1. Problem Setup
House elevation under flood risk
After completing this tutorial series, you will be able to:
- Build a time-stepped simulation model for a decision problem under deep uncertainty
- Use exploratory modeling to understand how a policy performs across many plausible futures
- Use multi-objective optimization to find Pareto-optimal trade-offs
- Interpret scenario maps and Pareto fronts to support real-world decision-making
Why This Tutorial?
Many real-world decisions—infrastructure investments, climate adaptation strategies, resource management policies—must be made before key uncertainties are resolved. Standard cost-benefit analysis assumes we know (or can estimate) the probability distribution of future outcomes. But for many problems in climate risk, water management, and infrastructure planning, we face deep uncertainty: we cannot confidently assign probabilities to future states of the world.
Decision Making under Deep Uncertainty (DMDU) is a family of approaches that takes this challenge seriously. Rather than optimizing for a single “expected” future, DMDU methods stress-test candidate decisions across a wide range of plausible futures to identify strategies that perform well across many of them.
This tutorial teaches SimOptDecisions.jl through a realistic DMDU workflow, using the example of deciding how high to elevate a house in a flood-prone area (inspired by Zarekarizi, Srikrishnan, and Keller 2020; see also Doss-Gollin and Keller 2023).
The Decision Problem
You own a house in a flood-prone coastal area. Each year, storm surges threaten your property. You must decide: how high should you elevate your house?
Elevating higher costs more upfront, but reduces future flood damages. The challenge is that you face deep uncertainty about:
- Future storm surge intensity — The frequency and severity of future floods depends on climate variability and change, which we cannot precisely predict
- The depth-damage relationship — How much damage a given flood depth causes depends on building characteristics and contents, which are uncertain
- Future economic conditions — The discount rate used to compare present costs against future damages reflects contested value judgments
These aren’t just “noise” around a known distribution. Reasonable experts disagree about the form of the distribution, not just its parameters. This is what makes the problem one of deep uncertainty rather than well-characterized risk.
The Trade-off
- Low elevation: Cheap upfront, but expensive flood damages over time
- High elevation: Expensive construction, but minimal future damages
- “Optimal” elevation: Depends on your assumptions about the uncertain future
Different assumptions about storm intensity, damage functions, and discount rates lead to different “optimal” recommendations. There is no single right answer—only trade-offs to understand.
The DMDU Workflow
This tutorial walks through a complete DMDU analysis:
| Step | Tutorial Section | What You’ll Do |
|---|---|---|
| Build the model | 2. Running a Time Step | Define the physics: depth-damage, construction costs |
| Add time | 3. Types and simulate() | Structure the model with types and run multi-year simulations |
| Aggregate results | 4. Outcomes and Metrics | Summarize outcomes across scenarios; choose metrics |
| Explore | 5. Exploratory Modeling | Map out where policies succeed and fail across the scenario space |
| Optimize | 6. Policy Search | Find Pareto-optimal trade-offs with multi-objective optimization |
Each section builds on the previous, using the same house elevation context throughout.
Next Steps
In the next section, we’ll start building the model from the inside: what happens in a single year when a storm hits?