A System-Level Gravel Loss Prediction Model

Granular roadways deteriorate over time due to both external factors (repeated roadway traffic loads and precipitation) and internal factors (weak subgrade soils, poor drainage, and low quality gravel/rock materials). As a result of these factors, granular roadways experience a significant amount of gravel loss during their lifespan. This is one of the main engineering challenges associated with maintaining such roads. Therefore, it is very important for engineers to be able predict annual gravel loss, to enable more reliable predictions for maintenance budgets. A gravel loss prediction model could help predict the amount of roadway surface materials expected to be lost after a given time, and therefore, to a greater extent, indicate to engineers the expected amount of materials to be purchased for the upcoming fiscal year. This tool can play a vital role in making proactive budgetary decisions for roadway maintenance. Existing gravel loss prediction models have mainly been developed to study the roadway material declination patterns on one or more segments in a roadway system. However, section-based gravel loss prediction models may not always be suitable to make network-level material declination predictions, because these models often require site-specific information which may lead to biased results caused by local calibration. In this study, granular road condition and performance data were collected from Iowa counties, and historical information on factors impacting roadway deterioration were obtained from federal government databases. Historical trends of daily traffic and precipitation variations, along with gravel/rock material quality test reports for quarries and soil test reports for counties, were analyzed to determine county-level representative values for each of those factors that contribute to granular road deterioration. Then the system-level roadway gravel loss was modelled using data from a full factorial experimental design. Multiple linear regression (MLR) analysis was also conducted considering county-level averages of daily traffic, precipitation intensity, gravel/rock material quality, and subgrade soil plasticity index as influencing variables. Out-of-sample data prediction and accuracy of fitted models were assessed using cross validation. The developed model can act as a standalone program because site specific data is not required to run the model and therefore, it provides the local agencies with a cost efficient annual gravel loss approximation tool.