Method to Reduce Uncertainty in the Prediction of Pavement Condition With a Lower Sample Frequency

Abstract

In this paper, the effect of missing pavement condition observations in the predictions of the future state of a road network was evaluated. Real data from North Carolina were used for this purpose. First, the auto-regression method was compared against the most common “family-curve” modeling approach. It was found that the auto-regression method improves the predictive accuracy of predictions, at both project and network levels. By using the auto-regression method over the “family-curve” approach it is possible to reduce, on average, the Mean Absolute Percent Error of the predictions by 40%. Second, this paper evaluates the case in which a reduced survey frequency is unavoidable, and state highway agencies might need to plan the network maintenance based on historical observations and the subsample of the current condition. Observations of the Pavement Condition Rating for years 2013–2019 were used to define four different scenarios of reduced survey frequency: Scenario 1—“business-as-usual,” where the entire network is surveyed every year; Scenario 2—“reduced-sampling,” analyzed the case where the entire network is surveyed every other year; Scenario 3—“halfway-sampling,” evaluates the case where only half of the network is surveyed every year; and Scenario 4—“least-sampling,” considers the case where only a third of the network is monitored every year. Scenario 1 was used as the baseline of comparison, and as expected it was found that whenever possible the network should be monitored annually; however, if that is not feasible the best option, from the ones evaluated in the paper, should be Scenario 3.