Uncertainty analysis of national highway performance measures in the context of evolving analysis models and data

Effective management of national infrastructure systems requires analysis to develop projections for system performance and estimates of investment needs. Beginning in the late 1960s, the Federal Highway Administration has been required by law to estimate and report biennially on the nation's future highway investment needs. This infrastructure agency has developed a quantitative decision support system for monitoring and projecting highway system performance and investment needs. The agency continues, incrementally, to refine the decision support system. Prior to the 1990s, few formalized efforts had been made to understand uncertainty in the national highway decision support system and its impacts on highway performance estimation. Uncertainty is important in highway system management because it affects highway performance estimates. In the 90s, there have been various efforts to analyze uncertainty and associated risks in infrastructure decision support systems, using computer simulation techniques. These efforts have focused on data uncertainty without attempting to capture the uncertainty about analysis models. This research develops a simulation-based approach for uncertainty analysis of highway performance measures while addressing the impact of evolving analysis models and data within the highway decision support system. The approach characterizes the uncertainty introduced into performance estimation through evolving analysis models by capturing the relative repeatability and efficiency of the models over time. The approach is applied to analyze changes, and associated risks, in the performance of a portion of the nation's highway system. The analysis results show that changes in both the analysis model and data comparably affect the expected value and variability of highway performance estimates, giving evidence of the relative importance of model uncertainty. The results also show that there is consistency in the trends of highway performance estimates notwithstanding variation in the dominant sources of model output variability and the evolution of analysis models. These observations indicate that there may be value in developing a retrospective approach that systematically validates performance estimates relative to the actual performance of the highway system. The approach for relative model validation, developed in this research, contributes a useful framework for such a retrospective approach.