Clustering Pavement Roughness Based On the Impacts on Vehicle Em
Journal of Ergonomics

Journal of Ergonomics
Open Access

ISSN: 2165-7556

Research Article - (2016) Volume 0, Issue 0

Clustering Pavement Roughness Based On the Impacts on Vehicle Emissions and Public Health

Qing Li*, Fengxiang Qiao and Lei Yu
Innovative Transportation Research Institute, Texas Southern University, Houston, TX 77004, USA
*Corresponding Author: Qing Li, Innovative Transportation Research Institute, Texas Southern University, Houston, TX 77004, USA


Objective: This research intends to explore the correlation between pavement roughness and vehicle emissions, and classify the pavement roughness based on vehicle emissions and public health impacts.
Method: On-road tests were conducted to measure vehicle emissions by a Portable Emission Measurement System (PEMS), and collect the corresponding pavement roughness by a smartphone application. A total of 19,038 data pairs were collected during 325 km long test routes in the State of Texas. The correlation of the emissions and International Roughness Index (IRI) are analyzed and the roughness was classified into clusters by three pattern recognition algorithms.
 Findings: The pavement roughness could be classified into four categories based on the clustering features of emission factors. The average of Normalized Emission Factor (ANEF) started with 0.051 at a level of IRI between 0+ and 1.99 m/km (category A), then dropped to 0.032 with IRI between 1.99 and 3.21 m/km (category B), followed by a slight decline to 0.030 with IRI between 3.21 and 6 m/km (category C). When the IRI was greater than 6 m/km (category D), the ANEF increased to 0.039. Driving on the pavement categorized to C and D may induce higher invehicle noise and driving stress indicated by drivers’ higher heart rates. Conclusion: The relationship between pavement roughness and vehicle emissions is nonlinear. The even smoother (category A) and even rougher (category D) road surfaces may also induce higher vehicle emissions. The proposed categorization of pavement roughness for Texas incorporates the impacts on environment and public health. To minimize the ANEF, the roughness in categories B and C (IRI: 2-6 m/km) is optimal for pavement design. If the impacts on in-vehicle noises and drivers’ heart rates are concerned as supplemental factors, category B (IRI: 1.99-3.21 m/km) is the best. Switching a pavement from category A to B, up to 34% of vehicle emissions and fuel consumption could be achieved. This categorization can be used in the design, maintenance, and evaluation of highway pavements, as well as applied to other states and countries with further calibrations of clusters for local specific classification of roughness. 

Keywords: Human public health; Environmental impacts; Heart rates; In-vehicle noise; Vehicle emissions; Pavement roughness; Clustering algorithms


NOx: Nitrogen Oxide; CO2: Carbon Dioxide, *CO: carbon Monoxide. HC: Hydrogen Carbon


According to the surveys on road users at national and local level, pavement smoothness is identified as one of the main factors to evaluate national highways [1]. It has been believed that the smoother the pavement of highways is, the more comfortable that riders may feel. In practice, the pavement roughness is represented by an International Roughness Index (IRI), which is defined by Average Rectified Slope (ARS), a ratio of the accumulated suspension motion of a vehicle divided by the distance travelled by the vehicle [2]. The IRI has been introduced since 1986 [3]. The roughness indexes for different roadway surface are often classified into several categories, indicating the quality of pavement conditions. The U.S. Federal Highway Administration (FHWA) classified International Roughness Index (IRI) into three categories for interstate, non-interstate roads, and the state highways: good (4]. In 2002, about 98% of rural area interstates met the acceptable standards [5]. In Texas by 2014, 72.83% of interstate pavement is measured in good condition, 24.01% is acceptable, and 3.16 % is poor. The percentage of non-interstate National Highways System (NHS) decreases to 57.97% for good condition, 34.38% for acceptable and 7.65% for poor [6]. The Texas Department of Transportation classified the roughness into five categories based on different levels of smoothness and the construction specification for ride quality, respectively [7]. In the existing literatures, none of the current standard and classification of pavement roughness have considered the impacts on environment and public health. However, fuel consumption and vehicle emissions, which are associated with greenhouse gas (CO2)* and the formation of acid rain8 and toxic smog for humans (NOx)* [8], are possibly sensitive to pavement roughness [9-12]. Furthermore, hazardous levels (70 dBA above) [13] of in-vehicle noises to drivers were spotted while vehicles are driving on a rougher pavement [14]. Chronically, such noise levels may damage the sensitivity of hearing and raise drivers’ heart rates for intense attention to the control of the vehicle in a dynamic traffic. Consequently, drivers may suffer from serious physical and psychological health effects. In other words, vehicle emissions, invehicle sound levels, and human health are crucial factors and shall be taken into account in the classification of pavement roughness during the process of pavement design, construction, maintenance, and evaluation. With regard to this, this paper intends to explore the relationship between vehicle emissions (CO*, CO2, NOx, and HC*) and pavement roughness, and cluster the roughness into categories using suitable pattern recognition algorithms. The classification by the identified algorithms with the highest silhouette is tabulated, which is further evaluated with the incorporation of existing studies regarding the effects on drivers’ heart rates and in-vehicle noises.

Tests and Method

Test design and data process

A set of on-road tests was designed in various highways in three metropolitan areas in Texas: Houston, Austin, and El Paso, twenty tests in thirty days spanning from November 2014 to June 2015 during sunny days. The length of test routes was approximately 325 km, covering new pavements, older pavement, and work zones with rougher pavement. One test driver was recruited to drive a dedicated vehicle with a starting mileage of 16,496 km. A Portable Emission Measurement System (PEMS) was equipped inside the test vehicle to measure and record the second-by-second vehicle emission rates. To collect real-time calculated IRI simultaneously, a smart phone application was used, which is functioned by a plethora or phone models with very different sensors and extreme car models. The calculated IRI by the smartphone application is highly correlated to the roughness measured by Laser IRI surveys (R=0.77) [15]. The collected IRIs was aggregated afterwards into an evenly distributed data set with 20 meters as the interval. Before each test, forty-five minutes preconditioning was managed to allow PEMS to reach all the setpoints. The calculated IRI is based on the quarter-car simulation (QCS) for sampling during a driving trip [16]. A total of 19,038 data pairs were collected. Each includes IRI, geographic location, emission rates, speed, and engine information.

The recorded emission rates for the four measured emission indexes were also interpolated into a spatially distributed set of emission factors with a 20-meter interval. In order to eliminate the influence of un-matched magnitude ranges among different emission indexes, the interpolated emission factors were normalized so that all values are within [0, 1], and the Average of Normalized Emission Factor (ANEF) was proposed to combine emission factors of the four emission indexes. The normalization of a certain type of emission index i is calculated using Equation (1).


where, xi is the interpolated emission factor for emission index i, while xi, min, and xi,max are the maximum and minimum values of all emission factors for emission index i. The calculated image is the normalized emission factor for emission index i, ranging between 0 and 1. The ANEF is an average of normalized emission factor image for all emission indexes.

Clustering models identification

The process of classifying the IRIs based on their relevant ANEF is a typical clustering issue, the purpose of which is to group a set of objects so that objects in the same cluster are more similar to each other than to those in other clusters [17,18]. Three typical clustering models, including Two-step, Kohonen, and K-means, were selected to identify the collected data pairs in groups based on ANEF and IRI. The three models are based on different algorithms to cluster a large number of data quickly, and the accuracy for the same could vary. Thus, the silhouette of each clustering result is adopted to screen the optimal clustering models for this study.

With the Two-step clustering model, the collected data pairs were namely processed in two steps: being compressed into a manageable set of sub clusters, and merging the sub clusters into larger and larger clusters by a hierarchical clustering method [19]. The Kohonen clustering model is a type of neural network that can cluster the data pairs into distinct groups [20]. Data pairs will be fully trained first. The records with similar characteristics will be grouped together; otherwise, the records will appear far apart on the output map. The Kmeans clustering model clusters data pairs into distinct groups by calculating the nearest mean distance between records [21]. The records with the nearest distances were identified as a group. Whichever algorithm is employed, the clustering with the highest silhouette was selected. Silhouette is a method of interpretation and validation of consistency within clusters of data, which was proposed by Rousseeuw [22].

Results and Discussion

Algorithm identification

The clustering results suggested that, the pavement roughness could be grouped into four clusters by Two-step (silhouette=0.576), five clusters by k-means (silhouette=0.552), and twelve clusters (silhouette=0.31) by Kohonen. Twelve clusters by Kohonen may be too fine to apply in practice, therefore being screened out. Since the silhouette by Two-Step is higher than that by K-means, hence, the classification results by the algorithm Two-Step was adopt in this study.

Clustering calculated international roughness

The four clusters classified by the Two-step algorithm are shown in s 1(a). Between each adjunctive cluster, there is a boundary separating them, which forms four categories along the axis of ANEF, named as categories A to D. The ANEFs in category A is visibly more dispersive than those in other categories. More specifically, in category A, higher ANEFs (up to 0.6) are observed with the IRIs between 0+ and 1 m/km. When the IRI increases from 1 to 2 m/km, the distribution of ANEF become denser and are more concentrated below 0.10. In category B, most ANEFs are within 0.10. Category C is for IRI between 3.22 and 6 m/km, where the ANEFs were even lower with most values within 0.08. In category D, the IRI rises to more than 6 m/km, the ANEFs seem no longer to follow the decreasing trend as there are less points with very small ANEF.

To obtain an insight into the trend of the ANEFs’ distribution along IRI, an average of ANEFs for each bin of 0.5 m/km internal were calculated and plotted in Figure 1(b). In Figure 1(b), the trend line of ANEFs is highly correlated to the IRIs with a R-square of 0.70 (R=0.84), which shows that ANEF declines obviously with an increase in the pavement roughness for smaller IRI (like within category A in Figure 1, IRIFigure 1(a) (IRI=1.99 to 6 m/ km), the trend line in Figure 1(b) stays steadily, within the ANEF value of 0.04. When the IRI further increases to 6 m/km and above, the trend line rise gradually. In fact, IRI of 6 is identified by Road Roughness Measuring System (RRMS) as a threshold of unsatisfactory. Therefore, the classified four categories of IRI are with different features of ANEF, which are further summarized in Table 1.


Figure 1: (A) Three clusters of IRI based on ANEF and (B) Correlation between EI and IRI.

Category IRI ANEF In-Vehicle noise (dBA)1
Range Cluster Center Ave. Std Evaluation
A [0.00-1.99] 1.36 0.051 0.055 High 60-70
B [1.99-3.21] 2.54 0.032 0.017 Low 60-70

Table 1: Classification of IRI based on ANEF, In-vehicle noise, and heart rates. Note: 1The heart rates (HR) of a male and female driver were monitored during the period of driving on the pavement with a wide range of roughness. The two drivers’ heart rates were lower while driving on the pavement roughness (IRI

In Table 1, the ranges and cluster centers of IRI and ANEF are listed based on the classified categories A-D. In order to further compare the impacts of IRI on human health, the corresponding in-vehicle noises and heart rates (HR) based on the reported information in [23] are also tabulated in Table 1. The EIs in category A and D are higher than those in category B and C. Therefore if only emission reduction is concerned, the roughness of pavement is suggested to be within (1.99, 6.000. Nevertheless, based on the previous studies by [18] in-vehicle noise was estimated to be 60-70 dB (A) for lower IRIs (within categories A and B), and greater than 70 dB (A) for higher IRIs (within categories C and D). What’s more, drivers’ heart rates may increase while driving on the pavements with the IRIs fallen into category C and D. Hence, considering in-vehicle noise, drivers’ health, and ANEFs, category B is an optimal option for emission reduction and better ride quality. If only emission reduction is considered, both categories B and C are the best ones.

Improvement by designing the category of pavement roughness

Not to mention the reduction in maintenance and construction costs of category B pavement compared to the category A, a proper design of pavement roughness may significantly contribute to the vehicle emission reduction state-wide. Table 2 illustrates that there will be a great reduction in vehicle emissions and fuel consumption if the pavement roughness is switched from category A to B. The vehicle emissions driving on the pavement categorized to A are obviously higher than those categorized to B by 11% for CO2 and 34% for other vehicle emissions and fuel consumption. What’s more, a great portion of the 72.83% Interstate (5,203.7 km) and 57.97% Non-Intestate roads (22,609.2 km, IRI

Category IRI m/km CO mg/m CO2 g/m HC mg/m NOx mg/m FC g/m
A (0.00-1.99) 0.879 0.351 0.128 0.011 0.013
B (1.99-3.21) 0.581 0.314 0.085 0.007 0.009
Reduction 34% 11% 34% 34% 34%

Table 2: Emission and fuel consumption reduction by switching category A to B.


Based on the on-road collected vehicle emissions, the pavement roughness was classified into four categories A, B, C, and D, with the combination of Two-step clustering modelling and RRMS. Results show that the relationship between the pavement roughness and vehicle emissions is nonlinear. Smoother and rougher road surface may both induce higher vehicle emissions. Furthermore, the impact of the four categories on in-vehicle noises and human health indicated by the possible change in drivers’ heart rates were evaluated. When the pavement roughness is greater than 3.21 m/km (category C and D), drivers may be exposed to hazardous in-vehicle noise and suffer from driving stress. Therefore, vehicle emissions, in-vehicle noise, and the impacts on health effects are essential in the design and operation of the pavement roughness for public health and ride quality. The proposed classification of pavement roughness is able to enormously reduce vehicle emissions and improve the efficiency of fuel consumption from the highway system in Texas. The study is recommended to extend to nationwide pavement roughness for greater contribution to vehicle emission reduction and higher efficiency of fuel consumption.


The authors acknowledge that this research is supported in part by the Tier 1 University Transportation Center TranLIVE # DTRT12GUTC17/KLK900-SB-003, and the National Science Foundation (NSF) under grants #1137732. The authors appreciate the assistance in data collections by Wu Ying, Pengfei Liu, Mahreen Nabi, Mahbuba Khan, and the valuable comments from Mehdi Azimi.


  1. Michael W, Thomas DG, William DOP (1986) Guidelines for Conducting and Calibrating Road Roughness Measurements. World Bank Technical paper Number 46.
  2. Sayers MW, Karamihas SM (1998) Little Book of Profiling. University of Michigan Transportation Research Institute.
  3. Federal Highway Administration (FHWA) (2002) Pavement Smoothness Relationships.
  4. U.S. Department of Transportation Federal highway Administration (FHA) (2015) Highway History. Interstae System Conditions and Performance.
  5. Texas Department of Transportation (TxDOT) (2015) Preliminary MAP-21 Texas Transportation System Performance Results: Pavement Condition.
  6. Texas Department of Transportation Construction Division: Materials and Pavements Section (TxDOT) (2005) Condition of Texas Pavements. Pavement Management Information System (PMIS).
  7. Li Q, Qiao F, Yu L (2016) Impacts of pavement types on in-vehicle noise and human health. Air Waste ManagAssoc 66: 87-96.
  8. Kalembo C, Jeihani M, Saka A (2011) Evaluation of the Impact of Pavement Roughness on Vehicle Gas Emissions in Baltimore County. Department of transportation and Urban Infrastructures Studies.
  9. Oregon Department of Transportation (2014) The cost of Poor Highway Conditions to Oregon’s Economy.
  10. Chatti K, I Zaabar (2010) Estimating the Effects of Pavement Condition on Vehicle Operating Costs. NCHRP Report 720. National Cooperative Highway Research Program, Washington, D.C.
  11. Li Q, F Qiao, L Yu (2016) Calibrating Emission Factors for Highways Considering Pavement Roughness Information. Air Quality Measurement Methods and Technology, North Carolina.
  12. Li Q, F Qiao, L Yu (2015) Implications of Wireless Communication System for Traffic Operations on Vehicle Emissions. Air & Waste Management Association (AWMA), USA.
  13. Forslof L (2014) Automated Collection of Road Condition Data with Smartphones.
  14. Michael W Sayers, Thomas D Gillespie, Cesar AV Queiroz (1986) The international Road roughness Experiment: Establishing Correlation and a calibration standard for measurements. World Bank Technical paper, Washington DC.
  15. Bailey Ken (1994) Numerical Taxonomy and Cluster Analysis. Typologies and Taxonomies. SAGE PUBLICATIONS International Educational and Professional Publisher Thousand Oaks London, New Delhi
  16. Yang H, Qiao F (1998) Neural Network Approach to Classification of Traffic Flow States. The ASCE Journal of Transportation Engineering 124: 521-525.
  17. Kohonen T (1982) Self-Organized Formation of Topologically Correct Feature Maps. Biological Cybernetics 43: 59–69
  18. Hartigan J A, Wong M A (1979) Algorithm AS 136: A K-Means Clustering Algorithm. Journal of the Royal Statistical Society 28: 100–108.
  19. Rousseeuw P J (1987) Silhouettes: a Graphical Aid to the Interpretation and Validation of Cluster Analysis. Computational and Applied Mathematics 20: 53–65.
  20. Li Q, F Qiao, L Yu (2015) Testing of In-vehicle Noises when Driving on Freeways with Different Pavement Types. Air & Waste Management Association (AWMA), USA.
Citation: Li Q, Qiao F, Yu L (2016) Clustering Pavement Roughness Based On the Impacts on Vehicle Emissions and Public Health. J Ergonomics 6:146.

Copyright: © 2016 Li Q, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Top globaltechsummit