[Paper Review] Environmental policy in the context of complex systems: Statistical optimization and sensitivity analysis for ABMs

Dylan Munson, Arijit Dey+1 more|arXiv (Cornell University)|Feb 19, 2026

Complex Systems and Time Series Analysis0 citations

TL;DR

The paper develops a machine learning–driven framework to accelerate policy optimization in costly agent-based models (ABMs) via Gaussian-process sensitivity testing and Bayesian optimization, demonstrated on Sugarscape to derive fast, interpretable policies.

ABSTRACT

Coupled human-environment systems are increasingly being understood as complex adaptive systems (CAS), in which micro-level interactions between components lead to emergent behavior. Agent-based models (ABMs) hold great promise for environmental policy design by capturing such complex behavior, enabling a sophisticated understanding of potential interventions. One limitation, however, is that ABMs can be computationally costly to simulate, which hinders their use for policy optimization. To address this, we propose a new statistical framework that exploits machine learning techniques to accelerate policy optimization with costly ABMs. We first develop a statistical approach for sensitivity testing of the optimal policy, then leverage a reinforcement learning method for efficient policy optimization. We test this framework on the classic ``Sugarscape'' model, an ABM for resource harvesting. We show that our approach can quickly identify optimal and interpretable policies that improve upon baseline techniques, with insightful sensitivity and dynamic analyses that connect back to economic theory.

Motivation & Objective

• Motivate the use of ABMs for environmental policy under complex adaptive systems.
• Develop a statistical framework to test sensitivity of optimal policies to system state variables.
• Implement ML-based acceleration (Gaussian processes) for sensitivity testing and Bayesian optimization for policy search.
• Demonstrate the framework on the Sugarscape ABM to identify optimal policies efficiently.
• Link policy findings to economic theory through interpretability and dynamics analyses.

Proposed method

• Model the ABM response G(x,θ) as a stochastic function with policy x and state θ inputs.”
• Use Gaussian process models to test additivity of f(x,θ) and perform a likelihood ratio test to assess sensitivity of the optimum to θ.
• Formulate f(x,θ) = Ψ{G(x,θ)} and minimize it via optimization, noting costly ABM evaluations.
• Apply Bayesian optimization with Expected Improvement to efficiently search the policy space for fixed θ, updating the GP with ABM runs.
• Compute acquisition using EI to balance exploration and exploitation, and iterate until convergence.

Figure 2: Distribution of pollution, $t$ = 0

Experimental results

Research questions

• RQ1 Is the optimal policy x*(θ) sensitive to state variables θ?
• RQ2 Can Bayesian optimization efficiently identify optimal policies for costly ABMs compared with baselines?
• RQ3 Does additivity hold for the objective f(x,θ), indicating θ-insensitive policies?
• RQ4 How does the proposed framework perform on the Sugarscape ABM in terms of speed and interpretability?
• RQ5 Do sensitivity and dynamic analyses yield insights that align with economic theory?

Key findings

• The framework rapidly identifies optimal and interpretable policies that improve upon baseline techniques.
• Sensitivity testing reveals whether the optimal policy depends on state variables, enabling targeted policy optimization.
• Bayesian optimization with Gaussian processes accelerates policy search under costly ABM evaluations.
• The approach provides insightful sensitivity and dynamic analyses that can be related to economic theory.
• The methodology is adaptable to a wide range of ABMs beyond the Sugarscape example.

Figure 3: Distribution of pollution, $t$ = 500