Skip to main content

Where do household E-cargo bikes go

Date

Project overview 

E-cargo bikes, for household use, are novel transport mode, but we know very little about their movement patterns. There is research about e-cargo bikes being used for logistics, but much less about domestic use. E-cargo bikes have strong potential to contribute to transport decarbonisation particularly in suburban areas where usage is growing and car dependence is high. However, there is a lack of research on both domestic e-cargo bike use and suburban use.   

Little is known about the movement patterns of household e-cargo bikes.  Finding out more about where e-cargo bikes go helps planners and policy makers (e.g.  those designing cycle infrastructure, or secure bike parking), and helps us to understand who might use e-cargo bikes, where and what benefits they might have 

The ELEVATE project loaned e-cargo bikes with GPS trackers to 46 suburban households in Leeds, Brighton and Oxford, to better understand the potential to reduce car use and transport energy demand.  Reducing car use is essential to meet carbon reduction targets.   

Data and methods  

The bikes travelled around 8000km in summer 2023.  We analysed the data from the bikes’ GPS trackers.  

Our objectives were to process the raw data to answer the following research questions which are partly developed from previous qualitative research:     

  • What road types do domestic e-cargo bikes use and avoid? - Qualitative interviews give us the hypothesis that some users do avoid main roads, but other users feel the extra size makes them own the space.   
  • Do domestic e-cargo bikes use traffic free cycle routes? - Some cycle paths have barriers that prevent e-cargo bikes, pushchairs, and adapted cycles from easily using them.
  • Do people on e-cargo bikes avoid hilly routes? – Hilliness is quite a barrier to conventional cycling, so it is useful to see if e-cargo bikes help reduce that barrier to cycling. 

The issues

We collected GPS data using POW Unity trackers.  They are principally designed for security (locating a bike if stolen) rather than transport data science analyses.  The tracker data receives basic pre-processing through the API. Preliminary investigation of the data found errors in the raw data. We determined then conducted precise pre-processing and cleaning to ensure the accuracy in the dataset. The issues causing error are as follows: (a) Phantom Bikes: These erroneous GPS points were recorded while the bike was moving . (b) Time to first fix: This error in recording occurs when the GPS is first switched on as the device takes up to a few minutes to get an accurate position (also known as “Time to First Fix errors”). (c) Motion Map Gap: This error occurred where the GPS device didn’t record the route information for  part of the route whilst  the bike was on the move. (d) Speedy Bikes: Some of the points  had speeds higher than was realistic for an e-cargo bike. This may have occurred when the bike was either being transported on public transport or in a car. The following methods were adopted to remove the erroneous points. 

The phantom bikes points were removed based on specific thresholds related to rolling average of distance between four consecutive points, speed to the next point and the time difference with the next point. The thresholds were determined based on other literature ​(Jiménez-Meza et al., 2013; Mohamed and Bigazzi, 2019; Koh et al., 2022; Maurer et al., 2024; Smith et al., 2024)​ and domain knowledge.  

To remove the points with time to first fix errors, we adopted a nearest neighbour algorithm based on a time-distance matrix. For ensuring further accuracy, we ran a further check based on the time difference and distance to the next point.  

The motion map gap occurred where route points were not recorded.   Users tend to follow a similar route for a round trip based on established habits or preferences for certain routes due to familiarity Prato ​(2009)​. If such round trips were present, the missing points were replaced with points recorded in another trip leg of the round trip.  

To address GPS points related to speedy bikes errors, we used a speed threshold of 60km/hr. Any points that recorded speed above the threshold speed were removed during the cleaning. 

We also had to group points into trips.  We experimented using unsupervised clustering but found that using the experiential knowledge of the Elevate research team was a more effective way of setting thresholds such as the minimum allowable stationary time to denote a stop between different trips, and a minimum trip distance so we could remove points where a bike was simply being moved from the garden to the garage.   

Key findings  

 The errors observed in the real data were generally a combination of two or three of the categories mentioned above. The total number of recorded GPS points in the original dataset was 238,351. 9896 points were removed during cleaning. We also imputed 790 points which were missing because of bad GPS signal. 95% of the GPS points were retained after the cleaning and processing. The most common error was “phantom bikes”. Table 1 shows the number of points that were concerned by each error cases. 

Item

% of total GPS points 
Duplicate points  1% 
Phantom bikes  2.5% 
Time to first fix  0.6% 
Motion map gap  0.3% 
Speedy bike  0.4% 

Value of the research  

This case study contributes to understanding the movement patterns and route preferences for domestic e-cargo bike users.  These preferences may differ to those of people on ordinary bicycles or e-bikes.  Because this data science project sits within a large Mixed methods research project then this work will contribute to a set of synthesised project findings that draw on quantitative and qualitative work.  This data science project contributed to:  Developing a methodology to clean data collected through GPS trackers (here specifically POW unity trackers for use on e-bikes), Contributed to a reuseable codebase which could be used by other organisations with POW trackers and more specifically, provide a clean GPS dataset that will inform research into the distribution of trip lengths, the extent to which e-cargo bikes avoid hills or not and inform what road and infrastructure types e-cargo bike users prefer.   

Insights 

  • What is novel in the research? Developing a comprehensive methodology to clean GPS tracker data that can be applied to e-cargo bike research.   
  • What have you added to the research area? The code developed contributes to a better overall understanding of the usage of domestic e-cargo bikes within a larger mixed methods peoject 
  • What have you learned? There are a number of error types which can be dealt with in the data cleaning. Contextual knowledge helped to set parameters and thresholds used to determine when a point was accurate.   
  • How might this be taken forward? The code developed will enable the research team to determine statistical measures on the trips undertaken by the participants. Moreover, it will also reveal the movement patterns of the e-cargo bikes. 

Research theme  

  • Societies 
  • Environment 

Programme theme  

  • The Science of Data Science 
  • Mathematical and Computational Foundations 

Team 

  • Dr Jayita Chakraborty, Data Scientist, LIDA 
  • Dr Ian Philips, Associate Professor Institute for Transport Studies, University of Leeds 
  • Dr Alice De Sejournet, Research Fellow Institute for Transport Studies, University of Leeds 

Partners

  • University of Brighton
  • University of Oxford
  • Technical University of Eindhoven
  • Technical University of Dortmund

Funder 

ELEVATE  is funded by UKRI  (UKRI EP/S030700/1) 

This work has been facilitated by the Leeds Institute for Data Analytics (LIDA) Data Scientist Development Programme, which employs early-career data scientists to deliver real-world data-driven impact in the interests of the public good. 

References 

  • ​​Jiménez-Meza, A., Arámburo-Lizárraga, J. and de la Fuente, E. 2013. Framework for Estimating Travel Time, Distance, Speed, and Street Segment Level of Service (LOS), based on GPS Data. Procedia Technology. 7, pp.61–70. 
  • ​Koh, Z., Zhou, Y., Pik Lik Lau, B., Liu, R., Hua Chong, K. and Yuen, C. 2022. Clustering and Analysis of GPS Trajectory Data using Distance-based Features. IEEE Access. 10, pp.125387–125399. 
  • ​Maurer, L., Meister, A. and Axhausen, K.W. 2024. GPS-based speed profiles for cyclists in Zurich, Switzerland [Online]. Available from: https://doi.org/10.3929/ethz-b-000689090. 
  • ​Mohamed, A. and Bigazzi, A. 2019. Speed and road grade dynamics of urban trips on electric and conventional bicycles. Transportmetrica B. 7(1), pp.1467–1480. 
  • ​Philips I, Azzouz L, de Séjournet A, Anable J, Behrendt F, Cairns S, Cass N, Darking M, Glachant C, Heinen E, Marks N, Nelson T, Brand C. 2024. Domestic Use of E-Cargo Bikes and Other E-Micromobility: Protocol for a Multi-Centre, Mixed Methods Study. International Journal of Environmental Research and Public Health. 21.12
  • ​Philips I, where do e-cargo bikes go , Royal Geographical Society Annual International Conference, London, August 2024 
  • ​Prato, C.G. 2009. Route choice modeling: past, present and future research directions. Journal of Choice Modelling. 2(1), pp.65–100. 
  • ​Smith, H., Akhtar, S., Caulfield, B. and O’Mahony, M. 2024. Validity of GPS data in driving cycles. IET Intelligent Transport Systems.