Scholarlink

<p>Accurate estimation of Annual Average Daily Traffic (AADT) on local roads is essential for transportation planning, infrastructure management, and safety analysis, yet traffic count data on these facilities are often limited. Existing AADT prediction models are typically specified as global relationships and may not adequately capture the spatially varying associations between traffic volumes and the built environment, accessibility, and road network characteristics across different geographic contexts. To address this, this study develops separate Geographically Weighted Regression (GWR) models for urban and rural local roads in Texas and compares them with traditional global regression. Candidate predictors derived from the Smart Location Database were combined with a segment-level measure of distance to the nearest non-local road, and the modeling set was selected using multicollinearity diagnostics. The results show that GWR improves model fit over global regression, with a larger gain for rural local roads than for urban local roads, and reduces prediction error across multiple metrics. Distance to the nearest non-local road emerges as the most important predictor in both contexts, indicating that proximity to the higher-class road network is strongly associated with local-road traffic. The local coefficient estimates vary in sign and magnitude across Texas, demonstrating that the relationships between AADT and its predictors are spatially contingent rather than fixed. Distinguishing between urban and rural contexts and explicitly accounting for spatial heterogeneity offers practical value for AADT estimation in data-sparse settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scalable prediction of local road traffic volume via macro-level indicators and spatial heterogeneity modeling

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

查看全文
  • Jinli Liu,
  • Subasish Das,
  • Swastika Barua,
  • Gaurab Chhetri

摘要

Accurate estimation of Annual Average Daily Traffic (AADT) on local roads is essential for transportation planning, infrastructure management, and safety analysis, yet traffic count data on these facilities are often limited. Existing AADT prediction models are typically specified as global relationships and may not adequately capture the spatially varying associations between traffic volumes and the built environment, accessibility, and road network characteristics across different geographic contexts. To address this, this study develops separate Geographically Weighted Regression (GWR) models for urban and rural local roads in Texas and compares them with traditional global regression. Candidate predictors derived from the Smart Location Database were combined with a segment-level measure of distance to the nearest non-local road, and the modeling set was selected using multicollinearity diagnostics. The results show that GWR improves model fit over global regression, with a larger gain for rural local roads than for urban local roads, and reduces prediction error across multiple metrics. Distance to the nearest non-local road emerges as the most important predictor in both contexts, indicating that proximity to the higher-class road network is strongly associated with local-road traffic. The local coefficient estimates vary in sign and magnitude across Texas, demonstrating that the relationships between AADT and its predictors are spatially contingent rather than fixed. Distinguishing between urban and rural contexts and explicitly accounting for spatial heterogeneity offers practical value for AADT estimation in data-sparse settings.