Concepts

Key ideas and vocabulary

The Problem Setting

SimOptDecisions targets a class of decision problems with three characteristics:

  1. Sequential dynamics: The system evolves over time through discrete steps (years, months, decision epochs). Actions taken now affect future states.
  2. Nonlinearity: Outcomes are nonlinear functions of inputs—threshold effects, saturation, path dependence—so closed-form solutions are unavailable. Monte Carlo simulation is the natural evaluation tool.
  3. Uncertainty: Key inputs (future climate, economic conditions, system parameters) are uncertain. Rather than optimizing for a single expected future, we evaluate decisions across an ensemble of plausible scenarios.

This setting is common in infrastructure design, climate adaptation, water resources planning, and natural hazard management. It draws on ideas from several traditions: stochastic optimization, optimal control, reinforcement learning, engineering design, and decision making under deep uncertainty (DMDU).

Key Vocabulary

Scenario
A plausible future state of the world. Not a prediction—a scenario represents one internally consistent set of assumptions about uncertain quantities. In SimOptDecisions, scenarios are concrete data objects containing pre-generated inputs.
Policy
A decision rule mapping system state to actions. Specifically, we use parametric policy rules, meaning policy rules that have parameters we can update, improve, or optimize.
Exploratory modeling
Running a model across many scenarios to understand system behavior, rather than to predict a single outcome (Bankes 1993). The goal is to discover where a policy is vulnerable.
Pareto front
The set of solutions where no objective can be improved without worsening another. In multi-objective optimization, the Pareto front characterizes the fundamental trade-offs in the problem.

How SimOptDecisions Fits

Workflow Step SimOptDecisions Feature
Define uncertain futures Scenario type with parameter wrappers
Define candidate decisions Policy type with tunable parameters
Simulate system response simulate() with user-defined callbacks
Explore across futures explore() → YAXArray result matrix
Search for optimal trade-offs optimize() → Pareto front

Further Reading

For a review of climate risk management and decision-making under uncertainty, see Keller, Helgeson, and Srikrishnan (2021) or Doss-Gollin and Keller (2023). For an overview of how these methods connect to broader multisector dynamics challenges, see Reed et al. (2022).

References

Bankes, Steve. 1993. “Exploratory Modeling for Policy Analysis.” Operations Research 41 (3): 435–49. https://doi.org/10/c7rgcr.
Doss-Gollin, James, and Klaus Keller. 2023. “A Subjective Bayesian Framework for Synthesizing Deep Uncertainties in Climate Risk Management.” Earth’s Future 11 (1). https://doi.org/10.1029/2022EF003044.
Keller, Klaus, Casey Helgeson, and Vivek Srikrishnan. 2021. “Climate Risk Management.” Annual Review of Earth and Planetary Sciences 49 (1): 95–116. https://doi.org/10.1146/annurev-earth-080320-055847.
Reed, Patrick M., Antonia Hadjimichael, Richard H. Moss, Christa Brelsford, Casey D. Burleyson, Stuart Cohen, Ana Dyreson, et al. 2022. “Multisector Dynamics: Advancing the Science of Complex Adaptive Human-Earth Systems.” Earth’s Future 10 (3): e2021EF002621. https://doi.org/10.1029/2021EF002621.