Modeling AADT on local functionally classified roads using land use, road density, and nearest nonlocal road data
Introduction
A rapid increase in population, the growth in demand for travel, and the subsequent traffic congestion and road safety challenges all require better utilization of existing road infrastructure. A federally funded state-administered program known as the Highway Safety Improvement Program (HSIP) was instituted for state agencies to adopt a data-driven and performance-based strategic approach to improve safety on public roads. Such an approach typically involves collecting traffic data and estimating annual average daily traffic (AADT) for all functionally classified major, minor, and local roads. There are nearly 700,000 local road links (typically between two adjacent intersections; small lengths) in North Carolina, USA. However, local road traffic count stations are available for only ~1.5% of the local road links in North Carolina. The data limitations can be offset using robust methods that help estimate AADT for all the local road links in North Carolina.
Most of the current research methods help estimate AADT for higher functionally classified roads (major roads) due to the availability of traffic counts for these roads (either AADT or Annual Daily Traffic, ADT). A few researchers in the past explored statistical methods (Mohamad et al., 1998; Xia et al., 1999; Zhao and Chung, 2001; Apronti et al., 2016; Raja et al., 2018), geospatial methods (Zhao and Park, 2004; Wang and Kockelman, 2009; Selby and Kockelman, 2013; Duddu and Pulugurtha, 2013; Shamo et al., 2015; Kusam and Pulugurtha, 2016), and machine learning approaches (Sharma et al., 2000; Sharma et al., 2001; Sun and Das, 2015; Khan et al., 2018; Khan et al., 2019; Das and Tsapakis, 2020) to estimate AADT.
The predictability of a model depends on explanatory variables used for modeling. The local road travel characteristics are very different from other functionally classified road travel characteristics. In general, the local roads are designed for land access (AASHTO, 2001). Traffic volume on a local road is influenced by the amount of land being accessed and the type of land use along the local road. Hence, it is important to consider land-use types and their area in the modeling process.
Urban areas have a higher road density than the suburban and rural areas. The rural areas have dispersed developments with low traffic volume on roads. However, in the case of urban local roads, there will be higher mileage of roads as they serve denser land uses. The road density is defined as the mileage of roads within a pre-defined radial distance and is an indication of how heavily the area around the local road link is developed. It could have an influence on the local road AADT.
The local roads also sometimes support through traffic from other local roads. The higher functionally classified roads with higher AADT typically have a higher interaction with the local roads. Thus, the distance to the nearest higher functionally classified road and its AADT could have an influence on local road AADT.
Although a statistical method like ordinary least square (OLS) regression is relatively easier and provides quick estimates of AADT, it generally provides results at a global level. Additionally, the characteristics of a local road and demographic/socioeconomic characteristics, as well as land-use characteristics within its vicinity, vary over space and at a local level. Hence, it is envisaged that methods/models accounting for the spatial variability in dependent and explanatory variables may give reliable estimates of local road AADT in a study area. Methods like geographically weighted regression (GWR) allow different relationships to exist between AADT and other explanatory variables at different point spaces. In other words, using data at traffic count stations within the vicinity of a local road link to model and estimate local road AADT may yield better estimates than using data for all the traffic count stations in the county. However, a comprehensive comparison of OLS regression and GWR to estimate local road AADT accurately at the county level was not performed in the past. There is a need to research, compare, and assess the applicability of GWR for modeling and estimating local road AADT at the county-level.
A majority of past studies focused on models to estimate AADT and validated their findings. However, not many studies discussed estimating local road AADT at the non-covered locations in a study area and assessing their applicability for transportation planning. Estimating AADT at all the local road links and assessing based on local travel characteristics (estimated local road AADT is lower at locations that are far from a higher functionally classified road, higher local road AADT estimates at locations with high road density, etc.) is important and increases the confidence practitioners may have in the AADT estimates when developing transportation plans.
In spite of best efforts, it may not be feasible to estimate local road AADT accurately due to higher variation in traffic volumes and relatively lower sample size compared to higher functionally classified roads. The median prediction error could also vary by the functional class or area type and the speed limit. There is a need to research and explore possible reasons for higher prediction errors in local road AADT estimates. This will help enhance the local road count-based AADT sampling methodology.
Therefore, the overall contribution of this research is, thus, four-fold: 1) a comprehensive comparison of OLS regression and GWR methods to estimate local road AADT at county-level; 2) incorporating the influence of land use, road density, distance to the nearest nonlocal road, and AADT at the nearest nonlocal road on local road AADT; 3) estimating AADT at all the local road links in a county and assessing the quality of estimates based on local travel characteristics, and 4) an evaluation of mean prediction errors by county, functional class, speed limit, and the number of local road traffic count stations.
The methodological framework proposed in this research to estimate local road AADT not only helps transportation planners develop safety performance measures and compute local road vehicle miles traveled (VMT), but also assists with planning, proposing, and prioritizing infrastructure projects for future improvements and in air quality estimates.
Section snippets
Literature review
Researchers in the past have developed various methods and models to estimate AADT when traffic counts from the field are not available for a road link. An overview of AADT estimation methods is summarized in four different parts: statistical methods, geospatial methods, artificial neural network (ANN) and machine learning, and other methods.
Regression analysis was adopted by various researchers in the past due its interpretability. Seaver et al. (2000) estimated traffic volume on the rural
Methodology
Developing a statewide GWR model to estimate AADT at all the non-covered local roads in North Carolina was first considered as a part of this research. However, there are 26 counties in North Carolina without any land-use data. Additionally, there are 4744 unique land-use descriptions of parcels in the county-level, land-use databases. Therefore, ten selected counties in North Carolina were considered for modeling based on the quality of land-use data, population density, and the number of
Results
The number of local road traffic count stations available for modeling varied from a low of 55 in Mecklenburg County to a high of 295 in Wake County (Table 3). The descriptive statistics such as minimum, median, mean, maximum, and standard deviation of count-based local road AADT are summarized in Table 3. The mean and standard deviation of local road AADT is found to be higher at urban counties like Mecklenburg County and Wake County.
The descriptive statistics of the selected explanatory
Conclusions
Estimating and reporting AADT is important for planning, designing, building, and maintaining the road infrastructure. As local roads account for a major proportion of the road infrastructure in the United States, AADT is an important variable in the road safety analysis and improvement programs. Therefore, the focus of this research has been on developing county-level local road AADT estimates.
Ten counties were considered for modeling based on the quality of land-use data, population density,
Disclaimer
The contents of this paper reflect the views of the authors and not necessarily the views of the University of North Carolina at Charlotte (UNC Charlotte) or the North Carolina Department of Transportation (NCDOT). The authors are responsible for the facts and the accuracy of the data presented herein. The contents do not necessarily reflect the official views or policies of either UNC Charlotte, NCDOT or the Federal Highway Administration (FHWA) at the time of publication. This report does not
Acknowledgments
The paper is based on a research project conducted for the North Carolina Department of Transportation (NCDOT). Special thanks are extended to Kent L. Taylor, Behshad M. Norowzi, Jamie L. Viera, Stephen P. Piotrowski, William S. Culpepper, Brian G. Murphy, and Lisa E. Penny of NCDOT and Mike Bruff of Capital Area Metropolitan Planning Organization (CAMPO) for providing excellent support, guidance and valuable inputs for successful completion of the project. The authors also thank Venkata R.
References (47)
- et al.
Estimating traffic volume on Wyoming low volume roads using linear and logistic regression methods
J. Traffic Transp. Eng.
(2016) - et al.
AADT prediction using support vector regression with data-dependent parameters
Expert Syst. Appl.
(2009) - et al.
Interpretable machine learning approach in estimating traffic volume on low-volume roadways
Int. J. Transp. Sci. Technol.
(2020) - et al.
Development of statewide annual average daily traffic estimation model from short-term counts: a comparative study for South Carolina
Transp. Res. Rec.
(2018) - et al.
Examining the influence of network, land use, and demographic characteristics to estimate the number of bicycle-vehicle crashes on urban roads
IATSS Research
(2020) - et al.
Spatial prediction of traffic levels in unmeasured locations: applications of universal kriging and geographically weighted regression
J. Transp. Geogr.
(2013) - et al.
Spatial residual analysis of six modeling techniques
Ecol. Model.
(2005) A Policy on Geometric Design of Highways and Streets
(2001)- et al.
Geographically weighted regression: a method for exploring spatial nonstationarity
Geogr. Anal.
(1996) - et al.
The potential use of big vehicle GPS data for estimations of annual average daily traffic for unmeasured road segments
Transportation
(2019)
Using geographically weighted models to explore how crowdsourced landscape perceptions relate to landscape physical characteristics
Landsc. Urban Plan.
Transit ridership model based on geographically weighted regression
Transp. Res. Rec.
Applications of artificial intelligence paradigms to decision support in real-time traffic management
Transp. Res. Rec.
Relationship between transport accessibility and land value: local model approach with geographically weighted regression
Transp. Res. Rec.
Principle of demographic gravitation to estimate annual average daily traffic: comparison of statistical and neural network models
J. Transp. Eng.
ArcGIS Pro 2.5.2
Local forms of spatial analysis
Geogr. Anal.
Geographically weighted regression and multicollinearity: dispelling the myth
J. Geogr. Syst.
Geographically Weighted Regression: The Analysis of Spatially Varying Relationships
Exploiting correlations between link flows to improve estimation of average annual daily traffic on coverage count segments: methodology and numerical study
Transp. Res. Rec.
Estimation of Annual Average Daily Traffic on Road Segments: Network Centrality-Based Method for Metropolitan Areas
Improved AADT estimation by combining information in image- and ground-based traffic data
J. Transp. Eng.
Development and evaluation of recurrent neural network-based models for hourly traffic volume and annual average daily traffic prediction
Transp. Res. Rec.
Cited by (15)
Network-wide road crash risk screening: A new framework
2024, Accident Analysis and PreventionLocal modeling as a solution to the lack of stop-level ridership data
2023, Journal of Transport GeographyComparison of models with and without roadway features to estimate annual average daily traffic at non-coverage locations
2023, International Journal of Transportation Science and TechnologyCrash proximity and equivalent property damage calculation techniques: An investigation using a novel horizontal curve dataset
2022, Accident Analysis and PreventionCitation Excerpt :Yet, smaller buffer sizes may not include all of the associated crashes that are associated with the safety of the target location. Previous studies have analyzed the same data with different buffer sizes to investigate this predicament (e.g., Zhang et al., 2015; Kang et al., 2019; Hobday and Meuleners, 2018; Pulugurtha and Mathew, 2021). However, the method of investigating the impact of different buffer sizes using varying models creates two issues that are unaccounted for.