Validation of Floating Car Data for Transport Policy Analysis and Transport Models



Validation of Floating Car Data for Transport Policy Analysis and Transport Models

Authors

Han Van Der Loop, KiM Netherlands Institute for Transport Policy Analysis, Marco Kouwenhoven, Significance, Peter Van Bekkum, MuConsult

Description

To validate Floating Car Data for model calibration and transport policy analysis, the floating car data of HERE and INRIX were compared with the detector loops data available for highways, regional and urban roads in the Netherlands from 2011-2016.

Abstract

Since 2010, floating car data for the Netherlands has been available from several companies: TomTom, INRIX, HERE (all large parts of the world) and BeMobile (Netherlands and Belgium). These data are usually used for navigation. Suppliers also publish lists of trends in congestion for urban regions, as based on FCD (TomTom, INRIX). It is unknown whether the quality of these data is good enough for use in transport policy analysis.

In a first study, trends in mean speeds per province per time of day of HERE data were compared with speeds calculated from detector loop data 2011-2014. It appeared that the speed level of HERE data increased significantly each year, while the speeds derived from detector loop data remained relatively stable. As HERE noted, these differences were caused by a change in the composition of vehicles in HERE’s data set, as well as by the method of speed calculation.

A second study is currently underway and compares INRIX data 2014-2016 for the Main Trunk Network, provincial and municipal roads in the Netherlands with available detector loop data. It appears that a time shift exists between the INRIX data and the detector loop data. The INRIX data react a few minutes later to changes in travel speed than does the detector loop data. The amount differs per road type. It also appears that INRIX data are correlated on a scale of one or a few minutes: INRIX’s algorithm produces an invisible kind of averaging or flattening of changes in travel speed, which questions the validity of these data at the level of minutes. Comparisons of the trends in time are not yet underway, but can be presented before summer 2017.

Given the fact that the amount of probes (e.g. cars equipped with navigation devices) increases each year (at present approximately 5% of all vehicles), the statistical reliability of FCD data has improved in recent years. A major advantage of FCD data compared to that of detector loops is that FCD data are available for all connecting roads of the network and provide indications for local and temporal decreases in traffic speed. These data can therefore be used to assess travel delay in the base year of a transport model and contribute to the calibration of travel delay. However, it remains uncertain whether these data will prove sufficient for identifying long term trends in speed. Also, the lists of congestion trends in urban regions published by FCD lack an explicit description of the calculation and validation. Moreover, they rely on a selective part of the vehicles, which also seems to change in composition over time.

In this paper we present a method to measure travel time delay by combining floating car data and detector loop data. Based on recent research, we present argumentation for how to use floating car data to identify other phenomena, such as the level of, and trends in, travel time reliability and the robustness of the road network. Finally, the possibilities for identifying the impact that policy measures (adding lanes and roads, traffic management, speed measures, financial bonus to avoid car use during peak hours) and other factors (socio-economic factors, flexible working) have on congestion will be discussed based on the empirical research conducted in recent years.

Publisher

Association for European Transport