find operations for four largest reservoirs that
and are robust to deep uncertainties
How should we translate and evaluate these narrative goals in our models?
9
10 June 2022
Red River System
Official Guidelines
Flood Season
Dry Season
Between Seasons
Determine SL release, utSL
Determine HB release, utHB
Determine TQ release, utTQ
Determine TB release, utTB
Unregulated. Use release from one of our optimized policies.
10
10 June 2022
Red River System
Official Guidelines
Flood Season
Dry Season
Determine TB release, utTB
If stTB < stTB, lower target
Else
utTB = 0
utTB = utTB, min
Between Seasons
Determine SL release, utSL
Determine HB release, utHB
Determine TQ release, utTQ
Determine TB release, utTB
11
10 June 2022
Red River System
Official Guidelines
Flood Season
Dry Season
Determine TQ release, utTQ
Determine TB release, utTB
If stTQ < stTQ, lower target
Else
utTQ = 0
utTQ = utTQ, min
Between Seasons
Determine SL release, utSL
Determine HB release, utHB
Determine TQ release, utTQ
Determine TB release, utTB
12
10 June 2022
Red River System
Official Guidelines
Flood Season
If stHB < stHB, lower target
Else
utHB = 0
utHB = utHB, min
If t!=(27,28,41,42,55,56)
Dry Season
Determine TB release, utTB
Between Seasons
Determine TQ release, utTQ
Determine Preliminary HB release, utHB
Determine SL release, utSL
Determine HB release, utHB
Determine TQ release, utTQ
Determine TB release, utTB
13
10 June 2022
Red River System
Official Guidelines
Flood Season
Else
utHB = 0
If t=(27,28,41,42,55,56)
Dry Season
Determine TB release, utTB
Between Seasons
Determine TQ release, utTQ
Determine Preliminary HB release, utHB
Determine SL release, utSL
Determine HB release, utHB
Determine TQ release, utTQ
Determine TB release, utTB
14
10 June 2022
Red River System
Official Guidelines
Flood Season
Determine SL release, utSL
utSL = max(0,q needed to raise stHB to stHB, lower target)
Determine Preliminary HB release, utHB
Dry Season
Determine TB release, utTB
Between Seasons
Determine TQ release, utTQ
Determine SL release, utSL
Determine HB release, utHB
Determine TQ release, utTQ
Determine TB release, utTB
15
10 June 2022
Evolutionary Multi-Objective Direct Policy Search (EMODPS)
Computationally efficient method for solving high-dimensional, multi-objective control problems
Step 1:
Parameterization
Mar
Dec
Sep
Jun
Release
Value of Inputs
16
10 June 2022
Evolutionary Multi-Objective Direct Policy Search (EMODPS)
Computationally efficient method for solving high-dimensional, multi-objective control problems
Step 2:
Simulation
17
10 June 2022
Evolutionary Multi-Objective Direct Policy Search (EMODPS)
Computationally efficient method for solving high-dimensional, multi-objective control problems
Step 3:
Optimization
18
10 June 2022
Red River System
EMODPS Policies
19
10 June 2022
Red River System
EMODPS Policies
20
10 June 2022
Red River System
EMODPS Policies
Mar
Dec
Sep
Jun
21
10 June 2022
23
6/10/2022
chora.space
Many-Objective Tradeoffs
Visual analytics:
MOEA Search (Red)
Target Solution Set (Gray)
Notice how it not only finds the solution, but also distributes itself across the solution.
Three-objective Test Problem
Borg MOEA Parallelization
25
6/10/2022
Hadka, D., and Reed, P.M., “Large-scale Parallelization of the Borg MOEA for Many-Objective Optimization of Complex Environmental Systems”, Environmental Modelling & Software, v69, 353-369, 2015.
Monte Carlo Simulation of Scalability of Search
26
6/10/2022
Theoretical Scaling from Discrete Event Simulation (accurate to within 0.1%)
Reed, P.M. and Hadka, D., "Evolving Many-Objective Water Management to Exploit Exascale Computing", Water Resources Research, v50, n10, 8367–8373, 2014.
Official Control Rules vs. EMODPS Polices
So, how do these approaches compare?
27
10 June 2022
Objective Comparison
Guidelines are fully dominated, and domination should increase with # of reservoirs
Barely provide protection to the 100-yr flood
28
10 June 2022
Let’s pick a few to highlight
Guidelines are fully dominated, and domination should increase with # of reservoirs
Barely provide protection to the 100-yr flood
29
10 June 2022
Let’s look in more detail…100,000 simulated years
Guidelines do not effectively coordinate operations to make use of reservoir storage for flood protection
30
10 June 2022
Guidelines are not coordinating operations well
This is troubling given we have only looked at stationary hydrologic uncertainty.
What if we experience major changes in human demands or monsoonal extremes?
31
10 June 2022
Generating alternative states of the world
Goal: Sample broad range of hydrologic and socio-economic factors to discover, a posteriori, the most important drivers of system dynamics and performance
…
…
…
…
…
…
…
…
SOW 1:
SOW 1000:
…
7 Hydrologic
Factors
4 Socioeconomic
Factors
µ1
σ1
ag1
aq1
µ1000
σ1000
ag1000
aq1000
32
10 June 2022
Annual mean flow and inter-annual variability
33
10 June 2022
Demand changes
34
10 June 2022
Factors influencing flood failures
Guidelines have more failures
35
10 June 2022
Factors influencing flood failures
Guidelines have more failures
Failures explained by 2 major factors:� Mean flow, μ
Inter-annual variability, σ
36
10 June 2022
Factors influencing hydropower failures
Same controlling factors, but failure regions are opposite
37
10 June 2022
Factors influencing deficit failures
Controlled predominantly by socio-economic factors:
Agricultural demand, ag
Other demand, o
38
10 June 2022
Defining a safe operating space (SOS)
SOS does not encompass base SOW
Cannot provide protection to 100-yr flood with 95% reliability
39
10 June 2022
Frequently Ignored Issues in Climate Assessments
Simple discrete if/then/else-based human systems abstractions lack fidelity and likely to inadvertently ignore major failures modes
Deterministic model “fits” to historical observations do not reflect rare events or the extrapolation of how they are changing. This is not a regression problem…it’s an extrapolation problem
Poor abstractions of sequential decision-making, coordination failures, sectoral conflicts, and poor use of information will cause severe errors in projecting candidate future pathways
Human institutions, land rights/competition, economic and technology transitions, infrastructure investments, etc. all can have huge landscape effects with small changes
1
2
3
4
40
10 June 2022
Thanks! Any questions?
Acknowledgements
NSF SCRiM #GEO-1240507
Julie Quinn
Matteo Giuliani
Andrea Castelletti
41
10 June 2022
Appendix
10 June 2022
Sensitivity of utHB with Different Policies
43
10 June 2022