MODAP: A Multi-City Open Data & Analytics Platform for Micromobility Research

The data reported through this work were accessed from three dockless micromobility operators, namely Tier, Lime, and Flamingo. Each of these operators runs dockless fleets of vehicles meaning that users can start or end a trip in any public space and the vehicles are not parked at dedicated docking stations. Data from only one operator per city were accessed. Table 1 provides an overview of the final micromobility data sets including the operators, number of trips, type of vehicle(s), and time period of data collection. These trip count values are from after the data has been cleaned.

The data were accessed via public-facing application programming interfaces (API). With every request, these APIs (Table 2) return a set of all available vehicles for the requested city, in JSON format. Relevant attribute information include vehicle identifier, geographic coordinates for the current location of the vehicle, vehicle type, and battery level. Each of these APIs were accessed every 60 seconds for the duration of data collection, stated in Table 1. At time of writing, the Lime and Flamingo APIs are still operational, however, Tier merged with another micromobility operator in mid-2024 and discontinued their API at the end of 2024.

Provided the set of available vehicles every 60 seconds, trips were identified by noting when a vehicle disappeared from the set of available vehicles (trip start) and when it reappeared in the set of available vehicles (trip end). Given the frequency of requests, this means that trips are accurate to a highest temporal resolution of one minute.

Provided an initial set of trips for each city, the data were then cleaned. Specifically, all trips where the battery level increased between the start and end of a trip were removed from analysis as an increase in battery level suggested that these were recharging/rebalancing trips completed by the operator. Similarly, trips where the average velocity exceeded 20km/hour were removed. All operators in the dataset limit their vehicles to a maximum of 20km/hour. Velocity was calculated as Euclidean distance between origin and destination divided by trip duration. Since full trajectories are not available, average speed is likely underestimated. In addition, trips shorter than 200m or five minutes were removed as well as those longer than 20 km or two hours in duration. This cleaning was done to remove vehicle adjustments and outliers in the data (see [15] for further details).

It is important to mention here that since the emergence of shared micromobility services, the ways in which data have been published via API has changed significantly. Today, many operators obfuscated their vehicle identifiers by randomizing them with every API call. Importantly, for all operators in the data set, I can confirm that the vehicle identifiers are not obfuscated. Lime does obfuscate the identifier in the vehicle_id field, but does not for the vehicle identifier in the rental_uris parameter. This allows for tracking a vehicle over API requests.

Completing the process above resulted in a set of micromobility trip for each of the five cities. Trips included the geographic coordinates of the origin and destination as well as the start time and end time, to the nearest minute.

I then identified three different geospatial units for aggregating the trip data. My motivation for aggregation is presented in the discussion section. The geographies include: 1. Socio-political boundaries. These are sub-city level administration units such as neighborhoods, districts, or traffic analysis zones within a city. As these are determined by the country or city, I understand them to be organized based on the characteristics of the population or physiographic features. 2. Hexagon grid at a 1,000 meter resolution and 3. Hexagon grid at a 500 meter resolution. These two hexagonal grids are uniform geometries that ignore population and physical geography. I felt it important to aggregate at a range of resolutions as these data can be used for different purposes by different stakeholders. The origins and destinations of all trips were intersected with the three different geographies to produce the spatial data sets available for download and analysis.

Temporally, all trips that started or ended within a geographic region were split into either weekday or weekend and origins were aggregated to the nearest hour. As with the spatial aggregation, this allows for dettailed temporal trend analysis but does not allow for the identification of individual trips or users within the data.

Finally, these data were cleaned to remove all geometries that contained no trips and all relevant data were compressed into a series of zipped folders for download. Each zipped directory includes four files: 1. A GeoJSON file containing polygons for the selected geography, 2. A meta data file containing relevant details on the source of the data and provenance information, 3. A Trip_OD.csv file containing counts and average duration between pairs of origins and destinations, and 4. A Region_Details.csv file containing temporal distribution of trips per geographic region. The data dictionaries for these last two files are presented in Table 3 and Table 4, respectively. These data have been prepared for all five cities and all three geographies and are published under a Creative Commons (CC BY 4.0) license.²²2https://creativecommons.org/licenses/by/4.0/

The MODAP platform boasts a number of features. These features have been numerically labeled in Figure 1 and are referenced in the following descriptive paragraphs in parentheses.

MODAP is built using completely open source software and the source code is also published as open source at https://github.com/grantdmckenzie/modap. On the server, the platform is running a LAPP³³3Linux, Apache, PostGreSQL, PHP architecture. Specifically, Ubuntu Linux running Apache 2 as the web server. All data are stored in a PostGreSQL relational database with the PostGIS extension allowing for spatial queries. Requests from the client (browser) are handled by a set of PHP scripts. Data downloads have been pre-processed, stored, and shared on the Open Science Framework at https://github.com/grantdmckenzie/modap. The front-end is a combination of HTML5, CSS3, and JavaScript. Two JavaScript frameworks are employed for various features. These are Leaflet⁴⁴4https://leafletjs.com for the web mapping functionality and Chart.js⁵⁵5https://chartjs.org for the dynamic graphs. All other functionalities were developed with native JavaScript.

As previously mentioned, the platform is designed for visual exploration of micromobility data, aiming to inspire new ideas and facilitate quick analysis of regional and temporal differences across cities, sub-regions, and vehicle types. For instance, one can select a region in a city such as Washington, D.C. and toggle between e-scooters and e-bicycles to identify the difference in trip duration between the selected region and all surrounding regions. Figure 3 shows such a selection. Since the legend remains constant between vehicle types, one can quite easily see that trips taken on e-bicycles travel to further destinations than e-scooters, while maintaining a similar duration. This is especially true for regions in the North of the city. By hovering over different destination regions, one can also view the mean and median duration differences between the two vehicle types.

Other analysis might focus on the difference in micromobility temporal patterns depending of if a region is a trip origin or a destination. Figure 4 shows a selected region in Wellington, New Zealand. Figure 4(a) shows the temporal patterns for the region when it is the origin of a trip. The dominant time period for trips originating in this region is during the morning commute on weekdays. We can see that it is higher than the city average, as shown in lighter blue. Compare that to the same region as a destination (Figure 4(b)). In this case, we find that the dominant time period is evening commuting hours, peaking at 17:00. Based on this exploratory analysis, one might make the initial assumption that the region is zoned to be residential. This might then spark further investigation into land use, socio-economic data, elevation, etc. in the selected and neighboring regions.

While the MODAP web platform was designed for visual exploration and analysis of the data, a user also has the ability to download the data sets and run their own analysis offline. These data support a wide range of analytical approaches and research questions. Here, I highlight a few simple examples to illustrate their potential.

One could compare the durations and times across all cities in the data sets to identify the most similar cities and most different cities. For instance, comparing various e-vehicle behavior has been a topic of interest recently [1, 17]. Through these open data, one can identify the difference in trip durations between e-scooters and e-bicycles in London, United Kingdom, for example. Figure 5 shows a density plot of each vehicle’s trip durations. This could be compared to other cities and used to inform policymakers on the suitability of these different vehicle types in their cities.

One could also use these data to identify the relationship between sex (male and female)⁶⁶6Reported as sex identified at birth in the UK Census. and micromobility trip volume in London, United Kingdom (Figure 6). To accomplish this, I accessed the demographic data from the UK 2021 Census⁷⁷7https://www.ons.gov.uk/census at the Middle layer Super Output Areas (MSOA) level and ran an areal interpolation [24] of the MSOA’s data to assign male and female population counts to the same 1,000 meter hexagon geography published through MODAP . Then, having data at the same geospatial resolution, I ran a Bivariate Moran’s I analysis comparing trip volume and percent female population in the city of London. The results report a Bivariate Moran’s I value of -0.2043 () with a z-score of -5.8137. This suggests that there is a a significant inverse spatial relationship between the two data sets, or rather that there is a strong relationship between micromobility trips and regions with higher male residential populations. This simple analysis is meant to demonstrate how these data could be used and further study might investigate additional factors related to the built environment and other socio-economic factors.

Finally, one could compare micromobility usage to automobile usage in a city. In this example I investigate spatial and temporal differences in duration of trips taken by the two modes of travel. The ride hailing company, Uber, previously published trip duration data at the level of Traffic Analysis Zones (TAZ) for a number of global cities through their Uber Movement platform.⁸⁸8https://www.uber.com/en-CA/blog/kepler-data-visualization-traffic-safety/ With access to these data, we can compare micromobility trip durations to those of passenger automobile trips. The Uber Movement data reports mean and median trip duration between all possible TAZ at every hour of either a weekday or weekend. Given the data in MODAP is published at the same spatial and temporal resolution (Administrative TAZ and hourly) for Washington, D.C., we can compare the two modes of travel. Figure 7(a) shows the difference in median trip duration between the two modes of travel for each origin TAZ in the city. This is calculated by subtracting the median automobile duration from the median micromobility duration for each TAZ. The results are the difference reported in seconds. We can see that there is some spatial clustering with the downtown core reporting the lowest difference and those on the outskirts of the city reporting larger differences. This is in line with existing work [16] comparing these data five years prior. Figure 7(b) reports these differences in duration by hour of a standard weekday. In this case, median automobile trip duration between all pairs of TAZ are subtracted from median micromobility trip duration and then the median of all of those is reported for each hour. We can see that while automobile trips are faster in all cases (y-axis is difference in seconds), the two modes of transport become more similar during peak commuting hours. This analysis showcases the types of comparison analysis that can be conducted between micromobility and other modes of transport within cities.

Micromobility services have grown substantially in recent years, now constituting a not-insignificant share of short, urban trips. As these services expand, they contribute to the evolving landscape of urban mobility, offering an alternative to, and often complementing, traditional transportation modes. In most cities, there is a regulatory requirement that operators of these services provide open APIs that publish real-time data on vehicle availability. In this work, I leveraged these data to reconstruct trips across five major cities over a three-year period (in most cases). Through the MODAP project, I am making these trip data openly available for download and analysis. To support these data, I developed an interactive geovisualization platform that enables users to engage and explore these data through their web browser. My objective is to provide researchers, policymakers, and urban planners with a rich and open source of new mobility data. My intention is for this to support evidence-based decision-making and contribute to the broader discourse on sustainable and equitable urban mobility.