Using sparse operational e-bike GPS data for corridor-level transport analysis: an uncertainty-aware map-matching framework
This paper presents an uncertainty-aware map-matching framework that evaluates the reliability of sparse e-bike GPS data for corridor-level transport analysis by classifying trajectory segments into support levels based on multiple diagnostic checks, thereby enabling planners to distinguish between robust route reconstructions and uncertain ones.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Every day, millions of people move through cities on bicycles and electric bikes, weaving through streets, parks, and shortcuts. For decades, city planners have tried to understand these movements to build better roads and safer paths, but they have mostly relied on fixed counters or surveys that capture only a small slice of the whole picture. In recent years, the GPS trackers built into shared rental bikes and subscription services have offered a new way to see these patterns in real time. However, these digital footprints are often incomplete. Because the devices do not record a position every second, but rather every few minutes, the data appears as a series of scattered dots rather than a continuous line. The space between those dots is a mystery: a rider could have taken a direct road, a winding park path, or a quiet residential street, and the dots alone cannot say which one it was. This gap between what is recorded and what actually happened creates a significant challenge for anyone trying to use this data to make decisions about infrastructure or safety.
A team of researchers at the University of Lausanne in Switzerland set out to solve this problem of missing information using data from a local electric bike rental service in the Canton of Neuchâtel. They faced a situation where the GPS points were so far apart that simply connecting the dots with a straight line or forcing the data to snap to the nearest road would create a false sense of certainty. Instead of trying to guess the single correct path for every ride, the researchers developed a new way to measure how much confidence planners should have in any reconstructed route. They treated every possible path between two GPS points not as a fact, but as a hypothesis, and then tested these hypotheses against each other to see if they agreed on the general direction of travel.
To do this, the researchers took 2,618 segments of bike trips, which contained over 11,000 individual GPS observations, and ran them through six different computer methods designed to figure out where a vehicle has been. These methods ranged from standard tools used by major navigation companies to specialized algorithms built specifically for bicycles. Each method produced its own version of the route, creating a set of competing stories for how the rider moved between the recorded points. The researchers then compared these stories to see if they converged on the same general corridor. If six different methods all suggested the rider took the same main road, the evidence was considered strong. If the methods disagreed, suggesting one took a highway while another took a side street, the evidence was flagged as uncertain.
The study found that while the computer programs could almost always produce a route for nearly every segment of the data, the quality of that route varied wildly. In fact, simply getting a result from the computer did not mean the route was reliable. When the researchers applied their new uncertainty-aware framework, they discovered that only about 56 percent of the segments had strong support, meaning multiple methods agreed on a clear path. Another 28 percent had moderate support, offering a usable but less certain picture. The remaining segments were either too ambiguous to trust or contained errors that made them impossible to interpret. This result challenged the common assumption that a successfully matched GPS track is automatically accurate; the researchers showed that for sparse data, a matched route can look complete while hiding a great deal of uncertainty about where the rider actually went.
Crucially, the researchers also found that the uncertainty was not just a minor technical glitch but a fundamental feature of the data. Even when the GPS points were relatively close together, the time between them was often long enough that a rider could have traveled a significant distance without being observed. The study revealed that many of the "routes" generated by standard methods were actually just guesses filling in large gaps. By grouping the results into categories of strong, moderate, weak, and unresolved, the team provided a practical tool for city planners. This tool allows them to keep the useful information—such as which general corridors are popular—while filtering out the noisy, unreliable data that could lead to poor planning decisions.
The findings suggest that electric bike data is a valuable resource for understanding how people move, but only if the limitations of that data are respected. The researchers demonstrated that it is possible to use these sparse records to identify broad trends in cycling activity without pretending to know the exact path of every single rider. Their approach offers a way to separate the clear signals from the noise, ensuring that when cities use this data to decide where to build a new bike lane or where to improve safety, they are acting on evidence that is as solid as the data allows. The study concludes that the value of this technology lies not in its ability to reconstruct every journey perfectly, but in its ability to tell us when we know enough to act and when we should remain cautious.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.