MODELING ANNUAL AVERAGE DAILY TRAFFIC FOR LOCAL FUNCTIONALLY CLASSIFIED ROADS
Analytics
54 views ◎18 downloads ⇓
Abstract
The rapid increase in population, the growth in demand for travel, and the subsequent traffic congestion and road safety challenges call for better utilization of existing road infrastructure. A federally funded state-administered program known as Highway Safety Implementation Program (HSIP) was instituted for state agencies to adopt a data-driven and performance-based approach to improving safety on public roads. One of the requirements of HSIP is for state agencies to report annual average daily traffic (AADT) for all functionally classified major, minor, and local roads.A considerable amount of resources are spent by various transportation departments to estimate AADT on major, minor, and local road links. The available AADT data are based on traffic counts collected at selected locations on these roads. However, time, money and other resource constraints limit agencies from estimating AADT for all the roads in the transportation network. The count-based AADT is available for all major and minor road links, but available for a relatively fewer number of local road links.The objectives of this research are: 1) to review AADT estimation methods for functionally classified major and local roads, 2) to examine the influence of road network, socioeconomic, demographic, and land use characteristics on local roads AADT, 3) to develop sustainable and repeatable methods to estimate AADT on local functionally classified roads, and 4) to validate and calibrate the models to improve their predictability. To achieve the aforementioned objectives, this research examined five different modeling approaches to estimate AADT for all local roads. They include traditional ordinary least square (OLS) regression, geographically weighted regression (GWR), and geospatial interpolation techniques such as Kriging, inverse distance weighted (IDW) interpolation, and natural neighbor interpolation. The available count-based AADT data at 12,899 traffic count locations on local roads in North Carolina during the years 2014, 2015, and 2016 was used as the dependent variable when developing the models. The road, socioeconomic, demographic, and land use characteristics for the year 2015 were considered as the explanatory variables. The explanatory variables were screened to minimize multicollinearity by computing and comparing the Pearson correlation coefficients.The model development was carried out in two levels: the statewide AADT estimation and county-level AADT estimation. The speed limit, road density, distance to the nearest nonlocal road, the count-based AADT at the nearest nonlocal road, and population density are significant explanatory variables used to develop the statewide models. The validation results indicated that the GWR model performed relatively better when compared to other considered statistical and geospatial methods. GWR can accommodate the spatial variations in AADT data, by geographic location, when estimating the local road AADT. The errors in estimated local road AADT are lower for locations with a higher number of nearby traffic count stations.Ten counties were considered for county-level analysis and modeling. The quality of land use data, population density, road density, and the number of local road traffic count stations available in the county were used in the selection process. The county-level models were observed to estimate local road AADT relatively better than the statewide models. The inclusion of land use variables for modeling can be mainly attributed to the improved performance of county-level models. The developed county-level models were used for estimating AADT at non-covered locations in each selected county.The median prediction errors associated with statewide and county-level models were compared and assessed to recommend future sampling requirements to improve the model predictability. The median prediction errors are higher for urban local roads and for local roads with a speed limit greater than 25 mph and less than 50 mph. In most of the cases, the median prediction error seems to depend on the number of available local road traffic count stations and county characteristics. These findings indicate that count-based local road AADT data from spatially distributed traffic count stations in North Carolina can improve the predictability of models.The prediction errors were also low at local road traffic count stations near single-family residential units, multi-family residential units, and the commercial area. Contrarily, they are relatively higher at local road traffic count stations near schools, institutions, government, office, and industrial land uses. This could be attributed to differences in the number of local road traffic count stations by land use area type (more the number of local road traffic count stations, lower the prediction error).Samples sizes were estimated based on the coefficient of variation in the available count-based local road AADT data and the number of local road links by the speed limit and link connectivity for each county at a 70% confidence level. A 15% prediction error rate was considered acceptable for local roads and used to estimate the sample sizes. A sampling plan based on the number of local road locations, functional classification type, speed limit ranges, and road connectivity type like dead-ends is recommended. To expand the local road traffic data collection program and estimate spatially distributed count-based local road AADT, sample data must be collected at around 12,000 (based on the speed limit) to 22,000 (based on the link connectivity type) different stations in North Carolina biennially. The simple random sampling criterion can be used when selecting locations based on the speed limit and link connectivity, in a county, while ensuring that they are geographically distributed in the county. This research proposes the use of county-level growth factors based on available count-based local road AADT for future AADT estimations. The count-based local road AADT and growth factor for the reporting year, for the county in which the local road is located, must be used if the count-based AADT was available for the previous year(s). For non-covered locations, the estimated AADT for the base year (2015 in this research) and growth factors from the base year to the reporting year must be used.It is recommended to update the base year local road AADT estimation model to 2020 once the statewide travel demand model is updated or census 2020 data (block-level) is available. Overall, the application of the proposed AADT estimation method and growth factors minimize the costs associated with lapses in traffic count data collection programs and plans. The estimated or actual AADT for each local road link can be used to compute the VMT for each local road link. The findings from this research can be used to proactively identify solutions and plan, design, build, and maintain the local roads.