What Do Local and Global Sensitivity Analyses Tell Agencies about Asset Management Decision Support Tools?

Abstract

Decision support models are widely used to make efficient decisions about the maintenance, repair, and renewal of aging transportation infrastructure. These decision support models often involve selecting several parameters to realistically represent the system while constraining the computational cost, which increases as the number of parameters increases. However, these input parameters are characterized by uncertainty in their properties and values, thus making it difficult to determine the correct parameters for the models confidently. Traditional or local sensitivity analysis provides insights into selecting appropriate values, but does not account for parameter interactions. Global sensitivity analysis helps identify the most influential parameters. Local and global sensitivity analyses were conducted using simulated input–output data from a life-cycle cost assessment model for planning highway infrastructure maintenance and rehabilitation. The sensitivity analysis helps explain changes in the magnitude of the output and the selected maintenance policy from changes in the model parameters. Local sensitivity analysis indicated that the optimal policy selected using default parameters provided results close to the optimal policy if any parameter changed. Using Sobol’s method, global sensitivity analysis showed the relative importance of all parameters in a model. The analyses provided insight into the selection of actual optimal policies under various scenarios characterized by the model parameters and the need to explore alternative inputs, such as the congestion level in the network. The findings underscore the need for adequate sensitivity analysis for model calibration and evaluation. Based on the findings, the paper includes a guide for using sensitivity analysis in practice.