The success of analysis and simulation in transportation systems depends on the availability, quality, reliability, and consistency of real-world data and the methods for utilizing the data. Additional data and data requirements are needed to support advanced analysis and simulation strategies such as multi-resolution modeling (MRM) and multi-scenario analysis. This study has developed, demonstrated, and assessed a systematic approach for the use of data to support MRM and multi-scenario analysis. First, the study developed and examined approaches for selecting one or more representative days for the analysis, considering the variability in travel conditions throughout the year based on cluster analysis. Second, this study developed and analyzed methods for using crowdsourced data vii to estimate origin-destination demands and link-level volumes for use as part of an MRM with consideration of the modeling scenario(s).
The assessment of the methods to select the representative day(s) utilizes statistical measures, in addition to measures and visualization techniques that are specific to traffic operations. The results of the assessment indicate that the utilization of the K-means clustering algorithm with four clusters and spatio-temporal segregation of the variables demonstrated superior performance over other tested approaches, such as the use of the Gaussian Mixture clustering algorithm and the use of different segregation levels. The study assessed methods for the use of third-party crowdsourced data from StreetLight (SL) as part of the Origin-Destination Matrix Estimation (ODME), which identifies the method resulting in the closest origin-destination demands to the original seed matrices and real-world link counts. The results of the study indicate that Method 3(b) produced the best performance, which utilized combined data from demand forecasting models, crowdsourced data, and traffic counts. Additionally, this study examined regression models between crowdsourced data and count station data developed for link-level estimation of the volumes. This study also examined the accuracy and transferability of the link-level estimation of the volumes to determine if the crowdsourced data combined with available volume data at several locations can be used to predict missing or unavailable volumes in different locations on different days and times within the network. Regression models produced low errors than the default SL estimates when hourly or daily traffic volumes were taken into account. For similar traffic conditions, the models predicted directional traffic volume close to the real-world value.