Article Preview
Top1. Introduction
New technologies have undoubtedly changed the way transport engineers understand the underlying dynamics of a transport network and therefore their role in planning, monitoring, predicting traffic and consequently influencing mobility decision making. Traditionally, transport planners were using information collected through dedicated measurement equipment such as loop detectors, cameras, passive infrared sensors and radars which are point detectors providing traffic data at installation points (limited coverage) (Antoniou et al., 2011; Klein et al., 2006). More recently, Automatic Vehicle Identification technologies, Bluetooth Point to Point detectors and area wide sensors - Automatic Vehicle Location technologies integrated with the global positioning system and geographical information systems have paved the way for the new era in transport planning, traffic monitoring and management (Herrera et al., 2010; Nanthawichit et al., 2003; Taylor et al., 2000). Floating car data (FCD) belong in the very last category referring to cars (either regular or dedicated fleets) acting as constant traffic sensors providing spatiotemporal information of their movement (coordinates, speed, timestamp – mapmatched data).
GPS data are linked to valuable spatiotemporal information giving a deep insight for a series of traffic parameters such as trajectories, travel time, time searching for parking, times stopped at signals, speed and can reveal driving behaviour related issues (Taylor et al., 2000). The datasets collected from floating vehicles (GPS-enabled) can, under conditions – i.e. sampling, considered as representative for the overall traffic and simultaneously as highly cost – effective. Given these advantages, the use of floating car data (FCD) has become increasingly popular in recent years and has triggered long discussions among the academic society for Big Traffic data, data fusion, transport modelling, prediction of spatiotemporal traffic phenomena and short-term traffic forecasting (Bhaskar et al., 2011; Dion, Oh, & Robinson, 2011; Dion, Robinson, & Oh, 2011; El Faouzi et al., 2011; Fries et al., 2012; Jintanakul et al., 2009; Ma et al., 2012; Oh et al., 2005; Van Lint & Hoogendoorn, 2010; Vlahogianni et al., 2014). Many studies have, however, based their results for mining traffic patterns and understanding network performance on special vehicles (e.g. trucks, taxi, buses) data (Bertini & Tantiyanugulchai, 2004; Moore et al., 2001; Schwarzenegger et al., 2008; Simroth & Zähle, 2011) opening the debate on coverage, penetration levels and driving behaviour influencing factors (i.e. can taxi data be representative for the whole traffic at an urban network?).
Many researchers have posed the issue of penetration rate for the ability of FCD to provide reliable traffic information; Dai et al. (2013) support that for a freeway, a penetration level of around 3% can be considered as acceptable for reaching reliable results, a percentage which increases however to 5% when examining ‘surface roads’(Dai et al., 2003). According to Cheu et al. (2002), 4% - 5% of active probe vehicles or at least ten probe vehicles crossing a specific link within the sampling period represent good statistics (Cheu et al., 2002), Hong et all (2007) supported that 2% probe penetration can guarantee information integrity, Chen and Chien (2000) supported that 3% of probe vehicles would be sufficient for reaching 95% significance level (Chen & Chien, 2001; Hong et al., 2007) and numerous of other researchers have also proposed around 4-6% penetration. It is also worth noting that additional potential limitations caused by different traffic conditions (variations) or traffic restrictions (i.e. congested or free roads, dedicated lanes for PuT or taxi) affecting roads’ capacity should also be taken into account.