APPLYING MULTIVARIATE GEOSTATISTICS FOR TRANSIT RIDERSHIP MODELING AT THE BUS STOP LEVEL

Abstract Travel demand models have been developed and refined over the years to consider a characteristic normally found in travel data: spatial autocorrelation. Another important feature of travel demand data is its multivariate nature. However, regarding the public transportation demand, there is a lack of multivariate spatial models that consider the scarce nature of travel data, which generally are expensive to collect, and also need an appropriate level of detail. Thus, the main aim of this study was to estimate the Boarding variable along a bus line from the city of São Paulo - Brazil, by means of a multivariate geostatistical modeling at the bus stop level. As specific objectives, a comparative analysis conducted by applying Universal Kriging, Ordinary Kriging and Ordinary Least Squares Regression for the same travel demand variable was proposed. From goodness-of-fit measures, the results indicated that Geostatistics is a competitive tool comparing to classical modeling, emphasizing the multivariate interpolator Universal Kriging. Therefore, three main contributions can be highlighted: (1) the methodological advance of using a multivariate geostatistical approach, at the bus stop level, on public transportation demand modeling; (2) the benefits provided by the models regarding the land use and bus network planning; and (3) resource savings of field surveys for collecting travel data.

摘要 出行需求模型历经多年发展与优化，旨在纳入出行数据中普遍存在的空间自相关性（spatial autocorrelation）特征。出行需求数据的另一项重要属性为其多变量特性。然而针对公共交通出行需求领域，目前仍缺乏能够兼顾出行数据稀缺性（通常采集成本高昂）与必要细节粒度的多变量空间模型。因此本研究的核心目标为：基于巴西圣保罗市某公交线路的站点数据，通过多变量地统计建模方法，对站点上车量（Boarding）变量进行估算。作为具体研究目标，本研究针对同一出行需求变量，采用泛克里金（Universal Kriging）、普通克里金（Ordinary Kriging）以及普通最小二乘回归（Ordinary Least Squares Regression）三种方法开展对比分析。基于拟合优度指标的分析结果显示，地统计学（Geostatistics）相较于经典建模方法是一种极具竞争力的工具，其中多变量插值器泛克里金的表现尤为突出。综上，本研究可总结出三项核心贡献：（1）在公交站点层面将多变量地统计方法应用于公共交通需求建模，实现了方法论层面的创新；（2）所构建的模型可为土地利用与公交线网规划提供决策支撑；（3）节省了采集出行数据所需的实地调研资源。