Infrastructure Reliability Assessment under Nonstationary Hazard Conditions: A Variance-Based Decision Support Framework

Abstract

Infrastructure systems with multidecade design lives face the fundamental challenge that hazard environments, organizational capabilities, and operational demands evolve over planning horizons that exceed the reliable range of historical calibration. This study presents a variance-based uncertainty decomposition framework for infrastructure reliability assessment, motivated by and designed for application in nonstationary hazard environments and demonstrated through application to naval dry dock facilities as representative critical infrastructure. The framework integrates process-based mechanism modeling, adaptive capacity evaluation, and variance-based sensitivity analysis within a Monte Carlo simulation architecture using Saltelli sampling with 2,560,000 model evaluations across 16 installation configurations. Applied to a synthetic demonstration model parameterized from literature-derived ranges, Sobol’ variance decomposition reveals that epistemic uncertainty—reducible knowledge gaps about organizational processes, component capacities, and degradation rates—accounts for 76.2% of total reliability variance under the exponential scaling relationship specified for this proof-of-concept. Structural sensitivity analysis across alternative functional forms confirms that epistemic dominance is sensitive to the scaling assumption ( S Epistemic : 0.788 under linear scaling, 0.252–0.333 under nonlinear forms), establishing that the 76.2% result is a model structural property rather than an empirical infrastructure finding. The 82% reducible uncertainty fraction identifies targeted data collection and organizational assessment as the highest-leverage investment strategies, with expected value of perfect information analysis illustrating decision value on the order of $18–28M over a 30-year planning horizon under the specified parameters. Scenario analysis demonstrates that configuration-dependent reliability differences are robust to calibration uncertainty ( S Parameter = 0.004 ). The framework provides a methodological foundation for risk-informed infrastructure decision-making where environmental loading distributions cannot be assumed stationary over the infrastructure lifecycle.