Modeling Dust Generation on Low-Volume Roads Based on Vehicle Speed and Surface Fines Content

Abstract

This study analyzes the role of vehicle speed and surface fines content on dust emission. Accordingly, fifty unpaved road sections in Iowa were evaluated; surface loose-aggregate samples were collected, and dust was collected using a Colorado State Dustometer at three speeds: 25 mph, 40 mph, and 55 mph. The data were analyzed using analysis of variance (ANOVA) test. Several dust-prediction models were developed utilizing multiple linear regression (ML), nonlinear regression with an interaction term (NLI), nonlinear beta regression (NLB), nonlinear curve-fitting regression (NLCF), and a multilayer neural network (MNN). The model predictors included vehicle speed and surface fines content. When models were evaluated using synthetic data and compared using post-hoc analysis, it was found that dust increases exponentially as vehicle speed increases and increases linearly as surface fines content increases. Also, at higher speeds, dust values will converge independently of the fines content in the surface materials. The ANOVA test results revealed that vehicle speed, surface fines content, and their interaction significantly affected dust emissions. The accuracy of models ranged from acceptable to good. The coefficients of determination (R²) for ML, NLI, NLB, NLCF, and MNN_Training models were 0.703, 0.718, 0.689, 0.696, and 0.776, respectively. Evaluation of the models showed that independent of the R² value, the MNN model was the most accurate in predicting dust emissions, followed by the NLCF model, the ML model, the NLB model, and lastly the NLI model. The post-hoc test showed that MNN_Training, NLCF, and ML models produced comparable results.