Capturing variability in pavement performance models from sufficient time-series predictors: a case study of the New Brunswick road network

Amador-Jiménez, Luis Esteban; Mrawira, Donath · 2011 · Crossref

DOI: 10.1139/l10-127

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the limitations of deterministic pavement performance models, which fail to quantify uncertainty and often require partitioning data into homogeneous groups, leading to unreliable predictions for groups with sparse data. The authors propose a multi-level Bayesian regression approach to calibrate mechanistic model parameters from historical data while explicitly capturing reliability through confidence intervals. This method allows for the simultaneous generation of deterioration models for various pavement families, incorporates expert criteria, and handles missing data by borrowing strength across groups. The study utilizes a case study of the New Brunswick Department of Transportation road network, comprising 6,580 km of asphalt concrete and 9,664 km of chip seal roads. The model predicts International Roughness Index (IRI) progression using six years of observations (1991–1996). Key predictors include traffic loading (Equivalent Single Axle Loads), environmental factors (moisture index and freeze-thaw cycles), and pavement structural capacity. Due to unknown layer thicknesses, the modified structural number was replaced by the "AREA" deflection basin parameter derived from Dynaflect readings. The analysis involved 3,790 road segments. Missing data, particularly for the AREA parameter, were addressed using a semi-random approach within the WINBUGS software, assuming missing values followed a normal distribution based on observed trends. The mechanistic model form related IRI to age, traffic, environment, and strength, with coefficients treated as stochastic variables estimated via Markov Chain Monte Carlo simulation. Results indicate that the Bayesian model successfully corrected biased initial priors and provided robust posterior distributions for model coefficients. Sensitivity analysis revealed that in New Brunswick’s low-traffic context, environmental factors are the most significant drivers of deterioration. The model demonstrated that chip seal roads have higher initial as-built roughness and deteriorate faster than asphalt roads. Specifically, posterior estimates showed initial IRI values of approximately 1.17 for chip seal versus 1.06 for asphalt, with deterioration rates (alpha) around 266 for both, though chip seal roads exhibited higher variability. When nesting environmental zones within surface types, the model could not distinguish significant differences in deterioration rates due to limited data for chip seal roads in certain zones, illustrating the model’s ability to borrow information from larger groups to stabilize predictions. The significance of this work lies in demonstrating that multi-level Bayesian modeling can effectively calibrate mechanistic models to local conditions while quantifying uncertainty. This approach overcomes the drawbacks of classical regression by handling missing data without discarding observations and improving predictions for data-sparse groups. By providing probabilistic estimates rather than deterministic means, the method supports more robust transportation asset management decisions, allowing planners to analyze the impact of prediction variability on treatment programs and resource allocation.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• road surface friction