Exploring the potentials of automatically collected GPS data for travel behaviour analysis – A Swedish data source

Understanding the regularity and the variability of individual travel behaviour over time has been one of the key issues in travel behaviour research for three decades. A deeper insight into the long-term mobility patterns of persons and households has been so far restricted by the limited availability of longitudinal data, though. This paper presents an innovative approach to gain longitudinal travel behaviour data by means of Global Positioning Systems (GPS) and furthermore, describes the outline of post-data-processing work as well as possible analysis and modelling directions. 1. TEMPORAL ASPECTS OF TRAVEL BEHAVIOUR: CONTEXT AND DATA NEEDS Transportation research has so far mainly focused on the traditional crosssectional analysis of persons’ and households’ mobility patterns. The main interest has been on differences of travel behaviour between travellers or groups of travellers (inter-personal level) which can be covered easily by one-day or few-days travel diary data (e.g. KONTIV). Hence, mobility patterns observed on single days have been interpreted as the optimal decision of the traveller and as a state of behavioural equilibrium – which is assumed to be existent for any point of time and any situation. The investigation of the intra-personal level of travel (see Figure 1) which describes the variability of single persons’ and households’ mobility patterns over time has been restricted so far by the absence of survey data longer than one week and suitable methodology to treat such data. From a S. Schönfelder, K.W. Axhausen, N. Antille and M. Bierlaire scientific point of view but especially taking into account the planners’ legitimate requirements of data and results concerning the temporal aspects of travel, the available knowledge is insufficient. This led to the inadequate explanation of travel motives and determinants and therefore to problems with the implementation of transport policies. Fig. 1: Inter-personal and intra-personal level of travel behaviour Based on the experiences made with earlier approaches to collect longitudinal mobility data, the research project Mobidrive (1998-2000) – funded by the German Federal Ministry of Education and Research (BMBF) – was established to address the described research deficit and to update data availability (Axhausen, Zimmermann, Schönfelder, Rindsfüser and Haupt, 2002). With the implementation of a continuous six-week travel diary as core of Mobidrive (see Axhausen et al., 2002), a current data set of long-term individual travel behaviour is now available for analysis. The extensive investigation of the data during the last years has led to the development and the adoption of a range of analysis and modelling approaches for long-term travel (see Zimmermann et al., 2001 for a comprehensive overview). 2. THE POTENTIALS OF AUTOMATICALLY COLLECTED GPS DATA Collecting long-term mobility data on the person and household level remains a challenge for travel behaviour research. Despite the Mobidrive success in data collection, it is questionable if for bigger samples the same extent of intensive support of the respondents could be guaranteed. The Exploring the potentials of automatically collected GPS data for travel behaviour analysis potential drawbacks of travel surveys based on self-administered paper based survey designs are well known, and are especially worth considering for the observation of long-term mobility patterns. These include among others • the limited pool of respondents for long-term travel behaviour studies • item and unit non-response • reporting errors / inaccuracy (especially times/durations and distances) • fatigue effects in longitudinal surveys Besides, most travel survey data lack information for the route choice decisions of motorised individual transport which is a substantial pillar of travel demand modelling. This has raised the interest in longitudinal data bases – eventually covering even longer periods of time than in Mobidrive – without the high expenses for data collection, though. Possible approaches to meet this requirement are a) to reduce the respondents burden in ordinary paper based instruments by frequent activities elements (see Massot, Madre and Armoogum, 2000; Schlich and Schönfelder, 2001 for examples) or b) the use of Computer Assisted Data Collection (CADAC) methods which has been promoted extensively in the last years (see Leeuw and Nicholls, 1996). An even further reaching technical approach is the automatic collection of travel behaviour data by in-vehicle or on-person GPS applications. GPS provides specially coded satellite signals that can be processed in a GPS receiver, enabling the receiver to compute location, speed and time. Trip data generated by GPS is getting especially appealing for travel analysis if it is connected with GIS applications which offer digital mapping (e.g. including detailed land use information). The combination of the two methodologies promises to accurately track movements of vehicles or individuals which may supplement or even substitute ordinary travel diary survey designs. The technique offers data accuracy and comprehensiveness which would never be reached by ordinary paper surveys, especially in the detection of micro space-time details (short trips, speeds etc.), obtaining route choice behaviour information and the extension of the observation period (Lee-Gosselin, 2002). Normally, the positioning data is accurate to only few meters. S. Schönfelder, K.W. Axhausen, N. Antille and M. Bierlaire Figure 2 shows an example of a technical implementation of automated GPS data collection in a Dutch pilot study (Draijer, Kalfs and Perdok, 2000). Similar settings have been applied elsewhere. Here, the mobile data collection system consisted of a GPS receiver with RDS/FM correction, a data storage device / micro-computer (called GEOMATE) and mobile power supply. The equipment used in this study was portable, and had the total size of a video-camera, whereas in several other studies the instruments were installed permanently in cars. For each trip (irrespective of made by car, foot or public transport), the respondents switched on the system independently which started data transmission to the computer (GEOMATE) in intervals of 4 to 10 seconds. After data collection, the data was transferred from GEOMATE to a conventional PC for processing. Fig. 2: Automated GPS data collection: Example of technical set-up; adopted from: Draijer et al. (2000) 148 2.1 Recent experiences Since mid 1990s, GPS data loggers in connection with GI systems have been applied to data collection in several initial feasibility studies (see Table 1 RDS/FM: Radio Data System, FM: Frequency Modulation; RDS/FM signals correct the various inaccuracies in the GPS system which finally leads to extremely high spatial accuracy compared with conventional GPS. This concept is called Differential Global Positioning System (DGPS). Power supply (battery + charger) ON / OFF GEOMATE Data logging Coordinate transformation Trip purpose Mode Number of passengers Public transport line Day Time RDS / FM correction GPS Coordinate quality signal FMantenna GPSantenna Online datatransfer Offline datatransfer PC with software for Trip records GIS-applications Mobile equipment package Stationary postprocessing Exploring the potentials of automatically collected GPS data for travel behaviour analysis 1 for an overview). Lee-Gosselin (2002) summarises the existing studies by the categories • “Imitating” traditional travel diaries, • Passive monitoring and • Hybrid approaches. The first category may be defined as a mixed data collection approach: Travellers are equipped with a GPS device (mostly fitted in the person’s vehicle; for an exception see Draijer et al., 2000) and a hand-held computer (e.g. Personal Digital Assistant, PDA). The latter device is provided in order to substitute the ordinary travel-diary form by the digital/computer assisted input of further trip attributes. Hence, there are two base sources of information: (1) the GPS tracking data covering start and end times, position and speed recorded in short intervals (mainly 1 to 10 seconds) as well as travel distances and (2) activity purposes, car occupancies etc. reported by the travellers themselves via PDA input. Passive monitoring describes the collection of trip data by passive invehicle GPS systems without any further information input by the drivers (i.e. no driver-computer-interaction). The data collection of those studies is mainly traffic safety driven, i.e. the research focus lies on the investigation of the relationship between driving behaviour (speeds) and crash risks. Finally, hybrid approaches of passive monitoring plus interim or subsequent contacts with the respondents have been developed in the last few years. In a Canadian study which aims to obtain deeper insight into activity-scheduling processes of travellers, personal traces collected by invehicle GPS devices was made available to respondents by means of an inter-active GIS application (Doherty and Lee-Gosselin, 2000). The idea was to “help” the respondents – by visualising their own spatial movements patterns – to optimise their daily activity patterns and to eventually provide recommendations for a change of spatial behaviour. S. Schönfelder, K.W. Axhausen, N. Antille and M. Bierlaire Tab. 1: GPS and travel behaviour analysis: Selection of recent studies Study Technical details / collection procedure Main research objectives Lexington Area Travel Data Collection Test 1996 (Batelle, 1997) Passive recording plus interactive input by the drivers PDA); post-usage interviews (see above) Acceptance of using automatic collection devices; test of passive and interactive reporting; analysis of route choice information 1997-1998 Austin Household Survey (Pearson, 2001) Passive vehicle-based GPS recording; 200 vehicles; additional paper and pencil diary Feasibility; underreporting of trips in ordinary paper surveys; identification of trip ends in GPS data Transport Research Centre (AVV) experiment 1997, (Draijer et al., 2000); several cities in the Netherlands Mixed design: Passive GPS recording by mobile equipment plus paper-pencil diary as well as GPS/pencil diary only; total sample size: 150 Acceptance of survey methodology; test of mobile GPS devices (in hand-held computer); test of suitability for all travel modes Georgia Tech experiment 2000 (Wolf, 2000; Wolf, Guensler and Bachman, 2001); Atlanta / Georgia Passive in-vehicle GPS system plus paper trip diary for part of the sample; 30 respondents Possibility of total substitution of paper travel diaries; postdata processing issues SMARTRAQ / Drive Atlanta, start: 2002, Atlanta (Wolf, Guensler, Frank and Ogle, 2000; Sanders, 2002) Passive monitoring of about 1.100 vehicles, up to twoyears monitoring period plus paper travel diaries (1) Traffic safety and travel behaviour issues; (2) Physical activity of the respondents; (3) Air quality issues Summarising the outcomes of the recent feasibility studies, they prove the feasibility of automatic data collection for studying travel behaviour, but also technical and operational difficulties. The advantages of data collection by GPS are manifold (see e.g. Draijer et al., 2000; Wolf et al., 2001), such as • the reduction (or even elimination) of respondents’ burden • the availability of path choice information • the high level of spatial accuracy • the fact that data is generated in digital format which allows direct analysis. Exploring the potentials of automatically collected GPS data for travel behaviour analysis As potential drawbacks, the following aspects were identified: • the possibility of sporadic or even systematic technical problems of transmission, eventually leading to total loss of information (e.g. certain warm-up times before receiving signals) • costly post-processing of the GPS data, i.e. trip end, trip purpose and street address detection • still relatively high equipment costs • unlike most ordinary paper-pen surveys, no motives for travel are queried. Generally, all studies have in common that the GPS tracking is only one part of the overall survey structure. GPS monitored and therefore passively collected travel behaviour data needs to be framed by socio-demographic attributes of the travellers but especially by further information on trip purposes and the size of the company. Some methodological approaches to detect trip purposes – based on recent studies – are described in section 4. Future GPS applications will be possibly combined with more interactive techniques (such as the usage of PDAs to state trip purposes and car occupancies) to obtain a broader picture of the observed trips or travel patterns (Doherty, Nöel, Lee Gosselin, Sirois, Ueno and Theberge, 1999). The level of user interaction is believed to be an important issue for the development of future survey design incorporating GPS data collection elements. 3. THE RÄTT FART BORLÄNGE GPS DATA SET Based on contacts with transport psychologists from the universities of Dalarna and Uppsala (Sweden), the Institute of Transport Planning, Traffic, Highway and Railway Engineering (IVT) obtained access to the GPS data set Rätt Fart which promises to match nicely the research directions of Mobidrive.