Cycling behaviour in Germany: A dataset of cycling parameter across 100 cities based on City Cycling GPS trajectories Authors: Sven Lißner, Stefan Huber, Paul Lindemann Published: 2026 ABSTRACT AbstractThe dataset presented was created as part of the cycling campaign City Cycling in Germany during the years 2022–2024. The campaign involved >3000 municipalities across the entire country. Over a three-week period, which participants could freely select between May 1 and September 30, the goal was to replace as many car trips as possible with bicycle trips. Participants could use a smartphone application to record their data. The recorded GPS trajectories were transferred from the smartphone to a data lake as a database dump and were then regularly retrieved by a backend server for data pre-processing. During this process, trips, activities, modes of transport, and driving modes were identified. For the dataset presented, individual trajectories were processed with respect to several core variables characterizing cycling behaviour. The dataset was spatially filtered beforehand. A total of 109 cities and municipalities were selected based on variables such as population size, the number of campaign participants, topography, the modal share of cycling, and geographic location within Germany, in order to obtain a representative impression of cycling behaviour. As a result, >8 million trip records with 37 distinct variables were collected over the data years 2022–2024.The dataset was analysed within the scope of the project Bicycle Driving Behaviour in Germany, focusing on waiting and loss times for cyclists and the influence of various personal and city-specific parameters. Additionally, the data allows for evaluations of cycling frequency, trip distances, speeds, accelerations, daily and weekly patterns, and comparisons between the different data years regarding the influence of weather conditions. Moreover, the known starting and ending points of trips can be used in transport demand modelling for model development and calibration. Overall, the dataset offers extensive opportunities for scientific analysis due to its scope and wide spatial coverage, making it a valuable resource for research in cycling behaviour and transportation planning.Keywords: Bicycle trip statistics, Cycling speed, Acceleration, Waiting times, Cycling frequency, Cycling Behaviour Specifications Table SubjectEngineering & Materials scienceSpecific subject areaMobility Behaviour of cyclists in Germany. Trip Statistics from 109 Cities all over the country.Type of dataTable, Filtered, Processed, Codebook for Variables.Data collectionThe data was collected during the nationwide City Cycling campaign (2022–2024), a crowdsourcing initiative with self-selected participants. Recruitment was led by city administrations, companies (via mobility management), cycling NGOs, and schools. Data was gathered via a smartphone app (Android/iOS) using GPS sensors at 1 Hz frequency. Uploaded data was pre-processed on backend servers, splitting raw GPS data into trips and activities. Only bicycle tracks were published, with origins, destinations, and trip statistics replacing raw trajectory data.Data source locationData was collected in 109 cities all over Germany.Data accessibilityRepository name: OPARAData identification number: https://doi.org/10.25532/OPARA-1150Direct URL to data: https://opara.zih.tu-dresden.de/handle/123456789/2167Related research articleLißner, S., Wagner, F., Huber, S., Lubashevsky, K.; Okhrin, I.: No Bicycle Race – Cyclists Waiting Time and Delay in German Cities, Journal of Cycling and Micromobility Research, submitted 04/2026, [1] Value of the Data •With 8 million trips, the datasets represent a highly comprehensive set of real movement data from cyclists across different urban types [2]. Through the provided origins and destinations, as well as route characteristics such as time and speeds, they enable extensive insights into the mobility behaviour of cyclists. This applies to both urban and rural study areas of varying topographies and with differing cycling modal splits. The provided percentiles for speeds, accelerations, and waiting times at intersections allow, for example, a comparison of traffic flow quality across different municipalities. Hashed user IDs and sociodemographic attributes such as age and gender additionally enable analyses at the individual person level. With 8 million trips, the datasets represent a highly comprehensive set of real movement data from cyclists across different urban types [2]. Through the provided origins and destinations, as well as route characteristics such as time and speeds, they enable extensive insights into the mobility behaviour of cyclists. This applies to both urban and rural study areas of varying topographies and with differing cycling modal splits. The provided percentiles for speeds, accelerations, and waiting times at intersections allow, for example, a comparison of traffic flow quality across different municipalities. Hashed user IDs and sociodemographic attributes such as age and gender additionally enable analyses at the individual person level. •The campaign data and their processing are well documented. The structure of the participants was analysed and documented using a survey with a response of 100,000 participants [3]. The data preparation was carried out following an established schema and is oriented towards the reported results of mobility diaries [4]. The campaign data and their processing are well documented. The structure of the participants was analysed and documented using a survey with a response of 100,000 participants [3]. The data preparation was carried out following an established schema and is oriented towards the reported results of mobility diaries [4]. •Data could be compared to statistics of other GPS data sets in cycling like [5] or [6] or [7] to elaborate on behavioural differences across different countries. Data could be compared to statistics of other GPS data sets in cycling like [5] or [6] or [7] to elaborate on behavioural differences across different countries. •The data can also be used to generate synthetic route alternatives between real sources and destinations like elaborated in Bovy et al. [8], or for the calibration and further development of traffic demand models. The data can also be used to generate synthetic route alternatives between real sources and destinations like elaborated in Bovy et al. [8], or for the calibration and further development of traffic demand models. •The dataset's depth, spanning a total of 15 months over 3 years, enables the comparison of cycling behaviour within a single year or across multiple years, and thus, for example, the examination of the influence of weather or other exogenous variables more in depth like in recent literature [9]. The dataset's depth, spanning a total of 15 months over 3 years, enables the comparison of cycling behaviour within a single year or across multiple years, and thus, for example, the examination of the influence of weather or other exogenous variables more in depth like in recent literature [9]. Background Cycling behavior depends on a variety of factors. The individual level, in particular, is already relatively well-documented. For example, differences in trip distance and speed by gender and age are well-known [10], as is the influence of perceived safety on the frequency of cycling trips [11]. Many of these factors are not only related to the cyclist themselves but are a direct consequence of the built environment, i.e., the characteristics of a study area. Due to the lack of data availability so far, the built and experienced environment of cyclists has not yet been compared on a larger scale. Comparisons between cities in the literature currently only occur at the level of modal split values, accident data, or surveys [12]. The nationwide GPS data enabled the creation of a curated dataset containing many individual trips across different city types, allowing for the analysis of city-specific differences. The use of route statistics instead of actual trajectories ensures data privacy. In the research article associated with the dataset, this is demonstrated for the parameters of waiting and delay times. However, the dataset itself offers the potential for further analyses. Data Description The dataset consists of the following elements: ▪Processed campaign data from 2022 in .csv file format (Trip_stats_2022.csv), a datafile with 2.1374.436 trip statistics collected between 01.05.2022 and 30.09.2022, including 37 variables. Blank entries in variable “gender” indicate a lack of response. IDs are hashed smartphone-IDs and change with a new installation of the app. Therefor IDs can remain the same over all three campaign years, although this is not necessarily the case. Processed campaign data from 2022 in .csv file format (Trip_stats_2022.csv), a datafile with 2.1374.436 trip statistics collected between 01.05.2022 and 30.09.2022, including 37 variables. Blank entries in variable “gender” indicate a lack of response. IDs are hashed smartphone-IDs and change with a new installation of the app. Therefor IDs can remain the same over all three campaign years, although this is not necessarily the case. ▪Processed campaign data from 2023 in .csv file format (Trip_stats_2023.csv), a datafile with 3.181.781 trip statistics collected between 01.05.2023 and 30.09.2023, including 37 variables. Processed campaign data from 2023 in .csv file format (Trip_stats_2023.csv), a datafile with 3.181.781 trip statistics collected between 01.05.2023 and 30.09.2023, including 37 variables. ▪Processed campaign data from 2024 in .csv file format (Trip_stats_2024.csv), a datafile with 2.712.312 trip statistics collected between 01.05.2024 and 30.09.2024, including 37 variables. Processed campaign data from 2024 in .csv file format (Trip_stats_2024.csv), a datafile with 2.712.312 trip statistics collected between 01.05.2024 and 30.09.2024, including 37 variables. ▪City stats (city_stats_short_rep.csv), a list of all 109 involved cities, including the variables topography, inhabitants and modal_split_bike City stats (city_stats_short_rep.csv), a list of all 109 involved cities, including the variables topography, inhabitants and modal_split_bike ▪The trip-stats codebook in .pdf format (Dataset_Description_Tripstats.pdf), a list with the description of all variables included. It includes information about how specific variables were computed. The trip-stats codebook in .pdf format (Dataset_Description_Tripstats.pdf), a list with the description of all variables included. It includes information about how specific variables were computed. ▪The city-stats codebook in .pdf format (Dataset_Description_City_stats.pdf), a list with the description of all city related variables included. The city-stats codebook in .pdf format (Dataset_Description_City_stats.pdf), a list with the description of all city related variables included. This article describes the dataset provided in the linked repository of the project “Bicycle Driving Behaviour in Germany.” The core dataset comprises approximately 8 million trips collected from participants in a three-year cycling campaign across 109 German cities (Fig. 1). It includes variables such as trip start and end points (latitude and longitude) and times, trip distance, duration, waiting times, as well as various percentiles of speed, acceleration, and deceleration. The dataset therefore offers substantial analytical potential while remaining compliant with data protection regulations. Data protection procedures were approved by the data protection officers of TU Dresden and the federal states of Hesse and Saxony. Spatial distribution of 109 German municipalities which were selected for data preparation and analysis. Data collection The City Cycling (CC) campaign is a nationwide initiative taking place between May and October, in which >3000 municipalities participate. The aim is to cycle as many kilometres as possible within a freely selectable three-week period defined by each municipality, instead of traveling by car. Since 2017, participants have had the option to record their trips using a smartphone. The collected data are subsequently processed, analysed, and made available, for example, to municipalities for planning purposes. The dataset presented here is part of the research project “Cycling Behaviour in Germany” and comprises a representative sample of municipalities. The 109 cities included in this dataset vary in size (Table 1), ranging from the capital Berlin as the largest city to much smaller municipalities. They also differ in terms of topography (Table 2) and cycling mode share (Table 3), with the city of Münster exhibiting the highest share, at around 30% of all trips made by bicycle. The cities were selected to represent a broad spectrum of urban types in Germany, which are likely associated with differing cycling behaviours. Table 1Distribution of city sizes in the datasetTable 1 dummy alt textInhabitantsCities<10,0002910,000 to 100,00046100,000 to 1000,00030>1000,0004sum1093,755,251max1,329min61,043med Distribution of city sizes in the dataset Table 2Distribution of different topography in the datasetTable 2: dummy alt textTopographyCitiesflat38hilly35mountainous36 Distribution of different topography in the dataset Table 3Distribution of cycling mode-share in the datasetTable 3: dummy alt textMode_Share_CyclingCities<10%4510–15%3715–20%20>20%7Sum10930%max4%min11%med Distribution of cycling mode-share in the dataset We introduced four size related classification schemes adapted from established German planning frameworks. The population size classes are a simplified version of the official German federal classification, which distinguishes seven categories. Municipalities with fewer than 10,000 inhabitants are classified as small cities, those with 10,000–99,999 inhabitants as medium-sized cities, those with 100,000–999,999 inhabitants as large cities, and municipalities with more than one million inhabitants as metropolises. The topographic classification is based on the approach used in the German Household Travel Survey (SrV). It considers the proportion of roads within the primary road network with a longitudinal gradient exceeding 2%. Municipalities with <2% of their main network above this threshold are classified as flat, those with 2–20% as hilly, and those with >20% as mountainous [13]. Finally, the cycling mode share categories follow the classification used in the German National Cycling Plan 2020. Municipalities with a cycling mode share below 10% are classified as starters, those with a mode share between 10% and 20% as improvers, and those with a cycling mode share above 20% as front-runners. In a stratified selection process, we first ranked all approximately 3,000 German municipalities according to both the absolute number of City Cycling participants and the participation rate relative to the total population. Municipalities were then selected to achieve an approximately balanced representation across the predefined categories while ensuring a broad geographical coverage of Germany, including municipalities from both the former East and West Germany. Like the variables shown in Table 1, Table 2, Table 3, the number of participants and recorded cycling trips was also a selection criterion to include a city in the dataset. The number of participants ranges from 229 to 20,092 (in Berlin) with a median of 839 participants per city. Participants were counted using the domestic principle, which includes commuters and tourists. The average number of trips per participants amounts to 11,62 trips per person per city. Please note: all numbers are from the CC campaign 2022 and are subject to changes over the years. Data preparation The following data pre-processing is based on method developed by Lißner and Huber and can be read in detail in the respective publication [4]. The basic work-flow is shown in Fig. 2. The data pre-processing pipeline starts with the transfer of raw data from smartphones to a centralized data lake, from where it is forwarded to a backend system responsible for all subsequent processing steps. Within this backend, multiple procedures are applied to transform raw trajectory data into a structured and analysable format. Date pre-processing scheme according to Lißner & Huber 2021 The pre-processing includes filtering and smoothing of trajectories, trip segmentation, mode recognition, and driving mode detection. Filtering and smoothing address data quality issues such as outliers and missing values, ensuring consistent and reliable trajectories. In the filtering process recordings shorter than 30 s were discarded in the first processing step due to their limited information content, as they typically represented app test recordings rather than actual bicycle trips. In a second step, GPS points with a reported positional accuracy worse than 50 m were removed, as were points with a system-reported speed exceeding 25 m/s, since such speeds are not attainable under normal cycling conditions. Owing to the data-minimisation approach adopted in the overall processing pipeline, the total number of discarded data points cannot be quantified directly. However, in a test dataset comprising 11.4 million GPS points collected in the city of Dresden, 3.9% of all GPS points were identified as erroneous and subsequently removed [4]. Trip segmentation separates trips from stationary activities based on speed, travelled distance, and changes in directional vectors at the GPS point level. Mode recognition distinguishes between walking, leisure (sports) cycling, commuter cycling, and other transport modes using a decision tree heuristic. This approach relies on various indicators, including trip speed percentiles, total trip distance, and detour factors. To further characterize cycling behaviour, driving mode detection is applied. This classification is primarily based on speed and acceleration and differentiates between four states: stop, acceleration, deceleration, and constant movement. Points are classified as waiting events if the maximum speed within a 180-second sliding time window is below 0.2 m/s. Acceleration and deceleration are identified using thresholds of ±0.2 m/s², while all remaining points are categorized as constant movement. To ensure user privacy, the data is anonymized by truncating the beginning and end of each trip within a circular buffer of at least 100 m plus a random offset between 1 and 100 m The trip segmentation approach achieved an accuracy of 81% in correctly identifying individual bicycle trips. The remaining 19% primarily consisted of over-segmentation errors, in which single trips—typically longer leisure rides—were split into two or more trips due to extended stop durations. Mode detection performed satisfactorily, achieving an overall accuracy of 92% across the bicycle and sport cycling modes. Misclassifications occurred predominantly between the bicycle and sport cycling classes, as well as between cycling and walking, particularly for very short and slow segments that had been erroneously separated from otherwise continuous bicycle trips [4]. Afterwards, all non-cycling trips and stationary activities are removed from the dataset. Owing to the data-minimisation approach adopted throughout the processing pipeline, comprehensive statistics for the individual filtering steps are difficult to derive, as discarded data are neither retained nor quantified separately. To provide an indication of the data quality and filtering performance, statistics were generated for a representative subsample. The raw dataset comprised 876,036 recordings containing 1.658 billion GPS points. During preprocessing, 6.95 million GPS points (0.4%) were discarded because their reported positional accuracy exceeded 50 m. In addition, 108,451 recordings (12.4%) were removed because they contained fewer than 90 GPS points, while a further 24,797 recordings (2.8%) were discarded due to erroneous GPS jumps. The filtering process resulted in 854,258 processed tracks. This number does not correspond directly to the difference between the raw and discarded tracks, as recordings are split whenever the temporal gap between two consecutive GPS points exceeds 10 min. Such gaps typically occur on smartphones with automatic GPS pause functions or aggressive power-saving modes. Following trip segmentation, the preprocessing produced 988,111 trips, of which 903,764 (91.5%) were classified as bicycle, 55,661 (5.6%) as sport bicycle, 17,920 (1.8%) as walking, and 10,766 (1.1%) as other. Finally, trip-level statistics are computed for each bicycle trip. Speed-related metrics are calculated using only movement points, while acceleration and deceleration percentiles are derived from the respective classified points. Additional details on dataset variables and processing steps are provided in the accompanying data description documentation. Dataset statistics In total, the datasets from the three campaign years comprise more than eight million trips, including 7.54 million regular bicycle trips and 421,110 sports trips after filtering for age outliers under 10 and over 90 years before descriptive statistics. We suggest this, as for both groups it is very unlikely to have taken part in the data collection with a smartphone-app. Descriptive statistics for all three datasets and the mode “bike” are presented in Table 4. Notable differences can be observed in the distances travelled by male and female cyclists, which ultimately result in a substantially higher total cycling distance for male participants. This difference remains largely consistent across all three years of data collection. Table 4Descriptive statistics for all data set years, without sports cycling grouped by gender (2022–2024)Table 4: dummy alt textGenderYearNumber of personsTotal tripsAvg. trips per personMean ageMean km per tripTotal kmNA20223,27860,44818.433.44.97254,859female202252,757928,82917.641.75.524,138,103male202260,4461,016,54916.843.57.556,040,046NA20234,36389,90820.633.85.26412,777female202373,5551,388,37918.942.15.686,655,489male202383,8071,511,46018.043.57.569,360,251NA202412,186223,96918.442.66.561,223,592female202466,6551,112,72316.742.65.765,223,994male202474,5191,203,64116.243.57.427,285,755 Descriptive statistics for all data set years, without sports cycling grouped by gender (2022–2024) As it was not mandatory to enter the gender in the registration process there are a number of missing values, which we suggest to drop in dataset re-use. The proportion of missing gender values was higher in 2024 than in previous years due to a data transmission error which occurred early in the CC campaign. Although age and gender information were subsequently recovered using the participant registration data, this was not possible for all records, resulting in an increased share of missing values in the 2024 dataset. Table 5 presents the statistics for all trips classified as “sports cycling” during data pre-processing. It also indicates a higher number of sports trips and greater distances among male participants, with values remaining relatively constant across the years. Table 5Descriptive statistics for all data set years, sports cycling, grouped by gender (2022–2024)Table 5: dummy alt textGenderYearNumber of personsTotal tripsAvg. trips per personMean ageMean km per tripTotal kmNA20228282,3122.7936.819.1053,448female202211,60529,6002.5544.317.00545,393male202222,36477,1983.4546.525.902,170,537NA20231,1563,5123.0436.419.0077,538female202316,44846,2122.8144.417.30860,000male202331,723115,4613.6446.125.903,262,408NA20243,88013,6173.5146.423.40351,258female202414,57238,5782.6545.217.50723,895male202427,31894,6203.4646.426.102,670,478 Descriptive statistics for all data set years, sports cycling, grouped by gender (2022–2024) The contributing cyclists are predominantly everyday cyclists. More detailed information on the population is provided by a survey of >100,000 participants of the City Cycling Campaign (CC) [3]. Men and women are distributed approximately equally across the dataset years (see Fig. 3). The slight predominance of male participants is consistent with the findings of the German Household Travel Survey (Mobility in Germany), in which 10% of all respondents reported having used a bicycle on the survey day, specifically 11% of male respondents and 9% of female respondents [13]. Gender distribution of City Cycling participants in the 109 cities (2022–2024) The violin plot (Fig. 4) shows a consistent age distribution for both genders across all years. Due to methodological constraints, younger individuals are underrepresented in the sample. This is partly related to smartphone ownership and partly to legal restrictions on app installation, which typically require a minimum age of 14. As a result, the travel behaviour of children is only indirectly captured in the dataset, for example through escort trips to kindergarten or primary school undertaken by parents. This should be regarded as a caveat, as the German Household Travel Survey (HTS) Mobility in Germany indicates that bicycle use is particularly common among younger age groups. On the survey day, 13% of children aged 0–9 years and 17% of those aged 10–19 years reported using a bicycle, compared with an overall average of 10% across the population [13]. Age distribution (kernel density) for City Cycling participants as violin plots (2022–2024) A similar pattern can be observed at the upper end of the age spectrum. The share of older adults (65+) is noticeably lower than what is commonly reported in household travel surveys. The HTS reports that 9% of people aged 60–69 years and 6% of those aged 70–79 years used a bicycle on the survey day [13]. This is also likely linked to access to and usage of smartphones as the primary data collection instrument. To improve interpretability and reduce the influence of outliers or erroneous entries, the age range was further restricted to values between 10 and 90 years, and the violin plots were trimmed accordingly. The mean age of participants shows a slight increase over time. For female participants, it rises from 41.7 years in 2022 to 42.6 years in 2024, whereas it remains nearly constant for male participants (43.6 years in 2022 and 43.7 years in 2024). In addition to the sociodemographic characteristics of the participants, travel behaviour is also of interest and offers a wide range of analytical possibilities. The following section briefly presents the results for trip distance distribution and speed distribution, as these can be relatively easily compared with data from Household Travel Surveys (HTS). For Fig. 5, trips classified as sports cycling (mode_type = 2; dist>30 km or a detour factor > 3) during data processing were excluded in order to provide a representation that is as close as possible to everyday travel behaviour. In the 20 km category, all trips ≥20 km were aggregated. Overall, the results reflect patterns well known from HTS: the majority of trips take place within a local range of up to 10 km. An in-depth case study regarding the representativeness of the dataset of a German city can be found in Lißner et al. [14]. Distribution of trip distances after data pre-processing (2022–2024) The distribution of speeds for everyday cycling trips exhibits patterns comparable to those observed in household travel surveys and measurement-based studies. Across the data from 2022 to 2024, the distribution closely approximates a normal distribution. The unweighted mean speed across all trips is approximately 15.1 km/h, while the distance-weighted mean is higher at around 17.8 km/h. Distribution of average trip speed after data pre-processing (2022–2024). The datasets, with >30 analysable variables, offers considerable additional potential for gaining insights into urban cycling behaviour. In particular, linking the data with city-level information via the ASR variable enables comparative analyses across 109 German municipalities with diverse characteristics. However, a detailed or even exemplary exploration of these possibilities would go beyond the scope of this article. Experimental Design, Materials and Methods The experimental design can best be described as a self-selected sample of cyclists. Within the framework of the underlying cycling campaign, municipalities, companies, and NGOs are responsible for distribution and actively encourage participation. In several federal states (Baden-Württemberg, Hesse, Bavaria, North Rhine-Westphalia, Rhineland-Palatinate, Lower Saxony, Saarland, and Thuringia), the “School Cycling” initiative, a dedicated sub-program targeting schools, is additionally promoted. Participation is voluntary and not based on a random sampling design, which results in a sample that predominantly includes already active cyclists. In addition to trip recording via a smartphone application (1 Hz GPS), participants can also manually enter their travelled distances through an online interface. Approximately 40% of participants use the smartphone apps on iOS and Android devices. The structure of the campaign is based on a competitive framework aimed at promoting climate-friendly mobility. Participating municipalities compete within predefined city size classes to achieve the highest total cycling distance and, consequently, the greatest amount of avoided CO₂ emissions. Within each municipality, teams (e.g., from different employers) are formed and also compete against each other. The winning team can be additionally recognized or awarded by the respective municipality as an incentive. The dataset presented here can be analysed both descriptively and using inferential statistical methods. Examples of data filtering and descriptive analyses are provided as R code in the accompanying repository. Examples for dataset re-use For a potential analysis of trip purposes in bicycle trips, land-use information from OpenStreetMap (OSM) could be incorporated for the respective study area. For this purpose, the administrative boundaries of the municipality of interest, or of all municipalities, could be obtained as shapefiles from the German Federal Agency for Cartography and Geodesy (BKG) using the official municipality identifier (ARS key) and subsequently linked to the trip data. Within the defined spatial boundaries, land-use information can then be obtained either from the BKG or directly from OSM. A spatial overlay with the origin and destination coordinates of the presented datasets would provide information on the characteristics of trip origins and destinations, which could serve as a proxy for potential trip purposes. To analyse the influence of weather conditions, particularly precipitation, on cycling behaviour, daily weather data from the German Climate Data Centre (DWD Climate Data Center) could be spatially assigned to the respective study regions and subsequently evaluated in relation to cycling patterns. Limitations Despite the large sample size and high spatial resolution of the dataset, several limitations need to be considered. First, a potential self-selection bias exists, as the raw sample is not drawn using a probabilistic sampling approach. The smartphone-based data collection approach introduces a frame error, as people without access to or willingness to use a smartphone are excluded from participation. This particularly affects younger children and older adults. Consequently, the presented dataset is likely more representative of daily work-related mobility patterns, while shorter and slower trips by elderly people and children may be underrepresented. Besides that, experienced and frequent cyclists are likely overrepresented due to the recruitment approach of the data collection. The largely non-incentivized participation in a cycling campaign is expected to primarily attract already established cyclists. However, the recruitment through teams, for example within workplaces, can also motivate people who have not previously used the bicycle regularly to participate. Their share would likely be higher in a probability-based sample, although such a sample would still be subject to limitations regarding representativeness, at least with respect to cycling activity and cycling mileage. Furthermore, the dataset does not distinguish between conventional bicycles and e-bikes. This limitation should be considered when interpreting speed-related indicators, as e-bike users may exhibit different cycling behaviour compared to conventional bicycle users (see Maurer et al.) [15]. This issue is currently being addressed: datasets generated after the 2026 City Cycling campaign will include a distinction between different bicycle types. Another limitation concerns the spatial representation within individual cities. The dataset was not corrected for spatial sampling biases, and certain areas may be overrepresented due to local recruitment effects, for example, through particularly bicycle-friendly employers, universities, or other institutions with high participation rates. Such spatial biases should be assessed individually for each city. Poliziani et al. propose an approach based on Shannon’s entropy to evaluate spatial representativeness [16]. Since the dataset provides timestamps as well as GPS coordinates of trip origins and destinations, such analyses can be conducted for individual cities in future work. The data collection period between May and the end of September primarily covers the warmer months and largely excludes school holidays. Therefore, the dataset is not suitable for directly capturing seasonal variations in cycling behaviour. In addition, the number of participants per city may vary between years depending on the intensity of local campaign promotion and recruitment activities. Furthermore, the dataset does not provide a direct and reliable classification of trip purposes. However, such information could potentially be derived in future analyses by combining typical movement patterns like time and frequency with contextual information, for example, land-use characteristics at trip origins and destinations. Finally, the transferability of the dataset is limited, as it exclusively contains data from German cities. While findings may be transferable to comparable European and parts of North American urban contexts, cycling behaviour in Asian megacities or cities in Australia, South America, and Africa may differ substantially due to differences in infrastructure provision, urban form, climate, and cycling culture. However, the methodological approach itself, using a cycling campaign to collect and analyse data, seems highly transferable to other contexts. CRediT Author Statement Dr. Sven Lißner: Conceptualization, Methodology, Validation, Formal Analysis, Writing – Original Draft, Visualization, Project Administration, Funding acquisition. Dr. Stefan Huber: Conceptualization, Methodology, Validation, Project administration, Funding acquisition. M.Sc. Paul Lindemann: Conceptualization, Methodology, Software, Data curation. Ethics Statement The authors clarify that data collection within the cycling campaign was conducted only after users provided opt-in consent via a data protection-compliant declaration. Subsequently, the data were anonymized such that they can no longer be attributed to a natural person with reasonable effort. The authors further confirm that they have read and will comply with the ethical requirements for publication in Data in Brief. The use of the dataset does not involve human subjects in an experimental setting, animal experiments, or additional data sourced from social media. Funding This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project number(s) 514587991. Acknowledgements We would like to thank all participants of the City Cycling campaign for their voluntary data contribution. The same applies to Klima-Bündnis Services GmbH and the team behind the City Cycling campaign Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. References