Prediction of pavement distress index with limited data on causal factors: An auto-regression approach

Prediction of the pavement distress index (DI) for different road sections on a network scale over an extended time horizon is necessary for efficient maintenance planning and resource allocation. In many cases, however, sufficient information on causal factors of pavement distress is unavailable. Nonetheless, models with sufficient reliability have to be developed in these situations. It is shown in this research that with suitable statistical models and application, sufficient information on missing causal factors may be accounted for indirectly by incorporating preceding observed DI values as an independent variable when predicting future DI values. Among different statistical modeling approaches, autoregression models with recursive applications were shown to produce reliable DI predictions. Extended future DI values were projected recursively by first predicting DI values for the immediate future interval based on current pavement age and DI value, and then using the new predicted DI value to predict the immediate value following that, and so on. By repeating this process, DI values as far into the future as necessary were developed. Results show that current and past DI values can capture the impact of many of the causal factors and hence may be used to produce fairly reliable DI projections. The resulting predicted DI time paths generated were not S-shaped as most of the literature suggests. This, however, is operationally inconsequential since rehabilitation actions are often triggered by relatively low DI thresholds for which the simplification here is sufficient.